The AMD Radeon RX 7800 XT has finally arrived, much later than we'd normally expect. Traditionally, when something new like the AMD RDNA 3 architecture comes out, we'd get the halo parts, then high-end, mainstream, and finally budget GPUs. But AMD skipped from the RX 7900 XTX / XT to the RX 7600, likely due to a surplus of RX 6000-series GPUs that were still in the channel due to an industry-wide oversupply. Now, nine months after Navi 31 first appeared, AMD is ready to release the mid-tier Navi 32 to compete against the best graphics cards.
Navi 32 will come in two variants on the desktop: the Radeon RX 7800 XT and the Radeon RX 7700 XT. We'll have both cards in the charts, but we've posted a separate review of the 7700 XT (so we can easily score them separately, if you're wondering). Will there be other variations of Navi 32 down the road? Certainly, we expect some mobile solutions, but given current market conditions, we probably won't get any non-XT parts.
Here's a rundown of the specifications, including the previous generation Navi 21/22 parts and the current Nvidia competitors.
Graphics Card | RX 7800 XT | RX 7700 XT | RX 6800 XT | RX 6700 XT | RTX 4070 | RTX 4060 Ti 8GB / 16GB |
---|---|---|---|---|---|---|
Architecture | Navi 32 | Navi 32 | Navi 21 | Navi 22 | AD104 | AD106 |
Process Technology | TSMC N5 + N6 | TSMC N5 + N6 | TSMC N7 | TSMC N7 | TSMC 4N | TSMC 4N |
Transistors (Billion) | 28.1 + 4x 2.05 | 28.1 + 3x 2.05 | 26.8 | 17.2 | 32 | 22.9 |
Die size (mm^2) | 200 + 150 | 200 + 113 (150) | 519 | 336 | 294.5 | 187.8 |
CUs / SMs | 60 | 54 | 72 | 40 | 46 | 34 |
GPU Cores (Shaders) | 3840 | 3456 | 4608 | 2560 | 5888 | 4352 |
AI / Tensor Cores | 120 | 108 | N/A | N/A | 184 | 136 |
Ray Tracing "Cores" | 60 | 54 | 72 | 40 | 46 | 34 |
Boost Clock (MHz) | 2430 | 2544 | 2250 | 2581 | 2475 | 2535 |
VRAM Speed (Gbps) | 19.5 | 18 | 16 | 16 | 21 | 18 |
VRAM (GB) | 16 | 12 | 16 | 12 | 12 | 8 / 16 |
VRAM Bus Width | 256 | 192 | 256 | 192 | 192 | 128 |
Infinity / L2 Cache | 64 | 48 | 128 | 96 | 36 | 32 |
ROPs | 96 | 96 | 128 | 64 | 64 | 48 |
TMUs | 240 | 216 | 288 | 160 | 184 | 136 |
TFLOPS FP32 (Boost) | 37.3 | 35.2 | 20.7 | 13.2 | 29.1 | 22.1 |
TFLOPS FP16 (FP8) | 74.6 | 70.4 | 41.4 | 26.4 | 233 (466) | 177 (353) |
Bandwidth / Effective (GBps) | 624 / 2708 | 432 / 1995 | 512 / 1664 | 384 / 1278 | 504 / ? | 288 / 554 |
TBP/TGP (Watts) | 263 | 245 | 300 | 230 | 200 | 160 |
Launch Date | Sep 2023 | Sep 2023 | Nov 2020 | Mar 2021 | Apr 2023 | May / July 2023 |
Launch Price | $499 | $449 | $649 | $479 | $599 | $399 / $499 |
Online Price | $500 | $450 | $500 | $320 | $590 | $374 |
It's a curious lineup in many ways, with the RX 7800 XT using the full Navi 32 GCD (Graphics Compute Die) plus four MCDs (Memory Cache Dies) and priced at $499, while the RX 7700 XT disables six CUs (Compute Units) and one MCD and comes priced at $449. Clock speeds for both the GPU and the GDDR6 also factor in, so let's put it another way.
In terms of paper specs, the RX 7800 XT has just 6% more GPU compute than the 7700 XT, but it has 44% more memory bandwidth and 33% more memory capacity. It's rated to consume 7% more power as well. For all that, you have to pay 11% more money. Even without analyzing benchmarks, then, it very much looks like the RX 7800 XT will be the card to get.
Perhaps that's part of why there's no AMD reference model for the RX 7700 XT. All of the 7700 XT cards will come from AMD AIB (add-in board) partners, while the RX 7800 XT will be available both in a reference design and via custom AIB cards.
One thing to note on the specs sheets is the memory bandwidth and "effective bandwidth" figures. Nvidia and AMD don't appear to calculate effective bandwidth in the same fashion. Actually, the RX 7600 seems to have similar effective bandwidth compared to the 4060 and 4060 Ti, but the RX 7700 XT and above appear to be reporting the peak Infinity Fabric bandwidth. Don't read too much into those "effective" numbers, in other words, as ultimately, real-world performance will be what matters most.
Generationally, we've already been musing about exactly where the RX 7800 XT will land. It boasts 80% more theoretical compute than the previous generation RX 6800 XT, which looks great on paper... but then the RX 7900 XTX boasts 159% more compute than the RX 6950 XT on paper, while in practice, it's only about 30–40 percent faster. Do the math, and the RX 7800 XT looks as though it may not be much faster than the previous generation RX 6800 XT — a problem we've already noted with the RX 7600, which offers just slightly better performance than the RX 6650 XT for slightly more money.
It's not just about specs, though. RDNA 3 includes AV1 encoding and decoding support, DisplayPort 2.1 (UHBR13.5) outputs, improved AI processing hardware, and generally superior efficiency compared to the previous RDNA 2 architecture. Also, the RX 7800 XT comes priced $150 lower than the launch price of the RX 6800 XT — though current street prices are the same now, with cards like this PowerColor RX 6800 XT going for the same $499.
Looking at the Nvidia side of the equation, there are a few different competitors. One is the generally good RTX 4070, priced at $599, while the other is the lackluster and niche RTX 4060 Ti 16GB, priced at $499 — but now starting at $449 — or the lower priced RTX 4060 Ti 8GB that costs $399. AMD offers the same 16GB of VRAM as Nvidia's "upgraded" 4060 Ti, at the same MSRP, but AMD also has a 256-bit memory interface and 64MB of L3 cache, compared to the 4060 Ti's 128-bit interface and 32MB of L2 cache. Advantage: AMD.
But the RTX 4070 provides stronger competition, even if it costs more. You get 12GB of memory on a 192-bit interface, and a lot more GPU horsepower than the 4060 Ti. In our testing, the RTX 4070 beats the RTX 4060 Ti by 30–40 percent in gaming performance. We definitely want to see how it stacks up against the new RX 7800 XT in real-world performance, which is what we'll be looking at shortly.
From the raw specs, the RX 7800 XT has more theoretical FP32 compute (32-bit floating point is what most games use). It also has 33% more VRAM, and 24% more raw memory bandwidth. Nvidia's RTX 4070 meanwhile blows AMD out of the water in theoretical FP16 compute (16-bit floating-point gets used a lot in AI workloads), with over 3X the computational power thanks to its tensor cores.
Nvidia also offers DLSS 2 upscaling, which is currently more widely supported than FSR 2 upscaling, plus DLSS 3 Frame Generation is already in about 50 games while we're still waiting for the public release of FSR 3. These are factors that need to be considered, though we still put native rendering performance as the primary metric.
There are a lot of aspects to compare with modern graphics cards, in other words, and you can't just look at the specs. It's why we run the benchmarks.
- MORE: Best Graphics Cards
- MORE: GPU Benchmarks and Hierarchy
- MORE: All Graphics Content