AMD Radeon RX 7700 XT Review: Stuck in the Middle

The AMD AMD Radeon RX 7700 XT and Radeon RX 7800 XT cards officially launched today, priced at $449 and $499, respectively. It’s the usual story of a trimmed-down variant of the same core GPU — Navi 32 in this case — coming in at a lower price point. Unfortunately, some of the trimmings went too far, while the price remains too high for what you get. As the final chip in the AMD RDNA 3 architecture lineup, Navi 32 fills the middle of the product stack and looks to compete with the best graphics cards.

But if you’re looking at Navi 32, you’ll generally be far better served by the Radeon RX 7800 XT, at least at current prices. We have a separate review of that card, and we suggest you start there as we’ll cover more of the Navi 32 details there.

The short summary: the 7700 XT comes up short and costs too much. Here’s a rundown of the specifications, including the previous generation Navi 21/22 parts and the current Nvidia competitors. 

Swipe to scroll horizontally
AMD RX 7800/7700 XT and Competing GPUs Specifications
Graphics Card RX 7700 XT RX 7800 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 + 3x 2.05 28.1 + 4x 2.05 26.8 17.2 32 22.9
Die size (mm^2) 200 + 113 (150) 200 + 150 519 336 294.5 187.8
CUs / SMs 54 60 72 40 46 34
GPU Cores (Shaders) 3456 3840 4608 2560 5888 4352
AI / Tensor Cores 108 120 N/A N/A 184 136
Ray Tracing “Cores” 54 60 72 40 46 34
Boost Clock (MHz) 2544 2430 2250 2581 2475 2535
VRAM Speed (Gbps) 18 19.5 16 16 21 18
VRAM (GB) 12 16 16 12 12 8 / 16
VRAM Bus Width 192 256 256 192 192 128
Infinity / L2 Cache 48 64 128 96 36 32
ROPs 96 96 128 64 64 48
TMUs 216 240 288 160 184 136
TFLOPS FP32 (Boost) 35.2 37.3 20.7 13.2 29.1 22.1
TFLOPS FP16 (FP8) 70.4 74.6 41.4 26.4 233 (466) 177 (353)
Bandwidth / Effective (GBps) 432 / 1995 624 / 2708 512 / 1664 384 / 1278 504 / ? 288 / 554
TBP/TGP (Watts) 245 263 300 230 200 160
Launch Date Sep 2023 Sep 2023 Nov 2020 Mar 2021 Apr 2023 May / July 2023
Launch Price $449 $499 $649 $479 $599 $399 / $499
Online Price $450 $500 $500 $320 $590 $374

The RX 7700 XT looks quite good compared to the previous-gen RX 6700 XT. You get substantially more compute, which is the key selling point. Theoretically, the RX 7700 XT offers 167% more number-crunching prowess than the 6700 XT. Even though we’ve pointed out in other RDNA 3 reviews that performance doesn’t match that lofty figure, it’s a good step up from the 6700 XT.

The problem is that it’s also a big step up in price relative to the existing cards, where online pricing has dropped to $329 for the RX 6700 XT and $349 for the RX 6750 XT, and even the RX 6800 costs less than the 7700 XT at a current price of $429. The RX 6800 also has more memory and a wider interface, and as we’ll see later, it provides stiff competition for the newcomer.

The other problem is the RX 7800 XT that uses the full Navi 32 implementation. On paper, the RX 7800 XT offers 6% more GPU compute, but 44% more memory bandwidth and 33% more memory capacity. And you get all that for an 11% increase in price. Spoiler alert: Generally, the RX 7800 XT is up to 20% faster. Short of a price cut, it’s the better buy. Maybe that’s why AMD is leaving all the RX 7700 XT production to its AIB (add-in board) partners.

Nvidia GeForce RTX 4060 Ti Founders Edition photos and unboxing

(Image credit: Tom’s Hardware)

If you don’t want to step up to the $500 RX 7800 XT, then you have to look at other cards like Nvidia’s competing RTX 4060 Ti, which costs $50 less — or the same price if you get the 16GB variant. You can go lower than that as well, with the RX 6700 XT and RTX 4060. AMD’s 6700 XT and 7700 XT at least provide 12GB of VRAM with a 192-bit bus, so they should be good for most games for quite some time, but if you’re not dead set on maxing out all the settings in games, you can still get by with an 8GB card.

Of course, you also get the new features of RDNA 3, like AV1 encoding and decoding support, DisplayPort 2.1 (UHBR13.5) outputs, and improved AI processing hardware. Those are fine to have, though, for most people, they’re not going to matter enough to sway your purchasing decision. If you care about AV1 encoding, Nvidia does it better on RTX 40-series parts; if you care about DP2.1 outputs, then you’ll also need a non-existent DP2.1 monitor… or else just stick to 4K and 240 Hz or lower displays, which can work on all the previous generation parts as well.

There’s also the broader ecosystem to consider. Nvidia has been pushing ray tracing for five years now, and there are certainly games that make decent use of the new API. Alongside ray tracing, you have AI and deep learning, as well as DLSS — which now has upscaling, frame generation, and soon ray reconstruction. 

XFX Radeon RX 7700 XT QICK 319

As noted already, there’s no AMD reference RX 7700 XT. We received the XFX RX 7700 XT QICK 319 for this review, which should be fairly representative of what we’ll see from baseline 7700 XT models.

One thing that immediately stands out is the size of the XFX card, especially as we’ve been testing the reference RX 7800 XT at the same time. It’s very large for a “mainstream” card, probably larger than it needs to be. But if you’re not using a compact case, what else are you going to do with all that space?

XFX goes the standard route with packaging as well, which means lots of empty space in the box. That should keep the card safe during shipping, but we definitely think the toned-down AMD and Nvidia packaging on reference models should be more commonly used by the AIB partners.

The XFX RX 7700 XT measures 339x129x50 mm, almost into RTX 4080/4090 as far as dimensions go. Even so, the card isn’t particularly heavy, weighing just 1111g — 4g more than the 7800 XT reference card, if you’re keeping score. Basically, XFX is using a lot of volume and triple fans to provide more than sufficient cooling, which means the card should generally run cool and quiet — and it does.

XFX has three fans, with two 94mm fans on the outside, and an 84mm fan in the center. The fans don’t have integrated rims, which tends to be a higher quality design, but they’re large enough and the card’s 245W TBP (total board power) is low enough that it shouldn’t matter too much.

Like the RX 7800 XT, the 7700 XT comes with dual 8-pin power connectors. That’s not strictly necessary as an 8-pin plus 6-pin combination, combined with the PCIe slot, should be able to provide up to 300W of power. Still, 8-pin connectors are pretty ubiquitous now, and it’s safer to not inadvertently end up with a card that pulls too much power from the x16 slot.

XFX AMD Radeon RX 7700 XT card photos

(Image credit: Tom’s Hardware)

We mentioned the relative size of the card, and this shot hopefully puts things into better perspective. The 7800 XT reference card looks almost tiny by comparison, though we anticipate there will be plenty of compact 7700 XT designs from other vendors.

There’s no RGB lighting or other bling on the XFX cards, so this is as barebones as it gets. The only real extra is a switch near the IO bracket that selects quiet or performance mode. In terms of IO, you get the standard single HDMI 2.1 port and triple DisplayPort outputs.

XFX RX 7700 XT Overclocking 

XFX AMD Radeon RX 7700 XT card photos

(Image credit: Tom’s Hardware)

This is going to be a short section, as our overclocking attempts with the XFX card basically ended in frustration. We normally start by maxing out the power limit and then attempt to dial in stable core and memory overclocks. We repeatedly ended up with a green screen and a crashed GPU after even minor changes.

It goes further than that. AMD’s Radeon Settings feature auto-overclocking options for tuning the memory and GPU clocks, or you can go full manual. These are usually pretty conservative, but even clicking the auto-overclock RAM and GPU options resulted in problems.

Given we couldn’t even dial in a 50 MHz core or GDDR6 overclock without issues, we decided to call it quits and move on. Whether it’s our particular sample card, or something with RX 7700 XT in general, we can’t say. But overclocking of RDNA 3 cards in general hasn’t been all that great in our experience.

AMD Radeon RX 7700 XT Test Setup

(Image credit: Tom’s Hardware)

We updated our GPU test PC at the end of last year with a Core i9-13900K, though we continue to also test reference GPUs on our 2022 system that includes a Core i9-12900K for our GPU benchmarks hierarchy. (We’ll be updating that with the 7800/7700 XT numbers shortly.) For this review, we use the 13900K results for gaming tests, which ensures, as much as possible, that we’re not CPU limited. We also use the 2022 PC for AI tests and professional workloads.

GPU Test Equipment

We’re using AMD’s preview drivers (basically 23.8.2 but with support for the Navi 32 GPUs) for this review. Most of the results from other cards are from the past few months, and we’ve retested a few specific games where there were apparent anomalies in performance. We also retested Minecraft on every GPU over the past few weeks, as we found forcing vsync off in the game’s Options.txt file improved performance by 10% or more on most GPUs (with Intel Arc being the biggest beneficiary).

We’re including current generation and previous generation AMD hardware in our charts, with GPUs nominally priced between about $600 and $300. The same price bracket applies to the Nvidia and Intel GPUs, though we’re not including RTX 30-series parts here. (If you want to see RTX 30-series results, check our GPU benchmarks or the RTX 4060 Ti 16GB review.)

Our current test suite consists of 15 games. Of these, nine support DirectX Raytracing (DXR), but we only enable the DXR features in six games. The remaining nine games are tested in pure rasterization mode. While many of the games in our test suite support upscaling — twelve support DLSS 2, five support DLSS 3, and five support FSR 2 — due to time constraints we’ve skipped upscaling tests here. (Keep in mind that we have both the 7700 XT and 7800 XT launching today, so there were more tests to run over the past week or so than usual.)

As with the RTX 4060 Ti 16GB review, we’re also including a few additional games with limited testing: F1 2023, Hogwarts Legacy, and The Last of Us: Part 1. Just for kicks, we’re also showing Starfield performance, even though Nvidia performance appears seriously poor right now. These extra games are not factored into the overall performance ranking but are more for those who want to see some newer titles.

We tested the RX 7700 XT at 1080p (medium and ultra), 1440p ultra, and 4K ultra — ultra being the highest supported preset if there is one, and in some cases maxing out all the other settings for good measure (except for MSAA or super sampling). We’ll only have limited commentary on the 4K results, as the RX 7700 XT doesn’t do too well at that resolution. It primarily targets 1440p gaming, according to AMD, or even 1080p.

Our PC is hooked up to a Samsung Odyssey Neo G8 32, one of the best gaming monitors around, allowing us to fully experience some of the higher frame rates that might be available. G-Sync and FreeSync were enabled, as appropriate. As you can imagine, getting anywhere close to the 240 Hz limit of the monitor proved difficult, as we don’t have any esports games in our test suite.

We installed all the then-latest Windows 11 updates when we assembled the new test PC. We’re running Windows 11 22H2, but we’ve used InControl to lock our test PC to that major release for the foreseeable future (though critical security updates still get installed monthly).

Our new test PC includes Nvidia’s PCAT v2 (Power Capture and Analysis Tool) hardware, which means we can grab real power use, GPU clocks, and more during all of our gaming benchmarks. We’ll cover those results on our page on power use.

Finally, because GPUs aren’t purely for gaming these days, we’ve run some professional application tests, and we also ran some Stable Diffusion benchmarks to see how AI workloads scale on the various GPUs.

AMD says the RX 7700 XT targets 1440p gaming, so that’s where we’ll start. It does have 12GB of memory and a 192-bit interface, but it’s also going to be less capable than the 7800 XT, so there will be games where 1440p and maxed-out settings prove a bit too much, particularly in ray tracing games.

Our current test regimen gives us a global view of performance using the geometric mean of all 15 games in our test suite, including both the ray tracing and rasterization test suites. Then we’ve got separate charts for the rasterization and ray tracing suites, plus charts for the individual games. If you don’t like the “overall performance” chart, the other two are the same view we’ve previously presented.

Our test suite is intentionally heavier on ray tracing games than what you might normally encounter. That’s largely because ray tracing games tend to be the most demanding options, so if a new card can handle ray tracing reasonably well, it should do just fine with less demanding games. Ray tracing also feels increasingly like something we can expect to run well when optimized properly. Mainstream and certainly high-end graphics cards need to be capable of running demanding settings, including ray tracing, at reasonable frame rates.

AMD Radeon RX 7700 XT performance charts

(Image credit: Tom’s Hardware)

The overall performance at 1440p ultra sets the stage for what we’ll see repeated again and again. It’s like a modern dance recital for GPUs! The RX 7700 XT comes in a bit ahead of the existing RX 6800 but behind the RX 6800 XT. It’s a relatively large generational improvement over the RX 6700 XT — 35% faster overall — but then it also costs 36% more money based on current prices.

Like the other RDNA 3 GPUs, we’re mostly looking at a sideways and slightly upward movement in gaming performance, with a slight price increase as well. The problem is that almost every new GPU from AMD and Nvidia for this generation feels like the model name got bumped one notch higher than it should have been.

If we take the RX 6900 XT as an example, AMD improved performance by 40–50 percent (at 1440p and 4K) with the 7900 XTX. The 7700 XT delivers about 55% more performance than the RX 6700 10GB, so maybe AMD should have dropped the XT suffix and knocked the price down below $400. Then we’d have something to get excited about.

AMD does beat the RTX 4060 Ti, but again it’s at a price premium (unless you count the 16GB card). Paying 13% more for 8% more performance, while not factoring in DLSS and AI or the higher power requirements, just doesn’t make for a compelling argument in favor of the RX 7700 XT. It’s not bad, but neither is it great.

Toss out the ray tracing games and look just at rasterization performance and things do tilt quite a bit in AMD’s favor. The RX 7700 XT beats the 4060 Ti by 20%, and it’s less likely to hit VRAM limits thanks to having 12GB. AMD’s newcomer sweeps the charts, leading by anywhere from 4% in Total War: Warhammer 3 to as much as 38% in Borderlands 3.

By the same token, the 7700 XT only leads the RX 6800 by 3% (a range of -5% to + 13%), and the two are effectively tied. Plus you get the extra VRAM and bandwidth on the 6800. Claiming a victory over Nvidia but tying your existing part just doesn’t give anyone a lot of reason to get excited.

More critically, the RX 7800 XT offers 17% more performance for 11% more money. Unless the price changes for one or both Navi 32 GPUs, that’s the better pick.

Ray tracing skews performance the other way, and now the RTX 4060 Ti offers 9% more performance overall. It’s not a clean sweep, though, as AMD still offers slightly higher performance (without DLSS or FSR 2) in Spider-Man: Miles Morales and Metro Exodus Enhanced. Minecraft as usual heavily favors Nvidia, to the tune of 57%, and we’d probably see a similarly wide margin in Cyberpunk with RT Overdrive mode.

Plenty of gamers will argue that ray tracing so far hasn’t proved to be a killer feature. They’re not wrong. But when was the last time we actually saw any new graphics feature that radically improved image fidelity? It’s been incremental improvements with often disproportionately large performance hits for at least the past decade in my book.

And while I’m here banging on my drum, let’s note again that the 7800 XT offers 16% higher performance in ray tracing games at 1440p, for 11% more money. It used to be that graphics cards had diminishing returns — the GTX 980 cost 66% more than the GTX 970 back in the day, while offering an average performance improvement of only 15–20 percent. The current situation of paying more to get proportionately more performance is definitely not typical in the GPU realm.

Stepping down to 1080p ultra, the margins start to shrink when comparing the RX 7700 XT and RTX 4060 Ti. While it’s possible CPU and other system bottlenecks are becoming a factor, these GPUs are also far enough down the hierarchy that we should still be mostly limited by the graphics processing power.

AMD Radeon RX 7700 XT performance charts

(Image credit: Tom’s Hardware)

The RX 7700 XT now leads the RTX 4060 Ti by just 3% overall. There are certainly still games that favor AMD by 20% or more, but there are also games where there’s no real difference, and ray tracing games as usual tend to favor Nvidia. 1080p also doesn’t hit Nvidia’s 128-bit interface as hard, which likely helps to narrow the gap.

The rasterization rankings are pretty must the same as at 1440p, just with slightly less of a gap. RX 7700 XT beats the RTX 4060 Ti by 13%, and squeaks past the previous gen RX 6800 by 2% again. Every game also easily clears the 60 fps mark, with only Flight Simulator and Red Dead Redemption 2 failing to reach the triple digits.

AMD would probably like more comparisons with the equivalently (now) priced 4060 Ti 16GB, but even after the $50 price cut, that’s not really a great option. The added VRAM mostly helps at 4K ultra, at which point the GPU is already struggling quite a bit in most modern games. Which means the real competition is from the 4060 Ti 8GB that costs $50 less, and overall things still end up mostly as a wash.

Nvidia continues to lead in ray tracing performance, by 11% overall now. That’s only a 2% delta compared to 1440p, indicating that neither of these GPUs are running into VRAM limits so far. Minecraft continues to be the outlier, thanks to its full ray tracing graphics, but Cyberpunk also shows a large 22% lead for the 4060 Ti compared to the RX 7700 XT.

Otherwise, everything pretty much remains as before. The 7700 XT comes out 11% ahead of the RX 6800, so the improved ray tracing hardware in RDNA 3 does help, but it’s not a massive improvement.

We do have to wonder what performance might have looked like had AMD used a monolithic die with no MCDs but otherwise identical specs, because the external cache and memory controller chips have to introduce at least some additional latency. RDNA 3 is basically a proof of concept for GPU chiplets; RDNA 4 is where we’ll hopefully see more impressive gains from the chiplet approach.

The 1080p medium results, like the 4K ultra results on the next page, are more academic than meaningful. Some people might prefer the higher framerates over image fidelity, but in pursuit of those framerates, CPU bottlenecks become much more of a factor. We’re only providing limited commentary here.

AMD Radeon RX 7700 XT performance charts

(Image credit: Tom’s Hardware)

The overall margin between the 7700 XT and 4060 Ti shrinks to 2% at 1080p medium in favor of AMD. Ray tracing games still favor Nvidia, while rasterization games tend to favor AMD’s chip. Across our full test suite, the difference ranges from -34% (Minecraft) to +26% (A Plague Tale: Requiem). About half of the games show less than a 10% gap.

All of the rasterization games now easily clear 60 fps and even 100 fps, with with four games even pushing past 200 fps. The ray tracing games are a different matter, with three games — Bright Memory Infinite benchmark, Cyberpunk 2077, and Minecraft — still coming up shy of the 60 fps mark. Turning off ray tracing can of course get any of those games well above 60 fps at 1080p.

You can peruse the charts below, and if you see anything particularly interesting, let us know in the comments. Otherwise, let’s head to the opposite end of the spectrum and punish the GPUs with 4K ultra settings.

4K ultra is very much not the intended playground for the RX 7700 XT, at least not if you’re playing recent and relatively demanding games. If you’re looking to play through Half-Life again, though, that’s a different matter. Again, we’re only going to provide very limited commentary here, so enjoy the charts at your leisure.

AMD Radeon RX 7700 XT performance charts

(Image credit: Tom’s Hardware)

The RX 7700 XT averages just 32 fps across our 15 game suite at 4K ultra. As you can imagine, that means some games (hello, Cyberpunk) are almost into the single digit framerates. The extra VRAM compared to the RTX 4060 Ti does widen the overall gap to 14%, but all six ray tracing games fail to hit 30 fps. The rasterization games on the other hand all manage at least 30 fps, though Total War: Warhammer 3 and A Plague Tale: Requiem just barely hit that mark.

Of course there are other ways to make “4K” gaming accessible, like FSR 2 and DLSS upscaling. If you run FSR 2 in Performance mode (4x upscaling), you’re basically taking 1080p and upscaling it to 4K. The performance for FSR 2 upscaling in that case is generally around 20%, give or take, which means plenty of games can still clear 60 fps. But upscaling isn’t quite the same as native 4K, I’m sure most people will agree.

Anyway, enjoy the charts below.

In our RTX 4060 Ti 16GB review, we added three additional (relatively) recent game releases to our testing, to see if they showed any additional benefit from the added memory. We’re keeping those same games for the 7800/7700 XT reviews, and adding one more: Starfield — not because it’s a balanced and well-optimized game, but it does go into full release today and we have some numbers from other GPUs, which you can see in our Starfield GPU performance guide, so we figured why not?

The other three games consist of F1 2023, Hogwarts Legacy, and The Last of Us: Part 1. None of these are brand-spanking new, and after the initial launch woes, most of them have received some much-needed TLC. But Hogwarts and The Last of Us still have a reputation for wanting more than 8GB of VRAM at maxed-out settings, and Starfield will give us a look at a brand-new game on current and previous generation AMD GPUs as well as Nvidia GPUs — caveats apply to Nvidia performance right now.

We’ll do a gallery of the benchmarks from each game, so swipe through the images to see the results at the various settings/resolutions.

F1 2023 is a bit interesting in that the presets disable ray tracing at anything besides the ultra preset, so we get results that mirror what we saw in our rasterization and ray tracing suites, depending on the settings used. For reference, we tested on the Australia track, in the rain, for one loop of the track (130 seconds).

At 1080p medium, the RX 7700 XT beats the RTX 4060 Ti by 20%. That shrinks to just 3% at 1080p and 1% at 1440p, basically tied, thanks to the ray tracing effects. But then at 4K ultra, the AMD lead grows to 12% as the 8GB card starts to run out of memory. That doesn’t happen with the 4060 Ti 16GB, though  with both cards sitting at 29 fps it’s a bit of a pyrrhic victory for the 16GB Nvidia card.

Hogwarts Legacy doesn’t like running at maxed out ultra settings on cards with only 8GB of VRAM, and even 12GB can be a problem at 1440p ultra with maxed out ray tracing settings. The game will warn you about selecting settings that are too high, and then it will often crash to desktop. We skipped the 4K testing due to the instability, not to mention performance would have been very poor on all of the GPUs. But we did manage to complete at least one benchmark run where needed.

For our 1080p medium settings, we have RT reflections, RT shadows, and RT ambient occlusion enabled, plus the overall RT quality set to medium. Our ultra settings use the ultra preset with all the RT effects enabled and RT quality set to ultra. As you’d expect, all of those ray tracing effects give the RTX 4060 Ti the lead over the RX 7700 XT.

Nvidia’s RTX 4060 Ti 8GB card is 12% faster at 1080p medium, 7% faster at 1080p ultra, and basically tied (1% faster) at 1440p ultra. The 4060 Ti 16GB doesn’t fall off a cliff at 1440p, and maintains an 8% lead over the 7700 XT. Of course, both GPUs would benefit from enabling some level of upscaling, as 1440p otherwise dips below 30 fps at times.

The Last of Us: Part I favors AMD’s GPUs, and it’s an AMD-promoted game. Using the medium and ultra presets, the RX 7700 XT is 14–17 percent faster than the RTX 4060 Ti 8GB at 1080p and 1440p, and then the lead grows to 36% at 4K when the 8GB card runs out of VRAM.

The RTX 4060 Ti 16GB doesn’t have that problem, naturally, and so the 7700 XT lead falls in the 13–15 percent range at all tested resolutions and settings.

Last is Starfield, a game that officially launches today (it’s been in early access for purchasers of the premium edition for the past week). As we noted in our Starfield PC performance testing, there are clearly issues with non-AMD GPUs right now. Nvidia’s GPUs aren’t hitting their power limit and fall further behind than in other games.

The net results is that the RX 7700 XT beats the RTX 4060 Ti (8GB) by 22–34 percent. VRAM capacity doesn’t factor in, as the 4060 Ti 16GB shows a similar 25–35 percent lead for AMD’s 7700 XT.

We haven’t mentioned the 7800 XT of late, so let’s also point out that it’s significantly faster than the 7700 XT — by around 30% across all tested resolutions and settings in Starfield. Whatever the game engine is doing, it definitely seems to appreciate raw memory bandwidth.

Radeon RX 7700 XT AI Performance

GPUs are also used with professional applications, AI training and inferencing, and more. Along with our usual professional tests, we’ve added Stable Diffusion benchmarks on the various GPUs. AI is a fast-moving sector, and it seems like 95% or more of the publicly available projects are designed for Nvidia GPUs. Those Tensor cores aren’t just for DLSS, in other words, but AMD now has “AI Accelerators” in RDNA 3 to help boost FP16 performance for such workloads. Let’s start with our AI testing and then hit the professional apps.

We’re using Automatic1111’s Stable Diffusion version for the Nvidia cards, while for AMD we’re using a recent Nod.ai Shark variant — we’ve retested all the AMD cards for this review using build version 20230713_819, as results have improved substantially compared to our previous testing. Intel meanwhile has finally has an OpenVINO fork of Automatic1111’s webui, which shows about a 50% improvement over our previous testing. We’ve retested those cards as well, now with Stable Diffusion 2.1 and 768×768 results.

AMD’s GPUs continue to look relatively weak in Stable Diffusion, even with the latest Nod.ai software. We’ve retested all of the AMD GPUs, and performance of the RX 7700 XT can’t even match the RTX 3060, though it does at least manage to beat the RTX 3050. Also, the RX 7700 XT still beats every previous generation RX 6000-series GPU.

It’s not clear how much of that is due to custom tuning on the part of Nod.ai, but we did manage to get Automatic1111 running via DirectML. It was universally slower than the Nod.ai version, so we decided to stick with Nod.ai for now. Note also how much Intel Arc has improved over time, with performance that now rivals the RTX 4060 Ti — it’s slightly ahead for 512×512 image generation, but only matches the 4060 in 768×768 images.

Radeon RX 7700 XT Professional Workloads

SPECviewperf is a different story, as AMD has been more willing to release drivers that help boost professional application performance even on consumer hardware. There are eight different benchmarks, and we use the geometric mean from those to generate an aggregate score — that’s not an official score, but it gives an easy way of at least approximating overall performance. Few professionals use all these programs, however, so it’s generally more important to look at the results for the application(s) you plan to use.

AMD’s RX 7700 XT sweeps the RTX 4060 Ti in SPECviewperf, leading by anywhere from  20% (Maya-06) to as much as 1377% (SNX-04). Nvidia prefers to keep its best optimizations for some of these professional apps (SNX in particular) locked behind its professional drivers. Even the old RX 6700 XT typically outperforms Nvidia’s RTX 4060 Ti.

For GPU accelerated 3D rendering, the only app that currently supports all three GPU vendors is Blender. It now leverages the ray tracing hardware to boost performance, and that generally puts Nvidia into the pole position.

The RTX 4060 Ti easily beats AMD’s RX 7700 XT, more than doubling its performance. Intel’s Arc A770 16GB also edges past the RX 7700 XT. At the same time, the RX 7700 does surpass the entire previous generation family, including the RX 6950 XT, so there are clearly some benefits from the new architecture.

Radeon RX 7700 XT Content Creation Summary

The RX 7700 XT follows typical trends in content creation benchmarks for AMD GPUs: It does quite well in SPECviewperf, but struggles in the AI and Blender tests. Depending on your needs, it may or may not suffice as a professional GPU alternative.

(Image credit: Tom’s Hardware)

Our new test PC uses an Nvidia PCAT v2 device, and we’ve switched from the Powenetics hardware and software we’ve previously used to PCAT, as it gives us far more data without the need to run separate tests. PCAT with FrameView allows us to capture power, temperature, and GPU clocks from our full gaming suite. The charts below are the geometric mean across all 15 games, though we also have full tables showing the individual results further down the page.

If you’re wondering: No, PCAT does not favor Nvidia GPUs in any measurable way. We checked power with our previous setup for the same workload and compared that with the PCAT, and any differences were well within the margin of error (less than 1%). PCAT is external hardware that simply monitors the power draw of the PCIe power cables as well as the PCIe x16 slot by sitting between the PSU/motherboard and the graphics card.

We have separate charts for 1080p medium, 1080p ultra, 1440p ultra, and 4K ultra below. We’ll just take them in that order for this review. We also have noise level test results further down the page.

AMD’s RX 7700 XT has a 245W official TBP (total board power), and in testing it averaged 218W at 1080p medium up to 239W at 4K ultra. More telling is that the similarly performing RX 6800 averaged 211W at 1080p medium up to 238W at 4K ultra. In other words, the new RX 7700 XT can’t even clearly exceed the efficiency level of the previous generation RX 6800.

We have to think that a lot of the power must be going into the GPU chiplet interconnect — “a lot” being relative. With a 5nm GCD and 6nm MCDs, you’d normally expect a net improvement in efficiency, and that has generally been the case with the higher tier RX 7000-series parts. But perhaps the RX 7700 XT shows the limits of the scaling; certainly, using 93% of the TGP of the 7800 XT while delivering 85% of the performance (give or take) isn’t a great result.

Nvidia’s competing cards use far less power, with the RTX 4060 Ti averaging 137W to 147W across our test suite. That’s about 80W less power while gaming, give or take, and AMD still has some idle power issues with certain monitors — I know for example that when connected to an Acer Predator X27 G-Sync monitor, the 7700 XT idles at 40W or more. As bad as that might seem, the 7900 XTX is worse, drawing 80–90 watts at idle with that particular screen. Oops.

For now, if you’re trying to get optimal efficiency, Nvidia’s RTX 40-series GPUs are way ahead of anything AMD offers.

GPU clock speeds aren’t extremely important, but it’s interesting to see where they land. Take the RX 7700 XT, which has an official game clock of 2544 MHz. With Nvidia and Intel GPUs, you can normally expect the cards to meet or exceed the stated boost clock, but that’s not always true with AMD’s RDNA 3 GPUs.

The closest we got for our entire 15 game suite to that game clock was at 1080p medium, where the XFX card averaged 2538 MHz. Higher resolutions and settings dropped the average clocks a bit, with a result of 2508 MHz at 4K ultra.

You can see all of the average clocks for each game and setting tested in a table further down the page. The key takeaway is that AMD’s stated Game Clock looks to be reasonable accurate but often overshoots the mark. Nvidia’s stated boost clocks meanwhile often end up being 100–200 MHz lower than our real-world measurements.

The XFX RX 7700 XT isn’t what we’d call a beautiful card, but one thing it does have in ample supply is cooling potential. The peak temperature across our full test suite was 63C, and many games stayed below 60C (with an ambient temperature of around 22C).

Just as important as having low temperatures is keeping noise levels down. Triple fans combined with a large radiator should get the job down, and that’s what we found with our noise testing.

Unlike the reference RX 7800 XT, the XFX 7700 XT noise results are quite good. It’s not the quietest card we’ve tested, but after prolonged testing it stabilized at about 40 dB(A). The fans were just barely moving, at 23% fan speed — below what you can manually set in MSI Afterburner. There are quieter GPUs, but not by much unless they opt to go fanless, which is a whole other can of worms.

For reference, our noise test consists of running Metro Exodus Enhanced, as it’s one of the more power hungry games. We load a save, with graphics set to appropriately strenuous levels (1440p ultra in this case), and then let the game sit for at least 15 minutes before checking noise levels. We place the SPL (sound pressure level) meter 10cm from the card, with the mic aimed at the center of the back fan. This helps minimize the impact of other noise sources, like the fans on the CPU cooler. The noise floor of our test environment and equipment is around 31–32 dB(A).

Given the gaming tests were only pushing 23% fan speed, you can imagine there’s plenty of room for more cooling if the card needs it. We also test with a static fan speed of 75%, which ratcheted up the noise level to 67.4 dB(A). That’s one of the loudest results we’ve encountered, but you’re unlike to hit such high fan speeds unless you intentionally mess with the fan curves.

Here’s the full rundown of all of our testing, including performance per watt and performance per dollar metrics. The prices are based on the best retail price we can find for a new card, at the time of writing. These values can fluctuate, especially on previous generation cards — the RX 6950 XT, 6900 XT, 6800 XT, and 6800 have all changed by about $20–$30 in the past couple of weeks, with only the 6950 XT having gone up in price.

It’s also worth noting that, unofficially, the base price for the RTX 4060 Ti 16GB has dropped to $449. Whether that will be a long-term change and we can expect to see such cards maintain that price level, or if it’s just short-term shenanigans ahead of the RX 7800/7700 XT launch, remains to be seen. For now, that means the RTX 4060 Ti 16GB costs as much as the RX 7700 XT, while the 8GB card holds a $50 pricing advantage.

Based on the current retail prices, the best GPU value of the cards we’ve listed above (at 1440p and in FPS/$) is the RTX 4060, followed by the RTX 4060 Ti. RX 7800 XT ranks third, then the RX 6700 XT, 6750 XT, and finally the RX 7700 XT. The “value” metric may not be the best indicator of ranking, but it’s interesting regardless.

As for efficiency, again at 1440p in terms of FPS/W, the RTX 40-series GPUs range from 0.35 to 0.39 FPS/W, while AMD’s closest is the RX 7800 XT at 0.28 FPS/W, then the RX 7700 XT at 0.25 FPS/W. The previous generation RX 6000-series parts range from 0.19 to 0.25 FPS/W, so AMD did improve in efficiency, just not as much as Nvidia this generation.

(Image credit: Tom’s Hardware)

The Radeon RX 7700 XT isn’t a bad GPU in and of itself; it’s just sitting at the wrong price. Much like the RX 7900 XT at launch, the proximity to the next step up is simply too close. AMD could argue that the RX 7700 XT competes well against the similarly priced RTX 4060 Ti (8GB or 16GB, it’s not a major factor). The problem is that we feel the RTX 4060 Ti is also overpriced. If you have $450 for the 7700 XT, save up another $50 and step up to the superior RX 7800 XT.

Overall, performance from the 7700 XT largely follows the existing RX 6800. It’s a bit faster overall, but only by low single-digit percentages. Even the power is basically the same unless you want to mince hairs over 245W versus 250W TBP. What you get is theoretically more compute, which mostly helps in stuff like AI or Blender, and some new features.

The new features are nice to have, but as we’ve said before, none of them feel like must-haves. AV1 encoding quality is mostly on par with HEVC, except it’s royalty-free, so it’s gaining more widespread support. But for hardware AV1 encoding, which we looked at in the past, Nvidia’s RTX 40-series delivers superior quality. Buying AMD GPUs for their encoding quality isn’t something we’d recommend, in other words.

We can make similar arguments about the DisplayPort 2.1 outputs. Unless you’re really planning on purchasing a new (as-yet-unreleased) DP2.1 UHBR13.5 monitor during the life cycle of these graphics cards, and that also implies running at higher than 4K 240 Hz refresh rates, there’s just no real need for that, particularly on mainstream hardware. Sure, having that feature “just in case” is fine, but we don’t see it as a critical item — more like a checkbox feature.

XFX AMD Radeon RX 7700 XT card photos

(Image credit: Tom’s Hardware)

Besides the pricing, the story of the RX 7700 XT is the same one we’ve been telling for almost three years since the RX 6000-series launched. AMD offers more VRAM, generally better rasterization performance at most price points, and inferior ray tracing performance.

Nvidia also offers extras like DLSS, which remain a selling point as long as you’re not particularly worried about trying to take a stand against proprietary APIs. It would be nice, in theory, if some company were to create a DLSS competitor with all that AI training and other stuff, like DLSS and XeSS, but with support for all GPUs. But where’s the money in trying to do that? Plus, we’d probably just end up at the old XKCD “standards” joke.

For better or worse, Nvidia has the muscle and is using it to push new technologies and features into the graphics market: AI, upscaling, ray tracing, frame generation, and now ray reconstruction.

Even if AMD doesn’t officially cut the price of the RX 7700 XT, we can’t help but think the market will take on that task for them because, much like the RTX 4060 Ti 16GB, this simply costs too much for what you get relative to the competition. The 7700 XT needs to match the price of the RTX 4060 Ti 8GB. Get it down to $400 or less, and suddenly, the RX 7700 XT looks far more compelling. As it stands, you’ll need to have very specific requirements to warrant picking up the 7700 XT over one of the other options.

This post was originally published on this site