Shulou(Shulou.com)11/24 Report--

Thanks CTOnews.com net friend net said do not like to listen, ysl clue delivery! CTOnews.com November 4 news, after half a year of warm-up, AMD finally unveiled its RDNA 3 architecture, with the arrival of the Radeon RX 7900 XTX ($999) and Radeon RX 7900 XT graphics cards ($899) cheaper than the previous generation, A cards seem to have a chance to compete with N cards again.

First of all, most of the details are consistent with AMD RDNA 3 architecture and RX 7000 series expected memory that we have previously revealed. RDNA 3 will use TSMC's 5nm chip, which contains a main GCD (Graphics Computing Module) and six MCDs (Memory Cache Modules).

In addition, RDNA 3 also has a lot of underlying improvements in architecture, such as more compute units and more GPU shaders.

According to AMD, this generation of graphics cards continues to focus on power and energy efficiency, with RDNA 3 reportedly achieving 50% higher performance per watt than RDNA 2. This is a bit different from the NVIDIA RTX 4090, which has increased power consumption and compression performance. Foreign media think AMD should feel no need to raise the power consumption to that height by default.

Let's take a quick look at the core specs and compare them laterally to the previous generation RDNA 2 and Nvidia RTX 4090.

Graphics RX 7900 XTXRX 7900 XTRX 6950 XTRTX 4090 RTX 4080RTX 3090Ti Architecture Navi 313121AD102AD103GA102 Process TSMC N5 + N6 TSMC N7 TSMC 4N Samsung 8N Transistor Quantity (million) 580580 - 1MCD268763459283 size (mm2)300 + 222300 + 185519608.4378.6628.4SM / CU / Xe9684801287684GPU kernel (shader) 1228810752512016384972810752Tensor kernel///512304336 Ray tracing kernel 9684801287684 Acceleration frequency (MHz) 230020002310252025051860 Memory Speed (Gbps)20? 20？182122.421 Memory Size (GB)242016241624 Memory Bus Width 384320256384256384L2 /Infinite Cache 968012872646ROPs192192128176112112TMU384336320512304336TFLOPS FP32 (Boost) 56.543.023.782.648.740.0TFLOPS FP16 (FP8) 1138647.4661 (1321)390 (780)160 (320) Bandwidth (GBps)960? 800？57610087171008TDP (Watt) 355300335450320450 Date December 2022 December 2022 May 2022 October 2022 November 2022 March 2022 Guidance Price USD 999 USD 899 USD 1099 USD 1599 USD 1199 USD 1999 USD National Bank Price Conversion About RMB 7293 About RMB 6563 CTOnews.com learned that AMD RX 7900 XTX did not learn NVIDIA, or traditional 8Pin interface, power consumption 355W only two 8Pin, and RX 7900 XT 300W is also two 8Pin.

AMD has two versions of the Navi 31 GPU, with the higher spec model named RX 7900 XTX, which has full GCD and six MCDs enabled, while the RX 7900 XT has only 84 of the 96 compute units and five MCDs enabled.

Technically speaking, the sixth MCD of RX 7900 XT may only be physically shielded, but it cannot be enabled by itself anyway, and the card itself does not have an additional 4GB GDDR6 memory for it, so don't think about opening the core.

Compared with competitors, RX 7900 XTX still lags behind RTX 4090 in terms of computing power. After all, Nvidia also has Tensor support, which can achieve stronger AI processing capability.

Considering that the RX 6950 XT has almost caught up with the RTX 3090 Ti at 1080p and 1440p (4K is about 5% behind), it is theoretically 40% less raw computing performance. So, on paper, the RX 7900 XTX probably doesn't represent much with 32% less computation than the RTX 4090.

As for the other model, the RX 7900 XT should not be used against the RTX 4090, we just need to consider how it will compete with the RTX 4080.

In terms of parameters, it has more memory and higher bandwidth, plus 16% computing power advantage. So even if the performance of the RDNA 3 shader drops (see below), AMD's next generation card should look very competitive, at least on paper, against Nvidia's sub-flagship, and it's $200 lower.

AMD has repeatedly emphasized its energy efficiency advantages over the past two generations of RDNA architecture, and RDNA 3 continues this focus.

AMD claims a 54% improvement in performance per watt compared to RDNA 2, which has a 54% improvement in PPW over RDNA. AMD efficiency has risen almost exponentially over the past three generations.

We took the previous generation RX 6900 XT as an example, its performance is basically twice that of the previous generation RX 5700 XT (1440p super resolution), while the measured power consumption of foreign media is about 308W, while the 5700 XT is 214W, so the efficiency should be improved by 38%.

So in the face of the RTX 40 series, how should AMD continue to improve efficiency? Much of this may be due to TSMC's iterative evolution from N7 to N5 (7nm to 5nm), but architectural updates certainly helped a bit.

The new generation RDNA 3 unifies compute units, each with 64 dual stream processors (GPU shaders). That's twice as much as RDNA 2, and AMD can either send a different workload to each SIMD unit, or it can make both work together on the same type of instruction to increase efficiency.

Notably, the latest AMD, Intel, and Nvidia GPUs are now all 128 shader designs-AMD's Compute Unit (CU), Nvidia's Streaming Multiprocessor (SM), Intel's Xe Vector Engine (XVE).

Not only did AMD double the number of GPU shaders in CUs, the total number of CUs increased from 80 to 96. AMD's Navi 31, by contrast, has 2.4 times as many shaders as the Navi 21, while consuming only 18 percent more power.

AMD has also improved the performance of its AI accelerator, an area where the company rarely talks about being crushed by Nvidia. We can't say for sure how powerful the new model is, but AMD says its AI accelerator supports INT8 and BF16 (brain floating point 16-bit) operations. Therefore, they may have at least some Tensor cores similar to Nvidia, but the total number of instruction sets supported is not the same.

But anyway, AMD says its new AI accelerator offers 2.7 times the improvement over the previous generation-twice the number, more processors, and slightly higher throughput.

Finally, AMD says they've optimized the ray tracing accelerator, RDNA 3 for 1.5x better ray tracing performance, new dedicated instructions and improved BVH (light chase/box) sorting and traversal, which isn't entirely clear, but we're sure the new card is a huge leap forward in ray tracing and improved rasterization performance.

In addition to the compute unit, RDNA 3 also offers major upgrades to many other areas, such as AMD Radiance display engine, or basic video output support.

AMD also upgraded RDNA 3 to support DisplayPort 2.1 (NVIDIA (*^_^*)), which means AMD is now the second GPU company to support DP2.x (Intel Arc is the first).

Intel only supports 10 Gbps per lane or a total of 40 Gbps bandwidth, while DP2.1 actually supports 20 Gbps per lane or a total of 80 Gbps bandwidth.

However, AMD currently does not support 20 Gbps, but chose the intermediate level of 13.5 Gbps, so AMD's new generation GPU can achieve a total bandwidth of 54Gbps, which is basically twice that of DP1.4a.

Thanks to DSC (Display Stream Compression) technology, AMD can support refresh rates up to 480Hz on 4K resolutions or up to 165Hz on 8K displays using its DisplayPort 2.1 interface. It is worth mentioning that the first DisplayPort 2.1 monitors and TVs are expected to be available in early 2023.

AMD has also significantly improved its media engine with RDNA 3. This has been more or less exposed, but the Navi 31 surprise is that it has a dual media engine that is fully capable of supporting two 8K60 streams running simultaneously-either encoding, decoding, or in parallel/combination to improve the performance of encoding a single stream.

Another update to the video engine is support for AV1, which means that all three GPU vendors have since provided full hardware encode/decode support for AV1. Hopefully we will see more software solutions and streaming services supporting AV1 instead of H.264.

In addition, the frequency of the video engine is higher than before (we are not sure how much higher), AMD also pointed out that it has AI-enhanced video encoding capabilities, but the specific also needs to wait for the follow-up hardware evaluation to do some video encoding tests, so stay tuned.

AMD also gave some initial performance figures for the Radeon RX 7900 XTX, such as Call of Duty, Cyberpunk 2077, Watchdog, Resident Evil 8, etc., which are approximately 50% to 70% higher than the previous generation RX 6950 XT.

We cannot guarantee the accuracy of the above performance claims, but AMD definitely chose these games for a reason.

AMD once again introduced its Fidelity FX Super Resolution (FSR) technology. This is a cliché, so let's skip it.

AMD currently has more than 216 games and apps that support FSR, but most of them are FSR 1.X, which is open source after all, so it's easy to integrate, and it's been more than a year, giving developers longer time to prepare, while FSR 2.0 is much younger, released in May this year, and currently only available in a few games.

Looking ahead, AMD's FSR will continue to be "attractive." We certainly hope that FSR2 penetration rate can exceed FSR 1.x, after all, it can bring a better experience for users, no matter what platform, and the game can also support both technologies, perhaps in some scenarios (such as low-end graphics and integrated graphics) FSR1 is more suitable for some users.

FSR3 technology is said to be released sometime next year, and it will seek some form of frame padding or interpolation, somewhat similar to the direction Nvidia's DLSS 3 does. AMD didn't reveal much detail, partly because FSR3 hasn't quite figured out what to do next, but in early tests it was able to double the number of frames in some games.

Both graphics cards will go on sale Dec. 13, with the RADEON RX 7900 XTX priced at $999 and the RADEON RX 7900 XT priced at $899.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.