We tested AMD Radeon RX 7900 XTX and Radeon RX 7900 XT. You can enjoy the full results one day before his official launch party. AMD’s latest and greatest RDNA 3 architecture and the RX 7900 series graphics cards everyone is ready to party Nvidia’s Ada Lovelace Architecture and the GeForce RTX 4080 When they vie for a spot on our list best graphics card.
One thing AMD doesn’t do: GeForce RTX 4090 it’s on top of us GPU Benchmark HierarchyAMD says it doesn’t need to compete directly with $1,600 (or more!) graphics cards, but some AMD seniors say Nvidia’s AD102 chip is “bigger than expected” I point out that it is basically out of reach.
Out of reach for many readers is the new AMD RX 7900 series cards. They’re cheaper than Nvidia’s RTX 4090 and 4080, with prices starting at $899 for the slower and frankly undesirable RX 7900 XT, but these are clearly not chasing the mainstream gamer market. applies to future Navi 32 / RX 7700 series cards (or possibly 7800 series). For now, if you want the fastest consumer graphics card AMD has ever made, be ready with plenty of cash.
We’ve already covered the RDNA 3 architecture and card previews in detail. With the actual hardware in hand and a set of benchmarks under our belts, that’s the main event today. I have some additional thoughts though, so as usual I’ll start with AMD’s latest card specs and for comparison he’ll use Nvidia and some previous generation GPUs.
| graphics card | RX7900XTX | RX7900XT | RX6950XT | RTX4090 | RTX4080 | RTX 3090 Ti | RTX 3080 Ti |
|---|---|---|---|---|---|---|---|
| architecture | Navi 31 | Navi 31 | Navi 21 | AD102 | AD103 | GA102 | GA102 |
| process technology | TSMC N5+N6 | TSMC N5+N6 | TSMC N7 | TSMC 4N | TSMC 4N | Samsung 8N | Samsung 8N |
| Transistor (billion) | 45.6 + 6x 2.05 | 45.6 + 5x 2.05 | 26.8 | 76.3 | 45.9 | 28.3 | 28.3 |
| Die size (mm^2) | 300 + 222 | 300 + 185 | 519 | 608.4 | 378.6 | 628.4 | 628.4 |
| CU/SM | 96 | 84 | 80 | 128 | 76 | 84 | 80 |
| GPU shader | 12288 | 10752 | 5120 | 16384 | 9728 | 10752 | 10240 |
| AI/Tensor Cores | 192 | 168 | 80 | 512 | 304 | 336 | 320 |
| Ray tracing unit | 96 | 84 | 80 | 128 | 76 | 84 | 80 |
| Boost Clock (MHz) | 2500 | 2400 | 2310 | 2520 | 2505 | 1860 | 1665 |
| VRAM Speed (Gbps) | 20 | 20 | 18 | twenty one | 22.4 | twenty one | 19 |
| VRAM (GB) | twenty four | 20 | 16 | twenty four | 16 | twenty four | 12 |
| VRAM bus width | 384 | 320 | 256 | 384 | 256 | 384 | 384 |
| L2 cache | 96 | 80 | 128 | 72 | 64 | 6 | 6 |
| ROP | 192 | 192 | 128 | 176 | 112 | 112 | 112 |
| TMU | 384 | 336 | 320 | 512 | 304 | 336 | 320 |
| TFLOPS FP32 | 61.4 | 51.6 | 23.7 | 82.6 | 48.7 | 40 | 34.1 |
| TFLOPS FP16 (FP8/INT8) | 123 (123) | 103 (103) | 47.4 | 661 (1321) | 390 (780) | 160 (320) | 136 (273) |
| Bandwidth (GBps) | 960 | 800 | 576 | 1008 | 717 | 1008 | 912 |
| TBP (Watts) | 355 | 315 | 335 | 450 | 320 | 450 | 350 |
| Release date | December 22nd | December 22nd | May 22nd | October 22nd | November 22 | March 22 | June 21st |
| launch price | $999 | $899 | $1,099 | $1,599 | $1,199 | $1,999 | $1,199 |
There’s a lot to reveal about the specs, but we’ll mostly focus on AMD’s new chips. The RX 7900 XTX has a fully enabled Navi 31 GCD (Graphics Compute Die) and 6 MCDs (Memory Cache Dies), while the 7900 XT uses 12 Compute Units (CUs) in the GCD. Disabled and one of the MCDs is fused. Technically, there are still 6 MCD chips to even out the mounting pressure from the heatsink, but one of them is melted (possibly a non-functioning MCD).
The number of GPU shaders is a bit different than other architectures. AMD says there are still 64 Streaming Processors (SPs) per CU, but there are also 4 SIMD32 vector units per CU, two of which can only handle FP32 or matrix operations, and INT32 cannot be processed. Each of these is called a GPU shader. This matches AMD’s peak throughput data of 61.4 teraflops FP32 on the 7900 XTX. This is similar to what Nvidia did with Ampere (and now Ada), so know that official SP counts are not the same as potential GPU shader counts.
We also learned about the “AI Accelerator” that is part of the RDNA 3 architecture. A quick summary is to reuse SIMD32 units to do matrix operations instead of FP32 (or FP16). It also supports BF16 (16-bit brain floating point) format and INT8 along with FP16. These three (FP16/BF16/INT8) have the same peak throughput, which is double the FP32 single precision floating point throughput.
What is the difference between the previous half-precision FP16 shader support and the AI accelerator FP16 support? Basically, with some new instructions supported in matrix mode to optimize throughput and reduce power consumption will be Clearly, the peak rates of FP16/BF16 are significantly lower than what the RTX 4080 and 4090 can deliver. Finding software that specifically uses AI accelerators on AMD’s RDNA 2/RDNA 3 GPUs is also proving difficult at the moment, so we may have to revisit this at a later date. Hmm.
Of course, one of the highlights of RDNA 3 and Navi 31 is the move to chiplet architecture. Separating memory and cache from other GPU functions helps keep prices down. At the very least, we can reduce AMD’s costs. It may also open the door to future designs with higher performance. However, it’s important to note that at this point, there are all indications that using chiplets means some kind of performance compromise.
I can’t go any further here, but if you look at the specs on paper and see a 160% increase in theoretical compute and a 67% increase in memory bandwidth compared to the RX 6950 XT, it’s clear And it doesn’t really match AMD’s own benchmarks. 50-80% improvement in individual games (around 60% on average). I believe some of that is due to the additional overhead and latency associated with using chiplets. In other words, the RDNA 3 chiplet looks a bit like a Zen 2 chiplet rather than the Zen 3 or Zen 4 chiplets that AMD’s Ryzen CPUs are really ramping up.
There is another minor change from the previously published specs of the RX 7900 XT. When it was first announced, AMD listed his 300W TBP (Total Board Power) rating, but has since bumped it up to 315W. They found that performance improved significantly with only a relatively small increase in power usage.
Finally, let’s talk pricing and potential performance. I wasn’t particularly happy that Nvidia’s “step-down” RTX 4080 raised the price of his xx80 model GPU, but AMD’s RX 7900 XT is arguably just as bad. On paper, 17% less memory and memory bandwidth, and 16% less computational performance. That’s fine by itself, but the price savings are only 10%. In other words, most people eyeing the RX 7900 XT would rather spend the extra $100 on the XTX model as the RTX 4090 could actually be a better ‘deal’ from his Nvidia I think it’s good.
