Intel has revealed promising Falcon Shores chips - they will allow you to choose the ratio of CPU to GPU

Intel, along with the announcement of Rialto Bridge computing gas pedals and the disclosure of new details about Sapphire Rapids-SP HBM, has also revealed the upcoming Falcon Shores server chips.
They will combine CPU and graphics cores, which is extremely atypical for the server segment, where GPU computing gas pedals have mostly been produced separately from CPUs.Some time ago, Intel came to the conclusion that chips should be made from universal blocks, rather than developing each one from virtually scratch.
After all, it is possible to take CPU cores, GPUs, AI gas pedals, memory, I/O interfaces and other necessary elements and combine them in the right proportions.
This was the approach used to create Alder Lake chips, which is why there are so many of them and so diverse.
For the server segment, Intel also wants to offer freedom of choice.
Of course, it is unlikely that customers would be able to order chips with specific CPU to GPU ratios from Intel themselves, but the company would be able to offer a very wide range and everyone would be able to find the best match for their needs.
Customers will be able to choose, for example, solutions with a large number of processor cores, where graphics play a supporting role, or vice versa - a lot of graphics, little CPU.
And what is important, it will all be in the same packages (cases), which simplifies the construction of servers.What Intel promises will be seriously different from the current approach, because each Falcon Shores chip is a computational gas pedal.
These solutions promise significant performance gains for supercomputers and other HPC systems.
As of now, Intel promises that the Falcon Shores solutions will deliver more than a 5x increase in performance per watt of power, as well as a more than 5x increase in computational density, and more than a 5x increase in memory capacity and bandwidth.
Intel does not specify what process technology will be used at Falcon Shores but says it will be angstrom-level technology (an angstrom is a tenth of a nanometer).
Recall that Intel plans to move to 20A technology in a few years - that is, about 20 angstroms or 2nm.
Falcon Shores solutions will not come out before 2024.

Intel announced work on Rialto Bridge, a 160 Xe-core computing gas pedal that will replace the Ponte Vecchio

Intel today unveiled a host of new products for high-performance computing and also lifted the veil on some future developments.
In particular, it was announced that computing gas pedals called Rialto Bridge are in the pipeline to replace Intel Ponte Vecchio gas pedals as early as next year.Essentially, Rialto Bridge will be an improved version of Ponte Vecchio - the novelty will offer more Intel Xe graphics cores, more processing power, higher input-output bandwidth, be able to do more gigatransactions per second (GT/s) and offer other improvements.
The next-generation gas pedals will also consist of multiple crystals, and some of them will inherit directly from their predecessors.So far, Intel has said that Rialto Bridge will get up to 160 Intel Xe cores, 25% more than the current Ponte Vecchio (128 cores).
At the same time, it is not specified what kind of cores will be the same as in the current Intel GPU or new cores.
Also, it is not reported what new items will have memory, but most likely a new HBM3.Also, it was said that the new items will be available in the form factor OAM v2.
Current Ponte Vecchio gas pedals also use OAM form factor.
The new version of the packaging will provide even more power to the chips.
The slide also mentions that the new product will be made as part of the IDM 2.0 strategy.

Crush - "warm-up" supercomputer with AMD EPYC and Instinct MI250X

Oak Ridge National Laboratory (ORNL) is often mentioned in news articles as actively testing and implementing new supercomputing technologies. The OLCF (Oak Ridge Leadership Computing Facility) is currently in the process of installing the first exascale Frontier supercomputer in the US based on AMD processors and GPU.

Now a number of architectural features of Frontier have come to light, as the National Center for Computational Sciences' Crusher small cluster, which uses virtually the same HPE Cray nodes as Frontier, has come on line. The system is used as an early access platform and consists of just two cabinets. The first has 128 nodes and the second has 64. Total peak performance is claimed to be 40 Pflops.

The heart of each node is a dedicated version of the AMD EPYC 7A53. The 64 cores (with SMT2) are split into four NUMA domains served by separate memory controllers. Eight DDR4 channels (512 GB total per node) provide 205 GB/s of throughput. There are only four gpus in the system, but they are the latest AMD Instinct MI250X dual-chip, so the system sees them as eight separate GPUs. Each of the gpus is connected to a single NUMA domain via two Infinity Fabric channels, providing 36 GB/s in each direction. The chips inside the MI250X are connected to each other via a faster channel, giving 200 GB/s in both directions. All of the gpus are connected to each other in a «each-to-each» 50-Gbyte/s channel arrangement. At the same time they are directly connected to the factory & ; each one is entitled to its own HPE Slingshot adapter (200 Gbps). Only a pair of 1.92 Tbytes SSDs (4 Gbytes/sec write, 1.6 million IOPS in random operations) are connected to the CPU via a PCIe switch. Each NUMA domain is divided into two L3 sub-domains connected to a single gpu, which allows for flexible load balancing. The primary storage is an external IBM Spectrum Scale storage system with a total capacity of 250 Pbytes and a peak rate of 2.5 Tbytes/s.

Future Frontier supercomputer

The system also has access to the NCSS network, though not directly. In general NFS storage each project can get 50 Gbytes with a storage time of 90 days, and in GPFS on Spectrum Scale 50 Tbytes are already available. Crusher comes with a lot of preinstalled software. The user environment is modular and based on the Lmod system written in Lua. Load balancing is handled by Slurm. For authentication the RSA SecurID hardware token is used.

AMD BC-160 mining video cards are now available to everyone - 72 Mhash/s for 150 thousand rubles

Last October, information about the mysterious AMD BC-160 video card designed for mining appeared on the web. Now this gas pedal can buy anyone who has 150 thousand rubles - a novelty appeared on Aliexpress. And if the claimed characteristics of it are true, it's not even unreasonable.

A key feature of the AMD BC-160, the sellers claim, is the combination of high mining performance and low power consumption. It is claimed that the performance of the gas pedal is above 72 Mhash/s for Ethereum mining, while the power consumption does not exceed 130W. This can provide revenue above $4.5 per day. The performance of the new product is higher than that of the Radeon RX 6900 XT (around 64 Mhash/s) and comparable to that of the NVIDIA GeForce RTX 3080 LHR after "unlocking" (around 70 Mhash/s).

According to the product description, the AMD BC-160 gas pedal is built on the 7nm Navi 12 GPU in a 36 compute block version, which means 2304 stream processors. Earlier AMD released only two graphics cards on this GPU: mobile Radeon Pro 5600M for Apple MacBook Pro, and server AMD Radeon PRO V520 for cloud services. The Navi 12 GPU is equipped with 8 GB of HBM2 memory with a 2048 bit width interface. The frequency of memory, as well as the GPU itself, are not specified. However, it can be assumed that the memory runs at an effective frequency of 2000 MHz, resulting in a bandwidth of 512 GB/s. Actually, the high memory bandwidth provides a very high performance in Ethereum mining.

We should add that the BC-160 boasts a PCI Express 4.0 x16 interface as well as a pair of 8-pin auxiliary power connectors. Nevertheless, it is claimed that in mining its power consumption is only 130W. The cooling system with a tangential fan ("turbine") is responsible for heat dissipation. The AMD BC-160 graphics card is offered on Aliexpress at a price of 151,670.13 rubles. The sellers promise to deliver it to Moscow in a few weeks, and also mention a full warranty.

Latest generation AMD and NVIDIA graphics cards have fallen to their lowest price since February in Germany, but are still expensive

Over the past two weeks, there has been a sharp drop in the price of graphics gas pedals. According to 3DCenter resource, the price of graphics cards based on graphics chips with NVIDIA Ampere and AMD RDNA2 architectures decreased to the level of February this year.

The 3DCenter report uses data on the cost of graphics cards in the German market, but in general the given statistics is relevant for other European countries as well as reflects global trends. The data for the report is collected from 10 German stores and then compared month by month. At the moment, the cost of graphics processors is on average 53% higher than the recommended retail price. While this is nothing positive for users, it is in fact the lowest price since February of this year, when the market began the price spike. More importantly, new shipments of GeForce RTX 3060 Ti and Radeon RX 6800 have started to appear in stores, which has not been the case for quite some time. It may take a few more weeks to return to normal pricing, as retailers and distributors are reluctant to adjust prices downward in an effort to get the most out of the current market situation. Note that large U.S. retailers that work with PCPartPicker (PCP), including BestBuy,  Newegg,  Amazon and others have not seen such steep price declines. Unlike the 3DCenter report, PCP aggregates data from sites that offer affiliate links. However, it does take into account various promotions such as coupons, discounts and free or more affordable shipping. The charts below show prices for the GeForce RTX 3060 and Radeon RX 6800 XT, respectively.

AMD Instinct MI200 Accelerator gets 128 GB of HBM2e memory

AMD's next-generation AMD Instinct (MI-Next) computing gas pedals will feature unprecedented memory capacity & ; up to 128 GB per GPU. The announcement was made during the ISC 2021 HPC event, where Pawsey CTO Ugo Varetto talked about Setonix supercomputer specs.

The supercomputer, made possible by a $70 million investment, will allow Pawsey to improve the speed and processes of processing, data visualization and help improve other information operations. Processing large arrays will require both large amounts of memory and powerful graphics processors to increase system efficiency. The system will get more than 200,000 AMD Milan processor cores (Zen 3 architecture) as well as more than 750 AMD MI-Next compute gas pedals. The flagship Model of the MI-Next series, namely the Instinct MI200 will be the fastest performance gas pedal produced by AMD to date. The GPU, codenamed Aldebaran, has an MCM layout, meaning it combines multiple crystals on a single substrate. According to Pawsey, the GPU has 128 GB of HBM2e memory & ; four times more than the current Instinct MI100 gas pedal. The network has already published visualization of the circuit GPU MI200 c eight HBM2e memory stacks, associated with both crystals of the GPU:

AMD still hasn't confirmed the MI200 gas pedal specifications, but judging from the leaks, the full Aldebaran variant will have 128 compute modules, which means there are 8192 stream processors. The number of active modules in the MI200 is not yet known, but it is likely that a full GPU will be used here.

The market is flooded with next-generation NVIDIA and AMD graphics cards, but you should not buy them

Former cryptomainers from China began to throw on the market wholesale shipments of next-generation graphics cards. Nevertheless it is hardly worth buying «new» GPUs on the secondary market, at least & ; you should not hurry.

According to Hong Kong sources, the local «bulletin board» PTT has seen all sorts of options & ; from NVIDIA GeForce RTX 3090 and AMD Radeon RX 6900 XT to NVIDIA GTX 1066 and AMD Radeon RX 478 (aka NVIDIA GeForce GTX 1060 and AMD Radeon RX 470 respectively) in large numbers. «Repression against cryptominers in Inner Mongolia, Qinghai and Sichuan, as well as other provinces of China and more aggressive policy of the PRC authorities on the cryptocurrency market in general has already led to a drop in cryptocurrency prices around the world. China accounts for 65% of the global hash rate of cryptomainers mining currencies like Bitcoin and Ethereum.

techradar.com.

Now, unwanted used cards are being sold at manufacturer-recommended prices & ; in some cases, even cheaper than what the cryptomainers themselves paid for them. Moreover, this is the worst time to buy video cards on eBay and similar marketplaces. With the fall in cryptocurrency prices, the market runs the risk of a wave of used GPUs. It is hardly worth buying under any circumstances hardware that has been exploited for months in 24/7 mode, worked in many cases in conditions with elevated temperatures and without good cooling. In addition, many of these cards will be sold on the trading floors as new & ; trust such ads is not yet worth it, even in the presence of boxes and documents for the purchase. But most importantly, the flow of video cards from Chinese farms will force homegrown speculators, who have purchased really new copies in large quantities, to reduce prices, and this will undermine their business, which often involves unscrupulous representatives of trading networks. Therefore, it is possible that soon modern GPUs will be available at reasonable prices in stores.

China's Innosilicon has unveiled FANTASY GPUs that are 'as good' as those from AMD and NVIDIA

Chinese company Innosilicon & ; one of the world's largest specialists in ASICs and systems for miners & ; peculiarly congratulated everyone with the coming New Year. The company announced that it will soon start shipping discrete graphics cards based on FANTASY GPUs. These will be data centre solutions «not worse than AMD and NVIDIA offerings. But details have not been announced by the developer yet.

Innosilicon FANTASY GPUs were first unveiled a month ago at China's first global 5G conference. The conference was held in Guangzhou from November 26 to 28. According to the company, «China's first data centre GPU demonstrated by INNOSILICON was the centre of attention and a milestone in the technological breakthrough in major 5G applications». «The FANTASY processor is ideally optimised for graphical big data applications such as multi-server cloud desktop, cloud gaming, cloud office, etc. The FANTASY data centre GPU enables deployment of 5G infrastructure using the most advanced SOC innovations, such as chipset and high-speed GDDR6» & ; says Innosilicon in a press release.

If you remove all the «water» from the press release, we are talking about GPUs based on Imagination's high-performance BXT graphics cores. Imagination introduced new graphics cores in October of this year, and Innosilicon has reached an agreement with the company to use the most powerful of the proposed cores in GPUs and discrete graphics adapters for the high-performance computing market. We await details!