Showing posts with label graphics processors. Show all posts
Showing posts with label graphics processors. Show all posts

11.29.2022

NVIDIA took an even larger share of the discrete graphics market - Intel and AMD have 12%

NVIDIA took an even larger share of the discrete graphics market - Intel and AMD have 12%

NVIDIA took an even larger share of the discrete graphics market - Intel and AMD have 12%

The other day, thanks to a report from Jon Peddie Research (JPR), we learned that shipments of all types of graphics processors collapsed by 25% in the third quarter.
Now, thanks to the same analysts, we know how the discrete graphics card market changed over the same period.
In short - it fell even more, and AMD gave NVIDIA a serious share of this market.Image source: WCCFTechThe discrete graphics market sagged very strongly.
In the third quarter of 2022, the supply of graphics cards was only 14 million units, a drop of 41.6% compared to the same period last year, when the supply was 24 million graphics cards.
Even a quarter earlier the situation on the market was much better - in the second quarter of 2022 was delivered 19 million discrete graphics cards.Shipments of mobile and desktop graphics cards were divided equally - in each segment was delivered 7 million cards.
However, desktop cards previously dominated the market, with 10 million desktop boards shipped in the quarter versus 8.6 million mobile boards, and 13 million desktop boards versus 11.6 million mobile discrete graphics cards shipped in the third quarter of 2021.
NVIDIA saw its share of the discrete graphics market grow to 88% in the third quarter, up 8 percentage points from the quarter before.
As a result, AMD had only 8 percent of the discrete graphics market, down from 15 percent in the second quarter.
And the remaining 4 percent was taken by Intel.
In the desktop segment, AMD is doing slightly better - here it took 10 percent.
The remaining nine tenths of the market were divided between NVIDIA with 86% share and Intel with the same 4%.
As for the mobile segment, NVIDIA's dominance is even stronger - the company owns 90% of the market, while Intel and AMD equally shared the remaining 10%.Overall supply of graphics processors, both discrete and embedded, both mobile and desktop, is rapidly decreasing since the second quarter of 2021.
Then a sharp decline occurred in the third quarter, then in 2022, the market was relatively stable until the first quarter of this year, but after that again began to actively decline.
This is likely due to the economic and geopolitical instability in the world.Analysts expect that next year, sales of graphics processors will fall even more.
In 2024 the market will be steadily low, and the recovery will begin only in 2025.
But the growth will not be too significant.
These are tough times for the PC market: demand for PCs soared during the pandemic, and now the need for them is severely diminished.
The GPU market has been heavily influenced by mining, which has driven high discrete graphics sales.
Now the market is flooded with cheap used video cards from miners, and there are a lot of new gas pedals in stores now, but some are asking too much.

11.07.2022

AMD explained the structure of graphics processor Navi 31 and clarified the number of stream processors in its composition

AMD explained the structure of graphics processor Navi 31 and clarified the number of stream processors in its composition

AMD explained the structure of graphics processor Navi 31 and clarified the number of stream processors in its composition

At the disposal of VideoCardz portal appeared the image block diagram of AMD Navi 31 graphics processor, which formed the basis of the new graphics cards Radeon RX 7900 XTX and RX 7900 XT.Image source: AMDThe AMD Navi 31 graphics processor uses a chiplet design.
In the center of the chip is a large GCD (Graphics Compute Die) crystal based on the 5-nm process, which includes all the GPU compute blocks and other necessary elements.
Around it are six cache memory crystals (MCD).
They are built using the 6nm process.
Each MCD chip includes a dual 32-bit memory controller combined with L3 cache memory.To date, AMD has only confirmed that the full-size Navi 31 GPU contains 96 MB of L3 cache memory (Infinity Cache), but the manufacturer did not specify the size of L0, L1 and L2 cache memory.
This information is clarified by a presentation slide which appeared to VideoCardz and to all appearances was shown to journalists at the presentation behind closed doors.
It says that L0 cache memory is 3 Mbytes, which is almost 2.5 times larger than in Navi 21.
L1 cache memory is three times larger and is equal to 3 Mbytes.
L2 cache memory is up 50% to 6 Mbytes.
L3 memory (Infinity Cache) is 96 MB, 32 MB less than that of Navi 21.
Image source: VideoCardz / AMDThe diagram above shows Navi 31 with six shader engines using eight pairs of Compute Units.
The Radeon RX 7900 XTX was originally supposed to have 12,288 stream processors.
However, if you turn to the company's website, we can note the indication of the presence of 6144 stream processors, that is half as many.
The reason for this is that RDNA 3 architecture is based on Dual Issue design, which means that it can execute two FP32 arithmetic instructions simultaneously, so we can talk about 12,288 FP32 blocks or thread processors.
Meanwhile, the Navi 31 has upped its performance in single precision floating point operations (FP32) to 61 Tflops, more than doubling over the previous generation.The slide also mentions that the RDNA3 architecture is designed to operate at 3 GHz.
The same information comes from journalists who visited the presentation of new graphics cards and discussed this issue with AMD representatives on the sidelines.
However, it should be noted that the maximum Boost frequency of the reference Radeon RX 7900 XTX is claimed to be 2.5 GHz.
That's what it says on AMD's website.
How the company is going to implement the 3 GHz support in its RDNA 3 graphics cards is not known yet.
Image source: Twitter / @Locuza_ Above you can see an actual photo of the Navi 31 processor with annotations of each chip block added on top, provided by one Twitter user.