AMD will unveil next-generation EPYC server processors on the evening of November 10.The company officially announced this on its website.The manufacturer did not indicate which series the new chips will belong to, but it is very likely that we are talking about EPYC Genoa processors.Image source: AMDIt is currently known that EPYC Genoa processors will be able to offer up to 96 physical cores on the Zen 4 architecture with support for up to 192 virtual threads.AMD is going to release a wide range of chips with different numbers of cores.The minimum is unknown, so let's just note that the current EPYC Milan series includes models with the number of cores from 8 to 64. We also know from leaks that EPYC Genoa processors will get 1 Mbyte of L2 cache per each core.L1 cache size is 32 Kbytes for instructions and data, L3 cache size - 4 Mbytes per each core.New processors will be launched together with new AMD Socket SP5 platform (LGA 6096).The new processor socket will support 12-channel DDR5 RAM controller and PCIe 5.0 interface.AMD will present new EPYC processors during live broadcast starting at 20:00 PM Moscow time on November 10.You will be able to watch it on AMD website or on official AMD YouTube channel.
To a wide audience, China-based Hygon Information Technology is known for its ill-fated partnership with AMD, which began shortly after Lisa Su took over as CEO, but ended in 2019 after tightening US export control rules.This week, Hygon stock went public in Shanghai, valuing the company's business at $20.7 billion.Image source: Hygon Information TechnologyFactually, as the Nikkei Asian Review explains, Hygon Information Technology's initial public offering was priced 94 percent higher than estimated, indicating investor confidence in the company's continued business growth.This developer of central processors and gas pedals for machine learning systems was founded in 2014, the company's development and research spending reaches 70% of its rather modest revenue, which did not exceed $341 million last year.It cannot be ruled out that access to capital in the stock market will allow Hygon to open a new stage in the company's development.Until 2019, it had access to the technological capacities of TSMC and GlobalFoundries, as it received from them licensed clones of AMD EPYC and Ryzen first-generation processors with the encryption block modified to Chinese national standards.Later, the U.S.sanctions limited the ability of Hygon to produce these processors, forcing the company to focus on its own developments.Investors are likely to be looking at the prospects of increased state investment in the semiconductor sector of the PRC economy, since now it covers only 20% of the national market demand for components, which is far from the target.
Few industry analysts are under illusions about PC market dynamics in the coming months.Gartner predicts a 9.5% year-over-year decline in PC sales, a view shared by Micron Technology, so it's not surprising to see moderately pessimistic estimates about AMD's business in the customer segment.Profile revenue of the company next year could decline by $675 million, according to experts Northland Capital Markets.Image source: AMDExcerpts from the analytical note published by the resource Seeking Alpha.According to the authors of the forecast, AMD next year will reduce the sales of central processors in the client segment by 6%, and graphics - by 7% in value terms.On the one hand, the consumer demand will be affected by macroeconomic factors and the saturation effect after two pandemic years of rapid growth.On the other hand, the video card market could be impacted by lower demand from miners.In the aggregate, AMD revenue in the client segment next year could decline by $675 million.For the company's income as a whole, it will not be a serious loss, because revenue from sales of components for Microsoft and Sony gaming consoles should increase by $400 million from the previous forecast of $740 million.Thus, the supply of these components will bring AMD at least $1.14 billion next year.By the way, the company's management has already noted that next year should be a peak year for the current generation of gaming consoles.Both forecasts are in line with industry trends, as the Xbox Series X/S and PlayStation 5 game consoles will enter the third year of their life cycle.The merger with Xilinx, which has already taken place, will have a negative impact on AMD's business next year, as that brand's FPGAs will reduce profile revenue by 6%.At the same time, the expansion of EPYC processors in the server segment will be so successful that the growth of revenue by 55% to an impressive $3.1 billion will cover all possible failures in other areas of business.
When AMD first brought first-generation EYC processors to market, it set fairly modest market share goals: up to 5%-6% by the end of 2018 and more than 10% in the next year and a half.Having surpassed these milestones, it declined to forecast further, but did not rule out beating the historical record, which corresponded to 26% of the market in the middle of the first decade of this century.Image source: AMDExperts at JPMorgan after interviewing executives of companies that are major AMD customers in the server segment, took the liberty to say that this year it managed to take 20% of the server processor market, and it will only take a few years to double that share.At least in the short term, AMD can safely count on a share of 30%, according to experts.They also added that in the previous two years, revenue from AMD server processors has quadrupled, and only for the previous year it increased by 120%.The appearance on the market after the fourth quarter of this year's new EPYC Genoa generation processors will further strengthen AMD's position, because Intel is still having difficulty increasing the mass supply of competing processors Sapphire Rapids.The accession of Xilinx and Pensando to AMD only increases the former's chances of success, as JPMorgan experts summarize.
The transition of the dynamic Chinese electric car maker NIO to the server systems based on AMD EPYC generation Milan processors this week was referred by the Chinese representative office of AMD to the achievements of the company itself.At the same time representatives of the automaker a couple of days later rebuked AMD, claiming the illegality of its use of this fact for advertising.Image source: CnEVPost explains with reference to the director of corporate relations NIO Ma Lin (Ma Lin), the company is not working with AMD and does not discuss such a possibility.In addition, NIO has not given AMD's Chinese representative office the right to use such information for promotional purposes.As explained by the representative of the automaker, as a result of the appeal to the Chinese representative AMD of the latter company has agreed to remove the press release from its website, but at the same time left a video on its page in Weibo, which talks about the cooperation between the two companies.NIO representatives insist on the removal of this video, which has already gathered over a million views.In fact, as explained by NIO's director of corporate affairs, the company purchased third-party server systems that contained AMD processors, but the deal was formalized without the direct participation of the latter company.Nevertheless, the Chinese representative office of AMD used the fact of the deal purely in its own interests, reporting on direct cooperation with NIO, which irritated the representatives of the latter.
A fresh leak from the Geekbench synthetic test database indicates that AMD's Zen 4 server processors will get 1 Mbyte of L2 cache per core. This is confirmed by tests of an engineering sample of one of the EPYC Genoa series processors. Since AMD Ryzen desktop processors are made up of the same chipsets as the EPYC server chips, the presence of doubled L2 cache memory will probably be true for the Ryzen 7000 models on Zen 4 as well.
The aforementioned Genoa series chip with OPN number 100-000000479-13 has 32 physical cores on board with support for 64 virtual threads, and runs at a base frequency of 1.2 GHz. Considering that this is a very early engineering sample, it's safe to assume that the market version of the processor will have a much higher frequency. According to Geekbench data, the tested chip has one megabyte of L2 cache per core, which is twice the size of Naples, Rome and Milan series EPYC processors. At the same time the new chip has 32 Mbytes of L3 cache per chip compared to the same Milan series processor (CCX unit with 8 cores). But since Genoa series processors will be able to offer up to 96 physical cores, the total L3 cache volume will in any case be higher than that of their predecessors. The only exception will be the newer EPYC Milan-X chips, with 3D V-Cache technology, which triples the cache size. It should be recalled that EPYC 7004 (Genoa) processors will be manufactured according to the 5-nm process technology. They will be the first AMD server processors, which will support DDR5 RAM, as well as PCIe 5.0 interface. According to AMD, these chips should go on sale this year. The company's partners are already receiving samples for testing.
Late test samples (Qualification Samples, QS) of AMD EPYC 7773X processors have been found on Chinese Taobao marketplace, they are being offered for $2 823.97, which is almost 4 times cheaper than the official retail price.
AMD announced EPYC Milan-X series server processors with 3D V-Cache technology last November. The company didn't specify an exact release date for the Milan-X chips, only saying that they will be available in Q1 2022 - apparently, that's why the late test samples have already leaked onto the market. Milan-X processors are based on the same Zen 3 cores and the same 7nm process technology as the Milan chips. The flagship EPYC 7773X from China boasts 64 cores, 128 threads and 768 Mbytes of L3 cache due to 3D V-Cache technology. It has a TDP of 280W with a base clock frequency of 2.2GHz and a peak clock frequency of 3.5GHz. Back in August one of the biggest B2B retailers published the price of this processor on its website, and it was an impressive $10 746,99. It's not surprising that the Taobao seller is asking just over a quarter of the price for a pre-production EPYC 7773X. It's not quite legal to sell such kind of products, and the specifications of the prototype presented on the site don't match the final ones: the base clock frequency is 2.1 GHz, which is 100 MHz less than the manufacturer's one, and there's no information about the peak clock frequency. The Milan-X processors will not require a motherboard upgrade and will fit into the same SP3 socket as the Milan. But it is possible that a firmware upgrade will be required.
Unknown attackers took advantage of a vulnerability in the Log4J library to hack AMD EPYC processor-based servers and mine the CPU cryptocurrency Raptoreum on these resources. As a result, the hash rate of the entire Raptoreum network doubled for as long as the machines were taken offline.
The Log4J library vulnerability received the highest rating because it requires no physical access and allows attackers to establish connections, download data, or run arbitrary code on the underlying machine. HPE servers based on AMD EPYC were chosen as a target for a reason: Raptoreum cryptocurrency runs on Proof-Of-Work (PoW) model, and its GhostRider algorithm is optimized for CPUs and is resistant to ASIC systems. The most popular for Raptoreum mining are 12-core AMD Ryzen 9 5900X and 16-core 5950X & ; both have 64 MB of L3 cache, AMD EPYC Milan processors on Zen 3 architecture have twice as much & ; 128 MB (Milan-X has 768 MB). Raptoreum developers discovered an abnormal burst of hash rate on the network on December 9. The number of machines on the network was growing at an even pace, and performance dramatically doubled from 200 Mhash/s to 400 Mhash/s. The hack was discovered late: the compromised machines were taken offline only on December 17. During this time, hackers received approximately 3.4 million Raptoreum tokens, which were valued at $110 thousand as of December 21. Subsequently, 1.5 million tokens of this number were sold on the CoinEx crypto exchange. The rest of the assets remained in the wallet & ; probably, the attackers are waiting for the rise in cryptocurrency prices.
Investors took some time to be optimistic about AMD's quarterly results, but by the close of trading the company's share price rose by more than 7 %, reaching $97.93. The last time it was close to that level was on January 11, and since then it has not demonstrated any clear positive dynamics.
AMD's stock price rally was driven by both a successful second quarter and a good outlook for the second half of the year. Analysts surveyed by Barron’s resource were particularly encouraged by AMD's prospects in the server segment. As it was noted at the quarterly event, the company's revenue from server components exceeded the level of 20 % of the total, and this share will continue to increase in future periods. The Milan generation EPYC processors are spreading through the market faster than their predecessors, and will start to dominate the company's supply structure as early as this quarter. Some concern of New Street Research analysts is caused by the situation in the personal computer market. Next year, in their view, this segment will not grow at the same pace. Intel in this case can use its own production capacity for revenge. In the second half of this year, revenue in the PC market will grow by 20 % compared with the result of a year ago, but this growth will be held back by a shortage of components. Experts at Atlantic Equities have raised the target for AMD stock to $150, although colleagues at BMO Capital Markets limited themselves to the $80 level.
In the second half of the year, as AMD management admitted at the quarterly conference, the server segment will determine the company's revenue dynamics, so there was a lot of attention paid to relevant comments at the event. This quarter, 7nm EPYC processors of Milan generation will begin to dominate AMD's supply structure over their Rome generation predecessors.
The server segment is understandably inert when it comes to purchasing next-generation products, but AMD CEO Lisa Su noted at the quarterly conference that the Milan family's EPYC processors are outperforming their predecessors in terms of market expansion. In sequential comparison, their sales volumes more than doubled at the end of the second quarter. In the third quarter, the Milan processors will finally overtake the supply structure of AMD central processors for server applications.
If in the first half of the year the main volume of AMD server components purchases was realized by cloud segment players, the second half should be joined by general corporate customers. Already now the server segment brings the company more than 20- % of the total revenues, and in the current half-year this share will only grow, as the company's management believes. Graphics processor-based computing gas pedals have more than doubled AMD's profile revenue year on year.
The company has even started shipping the first samples of gas pedals with CDNA 2 architecture, which should come out closer to 2022, as well as offer some more advanced manufacturing process compared to the current 7-nm. In the server CPU segment, revenues have been renewing records for the fifth quarter in a row. Both the growth of the average selling price and the increase in the volume of sales in physical terms contribute to this.
Speaking about the competition on the part of Intel, Lisa Suh stressed that she counts on high activity of the main rival and that is why she sets the task for the company to be better than the competitor. AMD customers are quite excited about the Genoa 5-nm processors, which will come out next year, they are scheduled to be announced according to the earlier schedule.
Price competition from Intel doesn't really change anything for AMD's tactics, as Lisa Su explained. The server segment, she said, values unit performance and total cost of ownership more than the selling price of the processors themselves. In terms of performance, she is convinced, AMD now leads the server processor segment. The company is ready to fight «for every processor socket» but the price factor in this case remains in the background.
AMD's recent success in strengthening its position on the list of the world's most powerful supercomputers has made even skeptics optimistic, and now Citigroup experts predict that by the end of 2022 the company will increase its server processor market share to 20%, which is not too far from historical highs.
AMD management, recall, refused the practice of setting goals to increase its share of the server market by a certain date, but does not discard the idea of the possibility of updating the historical high, which in the mid-twenties reached 26 %, according to rough estimates. Citigroup experts cited by Barron’s see the possibility of AMD increasing its market share in the server segment over at least the next two years. At the beginning of 2019, the company controlled no more than 2.5 percent of the server segment, but in the first quarter of this year alone, that share increased by one and a half percentage points. At this rate, as the authors of the forecast argue, by the end of 2022 AMD could take 20% % of the server processor market. Such dynamics will continue, provided that Intel delays the market launch of its new server processors. The shorter lifecycle of Intel's processors as a result of these delays will undermine the company's revenue, as payback periods will become shorter. Citigroup still does not rule out a price war between Intel and AMD, but now they tie this moment to the cooling demand for central processors for personal computers, which will prompt the first of the companies to act proactively to protect its interests in the server segment.
Tonight AMD will unveil the new, already third generation of EPYC server processors, also known under the codename Milan. The presentation will take place online, which is not surprising given the pandemic. The broadcast will start at 18:00, Moscow time, on the official AMD YouTube channel.
AMD executive director Lisa Su, technical director Mark Papermaster, general manager of embedded systems department Forrest Norrod and other company representatives are expected to speak at the upcoming event. As for the subject matter of the presentation, it is currently known about the EPYC Milan processors that they will carry the Zen 3 architecture, which debuted in the Ryzen 5000 desktop processors last year. In addition to architectural improvements themselves cores new chips will offer a unified third-level cache for each chip (a total of 32 MB for the eight cores).
AMD server innovations will offer up to 64 cores and 128 threads. They will be made in familiar SP3 package, which will allow to use them in already existing systems. TDP level of new chips according to rumors will not exceed 280 W.
AMD introduced the third generation of EPYC server processors, also known by the codename Milan. The new products are based on Zen 3 architecture and are able to execute up to 19% more instructions per clock (IPC) as compared to their predecessors. AMD has presented a total of 19 models of EPYC 7003-series processors. There are chips with the number of cores from 8 to 64, all with multi-threading support & ; each core processes two instruction threads simultaneously. Note that this is the first generation where AMD offered chips with 28 and 56 cores, so users have even more freedom of choice.
The base frequency of the new EPYC processors ranges from 2 to 3.7 GHz. In turn, maximum Boost speeds can reach up to 4.1GHz, although in most cases they do not exceed 4GHz. The standard TDP level ranges from 120 to 280W. Layer 3 cache capacity also varies and may be 64, 128 or 256 MB, depending on the model. Moreover, the amount of cache does not depend on the number of cores. It should be also added that L3 cache is now shared per chip & ; up to eight cores may access all 32 Mbytes, located with them on one chip.
The common features of all EPYC Milan models apart from Zen 3 architecture are eight DDR4 memory channels supporting up to 3200 MHz, and 128 PCIe 4.0 lanes. In fact, this new product does not differ from its predecessors. They are also similar in CPU socket & ; AMD continues to use familiar from the first EPYC socket SP3. Our sister site ServerNews has already prepared detailed review of new AMD EPYC Milan processor family and made tests using EPYC 7763 and EPYC 7543 examples. The material can be found here.
Recommended prices for EPYC Milan processors start from $913 and that is not the 8-core, but the chip with 16 cores. The flagship EPYC 7763 with 64 cores will cost $7890. For comparison, last generation EPYC prices ranged from $450 to $6,950. That is, the chips have not grown too much, but still noticeably.
TSMC already offers services for the production of mobile processors using 6-nm technology. In any case, MediaTek's 6nm processors in off-the-shelf smartphones will hit the market this spring. Now AMD has turned its attention to this process technology, albeit so far in the comments of the management at one of the industry events.
Image source: AMD
The existence of the 6nm process in his speech at the virtual conference Morgan Stanley mentioned AMD CTO Mark Papermaster (Mark Papermaster). According to him, the 7nm technology has not yet reached its full potential, it has a derivative form in the form of 6nm technology. Directly to work with TSMC to master the 7-nm technology AMD began more than five years ago, and the corresponding technological process should demonstrate a long life cycle. In its framework AMD managed to double the number of processor cores without changing their heat pack. TSMC 6nm process implies increasing the number of chip layers manufactured by ultra-hard ultraviolet lithography (EUV) from four to five. That increases the on-chip transistor density by 18 % compared with the 7-nm technology, but also allows developers to easily migrate their designs from the 7- to the 6-nm process.
Image source: AMD
NVIDIA and Intel have yet to officially confirm their interest in using TSMC's 6nm process, and AMD's CTO spoke publicly about this lithography step for the first time this week. It may well be that the company will entrust TSMC to release on 6nm technology of its graphics processors of RDNA 3 and CDNA 2 generation, because in the presentations of AMD them yet no specific process technology & ; only indicated that it is more advanced compared to 7nm. As Mark Papermaster pointed out, AMD products can not master advanced lithographic technology immediately after the mobile processors for smartphones, it takes a certain time to lquo;maturity» process technology. On the other hand, AMD does not hide the fact that EPYC server processors of the Genoa family with Zen 4 architecture will be released on 5-nm technology only in 2022. Thus, the company's 6-nm products have a chance to appear either at the end of this year or early next year.
AMD boldly talks about its approach to the layout of modern processors - it has been using the so-called chiplets for years, and can provide the results of calculations confirming the feasibility of this approach. Interestingly, chiplet processors need more silicon than monolithic ones, but the former are still cheaper to produce.
Image source: AMD, EE Times
AMD representatives managed to publish the next results of calculations during the ISSCC 2021 event, the agenda of which was followed by EE Times website. A year ago, AMD has already explained that the hypothetical 32-core EPYC processor on a monolithic chip would be two times more expensive than the chipset-based model, and the company would not be able to produce a model with 64 cores using monolithic layout.
Image source: AMD, EE Times
Elaborating on the theme, AMD reported this year that a 32-core EPYC processor with four chips consumes about 852 mm2 of silicon, while its hypothetical monolithic chip counterpart would take up 777 mm2. It would seem that the silicon consumption when using chiplets is about 10 % more, but in practice the multicrystal processor turns out to be 41 % cheaper to produce. Firstly, compact crystals produce a lower reject rate, as the risk of on-chip defect concentration is reduced. Second, the processor can be assembled from heterogeneous crystals produced using different lithographic technologies. The most advanced and expensive technological processes are needed to produce crystals with computing cores and other high-speed logic. The auxiliary part, on the other hand, can be satisfied with older production technologies, which are cheaper. Nevertheless, the increase in silicon consumption during the transition to chipsets contributes to the burden on manufacturers and exacerbates product shortages.
Image source: AMD, EE Times
AMD reminds us that it's more than four times more expensive to produce a 250mm2 conventional chip using 5nm technology than it is using 45nm technology. A monolithic 7nm EPYC processor with 32 cores would be twice as expensive as a multichip one. Well, the silicon consumption increase by 10 % in this layout is explained by the additional logic, which is needed to transfer information between the chips and coordinate their work. In the future, AMD will think about the integration of memory chips on a chip with computing cores, as well as the use of higher-density interconnect methods. These ideas, as we know, are to a certain extent shared by Intel.
The statements of Intel's management regarding increased competition in the coming year are still fresh in the minds of industry analysts. AMD CEO Lisa Su believes the competitive environment in the server segment will be business as usual and the average selling price of processors will remain the same as last year.
Image source: AMD
At the quarterly reporting event, AMD representatives once again reminded that major customers have been receiving 7nm EPYC generation Milan processors with Zen 3 architecture since last quarter, and their formal announcement to the general public will take place before the end of the current quarter. At the moment, Lisa Su (Lisa Su) considers the competitive environment to be quite competitive, but does not see the preconditions for increased competition this year. According to her, in the server segment everything will be lquo;as usual». Analysts invited to the conference mentioned the decline in average selling price of Intel server processors, indicating the company is ready to compete with AMD on price. Lisa Su countered that AMD has never set out to compete at the expense of low selling prices for processors, in the server segment it is both total cost of ownership and the balance of price, functionality and performance that matter to customers. According to AMD's CEO Milan server processors are the company's most balanced product in the segment. They are equally as good for the enterprise sector as they are for cloud applications. This year, the bulk of AMD's server processors will continue to be purchased by cloud market players, so the average selling price of EPYCs will not change much. At the quarterly event, it was noted that AMD's revenues in the server segment reached about 19% of total revenues, a sign of the growing impact of server products on the company's business. Server GPUs have little impact on revenues so far, but AMD does not set overly ambitious targets in this sense yet. The expansion of the CDNA architecture and beyond has to be carried out in stages.
AMD is bringing processors based on the Zen 3 architecture to market a little slower than one would expect. So far, desktop and mobile Ryzen 5000s have been introduced, but no HEDT or server carriers of the architecture have been released. Nevertheless, today during a presentation at CES 2021 AMD CEO, Lisa Su, briefly mentioned future Milan generation EPYCs, which will find use of Zen 3 cores.
EPYC (Milan) processors are expected to be the logical successor to the current EPYC (Rome) processors, retaining compatibility with the existing processor socket. What is expected is a change from the Zen 2 core architecture to Zen 3, with a natural 19% increase in IPC, but no dramatic changes in the number of cores, memory support or PCI Express. That the release of the EPYC (Milan) processors is just around the corner was confirmed today at CES 2021 by AMD CEO, Lisa Su. During her presentation, she demonstrated how a dual-processor system with 32 cores of Milan performs when simulating weather phenomena in the dedicated WRF software package, which is widely used in the industry. The simulation speed was compared with a similar task in a dual-socket system with Intel Xeon Gold 6258R processors (28 cores, 2.7–4.0 GHz).
AMD claimed that when simulating a 6-hour weather pattern over the continental US, the AMD system was 68% faster than the Intel system.
As indicated, AMD plans to announce details of next-generation server products and ecosystems in Q1 2021.
Modern AMD processors in server and desktop segments offer no more than 64 cores, and with the new generation EPYC the situation will not change. This does not stop the head of the company to say that AMD processors will have more than 64 cores in the future, although she is not in a hurry to be frank about specific plans in this area.
Image source: AMD, YouTube
Interview with Lisa Su published by Tom’s Hardware and AnandTech in varying degrees, allows to understand that AMD specialists focused on performance enhancement in single-threaded applications, delay reduction and overall system performance improvement during development of 7nm processor generations Zen 2 and Zen 3. The average performance increase was more than twenty percent, while maintaining the same number of cores. As the head of AMD explains, this does not mean that 64 cores is any limit to performance gains. According to her, AMD's processors will have more cores in the future, but the entire system has to scale proportionally so that performance increases evenly. During the interview, Lisa Su also touched on the just introduced Ryzen 5000 mobile processors with Zen 3 architecture. They can be seen combining an advanced compute core architecture with more mature Vega generation graphics. According to AMD's head, the graphics with RDNA architecture these hybrid processors didn't have time to try on due to a tight schedule in preparation for the announcement & ; the Cezanne family was due to debut in the first quarter. AMD's limited engineering resources have nothing to do with such a combination of functional units. We know from unofficial sources that AMD hybrid processors will get RDNA graphics later.
How is the creation of computer graphics in film and animation? How is the rendering farm arranged and how does it work? And why are AMD EPYC processors good for these tasks, but why is the answer to this question not as obvious as it seems? We are analyzing on the example of test servers intensively used by the famous Russian studio of visual effects and animation CGF.
If you immediately remembered the victories of AMD EPYC or Ryzen in CINEBENCH and decided that everything is clear here, then take your time - work on CG-effects and animation in reality is not so easy and the presence of multiple cores still solves nothing. With the support of AMD Moscow office, ASBIS distributor and CGF studio we prepared a material about the studio's subtleties and its experience of trial operation of new AMD EPYC 7002 series processors, which the studio tested to assess the possibility of their further use in the server park.
The studio has hundreds of works on its books - from special effects for feature films to short commercials, as well as fully animated pictures. However, even the so-called VFX, from which the viewer learns what was real and what was not, does not fully reflect the depth and complexity of creating even a few exciting seconds of video, which in reality can take hours or even days. It will not be possible to tell about all the subtleties and nuances, because the studio has its own secrets, but you can get a general idea about the processes.
Working process It is not easy to describe a typical scenario of the studio's work on special effects for a movie, as everything depends on the specific project. In general, the task is to process the original shooting material - add missing objects, delete unnecessary ones or correct the filmed ones. Source video footage from shooting may comprise dozens of terabytes of data. As a rule, such video is stored on a set of LTO tapes. Along with increasing capacity of tape drives, the quality of video, its resolution and color depth are also growing, so one LTO-5 cassette with the size of 1.5 terabytes can hold only 5-10 minutes of recording (without compression).
In the most complex case, everything starts with digitizing physical or simulations of future "artificial" objects in the frame from scratch, as well as simulations, i.e. physical calculations of behavior and interaction of such objects. This is followed by rendering - drawing of the modeled objects. Compositing - combining calculated and drawn objects with the original video at the very end. Behind such formally simple description, of course, dozens of hours of work of a large number of specialists are hidden.
For initial modeling of objects, graphic stations are used, mainly from Autodesk Maya. For simulation, physical calculations, rendering and so on is responsible SideFX Houdini. For compositing the studio uses The Foundry Nuke. For each software has separate additional plug-ins and modules. In addition, there is its own local tools. Each stage is divided into separate tasks, units are quite independent from each other calculations. Almost all these tasks get on the render-farm.
"The problem of many small Hollywood studios is the limited budgets for the purchase of new server farms and workstations. And in the case of CGF, the cycle of updating the equipment, however, quite inexpensive, can be from 4 to more than 7 years. For economic profitability, quality of work and speed of order execution, the studio needs to make the most efficient use of all existing equipment," says Kirill Kochetkov, CGF's CTO studio, under whose leadership testing of new platforms took place.
As a result, CGF render farm is not fully unified, it consists of several types of once expensive blades and server systems. In free time, individual graphic stations of employees are also connected as additional nodes of the farm. And when choosing a new "hardware" it is important to select processors and platforms for specific loads and maintain a balance, because the tasks described above, noticeably different requirements to resources: single and multithreading, IPC core, memory bandwidth, and so on.
Many simulation tasks are very resource intensive - the count time per frame can be several hours. At the same time, there are both scripts that use all processor resources (8-24 threads) and specific tasks that actually use only one thread and for which the frequency and IPC of one core comes to the fore. These include some physical calculations where each next state is directly dependent on the previous one - for example, the disintegration of an object into parts with their subsequent interaction between themselves and with the environment, which they also influence. On average, the rendering itself parallels quite well.
For the majority of tasks, calculations on CPU are used mainly. In some cases, GPUs can provide acceleration, but this is not a universal solution. If a particular project is properly funded, some studios build systems with gas pedals for specific software. In the case of CGF, adding the GPU to the studio's existing farm no longer makes much sense, due to the obsolescence of the server part and the accumulated tools.
The second important resource is RAM. Its consumption depends on a specific task (and even a specific frame), and if it is insufficient, imbalance and slowdown of the overall work may occur. In particular, a task may consume almost all the RAM of a node, loading only a small fraction of the available cores. Finally, the counting time increases by many times when the free RAM of the node is over and it is necessary to use swap files on the disks.
Finally, do not forget about the disk subsystem. The requirements for it also depend on the tasks. Some of them require a large amount of input data, but the output is very small. For others it is exactly the opposite (usually physics simulations) and for others the results of intermediate calculations take up a huge amount of space but the input and output are relatively small. Another nuance is that common caching technologies are not effective in most tasks, as files are usually read only once. Some algorithmic computations can be run very efficiently in the cloud or on a remote cluster, as there is no need to race dozens of terabytes of data on the drives and between nodes and storage.
To simplify the task distribution process, each of them has a specific weight that reflects the resource requirements. Each node of the farm, respectively, has a certain capacity. If, for example, a node has a capacity of 1000 conventional units (c.u.), it can process ten tasks of 100 c.u. each or two tasks of 350 c.u. and two tasks of 150 c.u. each at a time. But a 2000-square-unit task will no longer be able to cope with such a node, and it will not get to it.
"Maintaining a balance of available and required resources allows CGF to efficiently use available hardware capacity. And if the load is not distributed throughout the farm, it is a direct loss of money due to downtime or insufficient use of equipment. Ideally the entire farm should always be 100% loaded, and in practice this is what CGF wants," says CTO of CGF Studios.
Render Farm CGF farm on the basis of blade systems (4U, 10 "blades") made it possible to denser the computational resources on the available areas of the studio. Currently, there are two versions of such nodes based on past generations of Intel Xeon:
dual-processor "blade" in two versions (hereinafter referred to as Blades 1 and Blades 2), but with the same configuration: 2 × Intel Xeon E5645 (6C/12T, 2.4-2.67 GHz, 80 W), 64 GB RAM; The "blade" is more useful, also dual-processor (Blades 3): 2 × Intel Xeon E5-2670 (8C/16T, 2.6-3.3 GHz, 115 W), 128 GB RAM. But that's not all, all the workstations in the studio are used when they are free from interactive use by employees. Their typical configuration includes an 8 or 12 thread high-frequency processor - Intel Core i7-6700K, i7-8700K or i7-9700K - and 64 GB of RAM. The total number of active nodes at peak reaches 150.
Simulation and rendering tasks, model calculation, as well as assembly tasks for the final image are launched on the farm. The rendering farm software runs on the Linux distribution Debian 10. The distribution of tasks is done by the CGRU manager, a popular open source software for rendering farm management. Because there are a lot of tasks and all of them are relatively independent of each other, the farm is ready to accept almost any suitable hardware and there is always something to load it with.
The load on the test server with virtualization and four VMs That's why test servers based on modern AMD EPYC and Intel Xeon Scalable processors were added to this render-farm to assess the prospects of use. First of all, it was interesting to find out how the studio tasks will be performed on the most advanced single-socket AMD server with the maximum number of cores and two double-socket systems with AMD and Intel processors with approximately the same total number of cores per machine.
Unfortunately, at the time of testing it was not possible to get the most similar configurations with the new Intel Xeon Gold 6248R, which have more cores and higher frequency at a lower price (1ku RCP $2 700), and the new AMD EPYC 7F72 (1ku RCP $2 450) with the same number of cores but higher frequency. As a result, configurations with slightly simpler but cheaper processors were taken, which in terms of cost of CPU (see below) seems to be quite acceptable option for testing.
Single-processor server based on AMD EPYC 7702 (hereinafter referred to as EPYC x1):
Gigabyte R162-Z11-00 (1U) platform; 1 × AMD EPYC 7702 (64C/128T, 2.0-3.35 GHz, 180W); 512 Gbytes (8 × 64 Gbytes) DDR4-3200; Seagate FireCuda 520 SSD (M.2, NVMe, 500 GB); 10GbE network controller; 2 × FP 1200W (80+ Platinum). Dual-processor server based on AMD EPYC 7402 (hereinafter referred to as EPYC x2):
Gigabyte R282-Z91-00 (2U) platform; 2 × AMD EPYC 7402 (24C/48T, 2.8-3.35 GHz, 180W); 1 Tbyte (16 × 64 Gbytes) DDR4-3200; Seagate FireCuda 520 SSD (M.2, NVMe, 500 GB); 10GbE network controller; 2 × FP 1600W (80+ Platinum). Dual-processor server based on Intel Xeon Gold 6248 (hereinafter referred to as Xeon x2):
Gigabyte R281-3C2 (2U) platform; 2 × Intel Xeon Gold 6248 (20C/40T, 2.5-3.9 GHz, 150W); 768 Gbytes (12 × 64 Gbytes) DDR4-2933; 2 × Samsung PM883 SSD (SATA, 240GB; LSI RAID); network controller 10GbE; 2 × FP 1200W (80+ Platinum). All machines worked with the performance profile, and for AMD left one NUMA-domain per socket by default. All machines had the same software stack as the one used on the farm; there was no problem with it. Only software was actually on the local disks, while the working data on the external NFS storage was accessed via 10GbE network. All machines received a 1DPC memory configuration with the highest possible frequency for each platform. For the current tasks of the studio, calculated on the farm, this amount of memory is most often redundant, but in the tests it will certainly not be a limiting factor for performance.
Testing On one machine of each type - EPYC x1, EPYC x2, Xeon x2, Blades 1, Blades 2 and Blades 3 - the same test scenarios were run, which include both exclusively synthetic loads, which in practice are only parts of larger tasks, and real calculations of those projects that the studio worked on during the test period of several weeks.
Synthetic tests Physical calculations (simulations) of different types of objects and substances in SideFX Houdini software were used for initial assessment of nodes performance:
Cloth - fabric and clothing; FLIP - fluid; Grain - bulk solids; Pyro - gas; RBD - dynamics of solid bodies; Vellum Cloth and Vellum Hair - dynamics of soft bodies, based on explicit links between points. Such calculations are relatively fast and have different requirements for the number and frequency of cores. For example, the Cloth tool is of single thread nature, so the maximum frequency of a single kernel is important only for it. Other tools know how to parallelize calculations to a greater or lesser extent. In this test, there was exactly one calculation task per node.
The graph above shows the time it took the nodes to calculate each simulation of the above types. Such a large difference between the old and new nodes can be explained to a greater extent those architectural improvements in processors and platforms in general, which occurred over the past decade. If we compare only modern platforms, taking for a basic level of the results of the machine with Xeon, the difference is no longer such a huge, although very significant.
On average, the advantage of systems based on a single processor AMD EPYC is about 20%, and systems based on two processors AMD EPYC - 30%. Solutions with AMD processors are ahead of the competition in all tests, except for the first one, which, as mentioned above, differs by the fact that mathematical calculations in it work in single thread mode. Recall that the Intel Xeon Gold 6248 processor has a turbo frequency of 3.9 GHz, while AMD solutions have only 3.35 GHz. Probably, this explains the 10% gap in this test.
Practical Tests: Simulation and Render The systems under test worked as part of the render-farm of the studio for several weeks, so it was possible to collect quite extensive statistics on tasks, which during this time were processed several hundred. Since, in general, the distribution of tasks by computational nodes was carried out in automatic mode, the most representative tasks were selected for the final report: with a large number of frames and counting time of one frame on the fastest machine not less than one minute.
In the vast majority of cases, servers based on EPYC were faster than the machine with Xeon, and a single socket EPYC-system was on average even slightly faster. On average, the advantage of systems based on AMD compared to the Intel server was 20-21%. But since these are real tasks that include different types of loads, the growth is not uniform everywhere. For example, in one case two AMD processors turned out to be three times faster than two Intel CPUs, in another case - almost twice slower. Such disparity is a direct consequence of the fact that not always single tasks are effectively scaled to a large number of cores.
In this case, it may be more advantageous to distribute "several tasks per node" rather than "one task per node", as in previous tests. To evaluate both approaches, a set of 10 frames for rendering was sent to each test system. In the sequential version, each frame was sent to servers one by one, and in the parallel version, all at once. The result is the total computation time. Testing showed that dual-processor servers with AMD and Intel in this task are about a third faster in case of simultaneous launch of frames for rendering, and single-processor AMD server - about 20%. In terms of overall speed, systems with AMD are leading.
In the next special test for a more accurate study of the paralleling efficiency, we used another rendering task, different from the past, consisting of eight frames. At the same time, the count time of each frame on the nodes of the studio farm (Blades) was usually 10-20 minutes depending on their performance. This task was run on a server with a single AMD processor (EPYC x1) with task management system settings for counting one, two, four and eight frames simultaneously. Thus, the first variant will be sequential and all others will be parallel.
Here you should pay attention to the work of the rendering software. One of the variants of automatic universal paralleling of its tasks is splitting of the target image field into several blocks and calculation of each separate rendering stream. In general, the user can choose for his task how many blocks it can or should be divided. In this case, renderer of each frame does not know that something else counts on the node, and focuses on the number of computational cores to select the number of threads to be launched for block counting.
As a result, it turns out that the simultaneous calculation of eight frames is already actively using the means of processors and the operating system to process the number of threads much larger than the number of cores. The table with the results shows that in this case you can get a good gain - the total time of getting the result is about a third less than if all frames were counted sequentially, although formally in both variants are actively using all processor cores.
Virtualization As mentioned above, for some computing scenarios, using one powerful server with a large number of cores and a large amount of RAM may not be very efficient because of the complexity of managing heterogeneous tasks on it. In particular, this is due to differences in resources such as cores and RAM. The former have a direct impact only on the speed of calculations, but it may be simply impossible to predict and manage memory consumption of heterogeneous tasks.
In this case, you can consider the option of "slicing" one "big" server into several virtual ones with a given allocation of resources. This approach also allows you to adjust resource allocation on the fly, selecting the best option for current tasks. To test this approach, a server with a single AMD processor (EPYC x1) has been installed with an open Proxmox virtual machine management system.
For the EPYC x1 machine - 64 cores/128 threads and 512 Gbytes of RAM - virtual machines (VMs) with configurations ranging from 16 cores and 64,000 MB of RAM in eight pieces to one VM with 128 cores and 512,000 MB of RAM were used. Unlike previous tests with real servers, this scenario may have some performance limitations due to the placement of virtual machine disks on external storage connected via NFS over a 10 Gbit/s network connection.
Total count time VM configuration on EPYC x1 1 frame 2 frames 4 frames 8 frames 8 VM: 16 cores + 64000 Mbytes 00:21:14 - - - - 4 VM: 32 cores + 128000 Mbytes 00:23:09 00:22:03 - - - 2 VM: 64 cores + 256000 Mbytes 00:26:53 00:24:45 00:23:30 - 1 VM: 128 cores + 512000 Mbytes 00:38:30 00:31:01 00:27:28 00:25:27 First of all, it should be noted that the best result in terms of speed is very little different from the best result obtained on a real system without virtualization. This suggests that multi-core AMD EPYC processors are well suited for such scenarios and will be able to cope with high loads. At the same time in virtualization is the best scheme of "8 virtual machines, counting each frame of its own". In general, we can say that solutions with AMD EPYC are most effective in terms of speed exactly in "overloaded" scenarios, when a large number of simultaneous resource-intensive tasks or threads are used.
Power consumption and density Energy consumption and energy efficiency are just as important parameters when evaluating systems as their performance. Old blade systems have their own specifics: A 4U chassis with ten blades consumes about 3 kW (about 260-300 W per node) and requires separate cables for each of the four power supplies. By contrast, test systems are much less demanding - they only need two cables per 1U or 2U node. Comparison of the density and consumption of modern systems and old blade servers is given in recalculation of the 4U-volume rack space - that's how much space each available studio blade system requires.
During the testing, all three servers were configured to remotely monitor consumption using sensors built into the platform. The graph above shows the maximum and minimum power consumption. The former are the highest recorded stable values under load, and the latter were obtained in the absence of load on the servers. The higher efficiency of AMD solutions is probably due in part to a finer process.
The single-processor 1U system with AMD EPYC 7702 is the most interesting in terms of resource density. Four of these systems offer 256 cores with a total consumption of about 1.3 kW. While the old blade system (Blades 3) in the 4U chassis has 160 cores and, as mentioned above, consumption at 3 kW. That is, the modern AMD solution has 1.6 times more cores with more than twice the difference in consumption.
Two dual-processor 2U systems with AMD EPYC 7402 are slightly less dense: 96 cores at 950 W, that is 1.66 times less cores and more than three times less consumption compared to 4U blade (Blades 3). Finally, for the test system with Intel Xeon Gold 6248 under the same conditions of comparison, we get twice as many cores and a little less than three times the difference in power consumption.
In addition, test 1U/2U servers are less cooling intensive compared to blades and can be upgraded with GPUs (provided that the platform supports this feature), which can speed up some calculations or be used to organize VDI. If such versatility is not needed and even higher density is needed, 2U4N nodes can be used.
The cost of . It is worth making a reservation at once that the calculation of the cost of test platforms is approximate, because usually such purchases are much more complex and take into account the entire project, rather than individual machines. ASBIS, which has provided servers for testing, has also provided prices for each of the base platforms (CPU + memory + chassis) in the case of a single purchase. Naturally, if more machines are purchased, the prices will differ.
Estimated cost of configurations, $ EPYC x1 EPYC x2 Xeon x2 Processors 7 435 2 × 2 185 2 × 3 050 Chassis 2 100 3 070 2 350 Memory 8 × 320 16 × 320 12 × 299 Total 12 095 12 560 12 038 The cost of all three platforms has the same order. Both AMD-systems in real load tests proved to be almost identical in performance and noticeably faster than Xeon servers. However, given the higher density and energy efficiency, it is the AMD single socket system that is the most profitable.
If we compare the actually tested systems with theoretically more suitable ones (Xeon Gold 6248R, EPYC 7702P and 7F72), the picture becomes even more interesting. AMD's pricing policy for P-series processors, which are suitable only for single-socket systems, aims to displace the two-socket configurations of Intel - at a lower cost of CPU can get a higher density and / or number of cores. And it concerns even more "simple" models, without the increased frequency and huge cache as in 7Fx2.
Processors, 1ku RCP $ EPYC x1 EPYC x2 Xeon x2 Test option 6 450 2 × 1 783 2 × 3 075 Optimal 4 425 (7702P) 2 × 2 450 (7F72) 2 × 2 700 (6248R) At the same time, AMD consciously does not segment processors by all other parameters: any EPYC 7002 has 8 DDR4-3200 memory channels and 128 PCIe 4.0 lines as opposed to 6 DDR4-266/2933 and 48 PCIe 3.0 lines at Intel. And dual-processor AMD systems may be needed already to provide a higher base frequency per number of cores, when the need for memory capacity greater than 4 terabytes, for certain requirements to the size and memory bandwidth per core, as well as for HPC / AI-systems.
Conclusion One of the main conclusions that the studio made for itself based on the results of the testing is not about "hardware" but software. To be more precise, it is fully compatible with AMD platforms. All software packages used by CGF worked without any problems or additional fine tuning. Including in more complex block and virtualized rendering scenarios. It is fears of software incompatibility that are still one of the psychological, not always reasonable barriers in the choice of hardware. There have also been no compatibility issues with the installation of additional network adapters or drives.
With hardware, AMD's strategy of promoting single socket systems in the case of the studio has hit its target exactly. Test platforms have one order of cost and at the same time much faster than the old blade systems, but the two-socket AMD system is still somewhat more expensive than the others. At the same time, both AMD platforms in the real studio simulation and rendering tasks are on average 20-21% faster than Xeon-based systems. If we take into account that using the P version of the processor in a single-socket AMD server will further reduce the cost, it becomes obvious that this platform is the most attractive in terms of price and performance ratio among all tested.
In addition, it is also more energy efficient and denser in terms of number of cores than other test platforms and old blade systems. Simplifying cabling and reducing electricity bills are undoubtedly very important aspects for the studio. The combination of all these factors led to the fact that even before the completion of all the tests in the rendering farm, the studio purchased another single-socket server based on AMD EPYC 7002 to explore its capabilities in other scenarios of the company's IT tasks.
According to the test results, the studio is considering the possibility of partial updating of the render-farm without a final rejection of blade systems, due to the use of single-processor AMD platforms with a height of 1U, but in a configuration other than the test system. The main factor is still the financials, as the studio's economy is largely tied to the requirements of the orders being fulfilled - the more and more technically complicated the projects are, the higher the requirements to the "iron".
In the current situation there is a need for a small, relatively inexpensive, but fast enough and versatile cluster for everyday studio needs. Including for development of new techniques and capabilities. Single super-heavy tasks of projects by the totality of many factors are most often economically more profitable to "offload" to external sites - large clusters or clouds.
"How much to hang in cores? The most optimal for us now are 1U systems with AMD EPYC 7502P 32-core processors (2.5-3.35 GHz, L3-cache 128 Mbytes, TDP 180 W). They strike a balance between cost, including chassis and memory, density, power consumption, performance and versatility to ensure efficient studio operation in the current environment," concludes CGF CTO.
AMD has so far not been very happy to talk about processors with Zen 4 architecture. At the October presentation of Ryzen 5000, it was noted that they are currently under development, and will be released on 5nm technology closer to 2022. As it turns out, the performance improvements of Zen 4 processors will be achieved by a combination of factors, as in the case of Zen 3.
Image Source: AMD, YouTube
This was stated by Rick Bergman, AMD's Executive Vice President of Computing and Graphics Products, in his interview to TheStreet. He noted that the company needed to improve the Zen 3 architecture in several ways to achieve a 19 % increase in specific performance. In the case of Zen 4, the list of improvements will be as long as he assured. Branching prediction, cache structure and length variation of the executive pipeline will all be in line with the increased energy efficiency of the processors enabled by the transition to 5nm manufacturing technology. Bergman made it clear that even with the 7nm technology, AMD processors will be subject to intermediate improvements; at least because TSMC is improving its lithographic technology. Switching to EUV lithography for AMD is not a problem, but it does not provide any visible benefits in terms of processor performance. In fact, as Bergman explained, EUV lithography brings more benefits to TSMC. Speaking about the 7nm capacity shortfall of the TSMC, AMD's representative admitted that demand hasn't yet been fully met, but there is a lot of effort and 2021 can make a difference in this sense. Ryzen Threadripper processors will continue to evolve, but Bergman did not specify when they will switch to the Zen 3 architecture. Functions to accelerate AI systems will be implemented by AMD in their own processors, and over time they will appear even in consumer solutions. As in the case of RDNA 2, when creating RDNA 3 graphics solutions the company is going to pay close attention to improving specific performance per watt of power consumption. At the same time, AMD graphics solutions to use 5nm technology will move to comparable terms with CPUs, which actually indicates 2022. Since the use of EUV and advanced technology processes in the production of absolutely all components at a particular time may have limited economic feasibility, AMD will continue to develop layout technology. Three-dimensional layout will first be tried on by the most expensive and productive products of the brand, but in time it will reach the desktop segment.
