The AMD Radeon RX 7900 GRE isn't a new graphics card — the 7900 GRE officially launched in July 2023 ... in China. Now it's coming to the U.S. and other regions, as the third variant of AMD's largest Navi 31 GPU using the RDNA 3 architecture , with a starting price of $549. With Nvidia's recently launched RTX 4070 Super and the price cut on the vanilla RTX 4070 , it seems AMD felt a need to bump performance in this market segment, and the RX 7900 GRE hopes to claw out some space for itself among the best graphics cards .
We thought AMD was basically done with new RDNA 3 cards when the RX 7600 XT arrived last month, but bringing the 7900 GRE to the U.S. and other areas of the world does make sense. It also makes the RX 7800 XT look just a bit less desirable. The 7900 GRE is similar in some ways to the 7800 XT, as both have four MCDs (memory cache dies) and 16GB of total VRAM, and both are priced in the $500~$600 range. But the 7800 XT only has 60 compute units (CUs) while the 7900 GRE has 80 CUs — a 33% improvement on paper.
Here's the rundown of specifications for the relevant current generation AMD and Nvidia GPUs. You can't just compare raw specs of course, but we'll have plenty of benchmarks to show how AMD's 'new' card stacks up to the competition.
| Graphics Card | RX 7900 GRE | RX 7900 XT | RX 7800 XT | RX 7900 XTX | RX 7700 XT | RTX 4070 | RTX 4070 Super | RTX 4060 Ti 16GB | RTX 4070 Ti Super | RTX 4080 Super |
|---|---|---|---|---|---|---|---|---|---|---|
| Architecture | Navi 31 | Navi 31 | Navi 32 | Navi 31 | Navi 32 | AD104 | AD104 | AD106 | AD103 | AD103 |
| Process Technology | TSMC N5 + N6 | TSMC N5 + N6 | TSMC N5 + N6 | TSMC N5 + N6 | TSMC N5 + N6 | TSMC 4N | TSMC 4N | TSMC 4N | TSMC 4N | TSMC 4N |
| Transistors (Billion) | 45.6 + 4x 2.05 | 45.6 + 5x 2.05 | 28.1 + 4x 2.05 | 45.6 + 6x 2.05 | 28.1 + 3x 2.05 | 32 | 32 | 22.9 | 45.9 | 45.9 |
| Die size (mm^2) | 300 + 225 | 300 + 225 | 200 + 150 | 300 + 225 | 200 + 113 | 294.5 | 294.5 | 187.8 | 378.6 | 378.6 |
| SMs / CUs / Xe-Cores | 80 | 84 | 60 | 96 | 54 | 46 | 56 | 34 | 66 | 80 |
| GPU Cores (Shaders) | 5120 | 5376 | 3840 | 6144 | 3456 | 5888 | 7168 | 4352 | 8448 | 10240 |
| Tensor / AI Cores | 160 | 168 | 120 | 192 | 108 | 184 | 224 | 136 | 264 | 320 |
| Ray Tracing Cores | 80 | 84 | 60 | 96 | 54 | 46 | 56 | 34 | 66 | 80 |
| Boost Clock (MHz) | 2245 | 2400 | 2430 | 2500 | 2544 | 2475 | 2475 | 2535 | 2610 | 2550 |
| VRAM Speed (Gbps) | 18 | 20 | 19.5 | 20 | 18 | 21 | 21 | 18 | 21 | 23 |
| VRAM (GB) | 16 | 20 | 16 | 24 | 12 | 12 | 12 | 16 | 16 | 16 |
| VRAM Bus Width | 256 | 320 | 256 | 384 | 192 | 192 | 192 | 128 | 256 | 256 |
| L2 / Infinity Cache | 64 | 80 | 64 | 96 | 48 | 36 | 48 | 32 | 64 | 64 |
| Render Output Units | 192 | 192 | 96 | 192 | 96 | 64 | 80 | 48 | 96 | 112 |
| Texture Mapping Units | 320 | 336 | 240 | 384 | 216 | 184 | 224 | 136 | 264 | 320 |
| TFLOPS FP32 (Boost) | 46.0 | 51.6 | 37.3 | 61.4 | 35.2 | 29.1 | 35.5 | 22.1 | 44.1 | 52.2 |
| TFLOPS FP16 (FP8) | 92 | 103.2 | 74.6 | 122.8 | 70.4 | 233 (466) | 284 (568) | 177 (353) | 353 (706) | 418 (836) |
| Bandwidth (GBps) | 576 | 800 | 624 | 960 | 432 | 504 | 504 | 288 | 672 | 736 |
| TDP (watts) | 260 | 315 | 263 | 355 | 245 | 200 | 220 | 160 | 285 | 320 |
| Launch Date | Jul 2023 | Dec 2022 | Sep 2023 | Dec 2022 | Sep 2023 | Apr 2023 | Jan 2024 | Jul 2023 | Jan 2024 | Jan 2024 |
| Launch Price | $549 | $899 | $499 | $999 | $449 | $599 | $599 | $499 | $799 | $999 |
| Online Price | $550 | $700 | $490 | $920 | $422 | $530 | $600 | $430 | $800 | $1,000 |
The RX 7900 GRE started trickling out to other markets around the world over the past month, after initially becoming available in China with a converted price of around $650. Today (February 26, 2024) however marks the official unveiling of the worldwide availability and pricing, with a $549 MSRP for the U.S. market. Performance hasn't changed, but we haven't had a card for review until now, so that's certainly something we want to investigate.
One of the interesting questions is why AMD decided not to launch the 7900 GRE outside of China initially. We asked but didn't receive an answer yet. The most likely cause is that there was still a glut of previous generation high-end RX 6000-series GPUs available, including cards like the RX 6950 XT, 6900 XT, and 6800 XT. Now that inventories of those have mostly cleared out, there's more room for a new card between the 7800 XT and 7900 XT. But still, why did China get such a card first? Perhaps as a test of the waters. But whatever the case, it's now launched everywhere — or at least, it will be officially available in the U.S. starting tomorrow, February 27, 2024.
The 7900 GRE uses AMD's largest Navi 31 GCD (graphics compute die), which gives it access to far more compute resources than the step down Navi 32 GCD. The 7900 XTX has the full chip with 96 CUs, while the 7900 XT has 84 CUs and only five of the six potential MCDs enabled. The 7900 GRE only drops the CU count to 80, so perhaps it's a case of AMD having enough Navi 31 GCDs with two defective MCD links.
AMD also reduced the boost and game clocks quite a bit, with a 2245 MHz boost clock (versus 2400 MHz on the 7900 XT and 2430 MHz on the 7800 XT). That in turn makes for a more efficient chip, and the 7900 GRE on paper comes in just 3W below the 7800 XT's 263W TBP (Total Board Power) rating. That's still 60W than Nvidia's similarly priced RTX 4070, and Nvidia's Ada GPUs will continue to reign as the most efficient GPUs right now.
Memory clocks are also slightly lower than on the other high-end AMD chips, which run at 20 or 19.5 Gbps. The 7900 GRE runs its GDDR6 at 18 Gbps. The 64MB L3 Infinity Cache means effective bandwidth should still be very good, but the reduced bandwidth likely creates a slightly wider gap between the 7900 GRE and the next step up 7900 XT.
Theoretical compute ends up at 46.0 teraflops FP32, 23% higher than the 7800 XT and 11% lower than the 7900 XT. Raw memory bandwidth meanwhile is 8% lower than the 7800 XT and 28% lower than the 7900 XT. Like I said, it seems AMD was aiming to create a GPU that would land midway between the other two cards, which is why memory speeds are reduced compared to the other two options.
- MORE: Best Graphics Cards
- MORE: GPU Benchmarks and Hierarchy
- MORE: All Graphics Content
