Official AMD Radeon RX 7800 XT & RX 7700 XT Performance Figures Leaked

[Jism]

Numbers are good. RT is not that mandatory. So you pretty much have a solid card able to beat a 4070 here.

 

[ratirt]

RDNA 3's RT doesn't look improved. Actually it's worse:

View attachment 312247

AMD Radeon RX 7600 Review - For 1080p Gamers

Today AMD is launching their Radeon RX 7600, which is targeted at gamers with 1080p Full HD screens. Our review confirms that thanks to the new RDNA3 graphics architecture, the new card is highly energy efficient.

www.techpowerup.com

Maybe I read it wrong but I see the RT somewhat improved. 7600 100% 6600XT 89%. That is an improvement.


The price is good for sure but the performance? Well, for the price is good but it is a 7800xt. I feel like all cards nowadays are slower even though the name would suggest otherwise.
I feel like the 7900 xt is the 7800 xt and AMD played a charade a bit with the naming.

 

[Cheeseball]

Maybe I read it wrong but I see the RT somewhat improved. 7600 100% 6600XT 89%. That is an improvement.


The price is good for sure but the performance? Well, for the price is good but it is a 7800xt. I feel like all cards nowadays are slower even though the name would suggest otherwise.
I feel like the 7900 xt is the 7800 xt and AMD played a charade a bit with the naming.

I think the 7900 GRE would've been a closer bet. 7900 XT with its 320-bit memory bus and 192 ROPs puts it way ahead of any x800 XT GPU.

 

[TumbleGeorge]

I can see a very heated argument over an obscenely small percentage difference number. In any case, many gamers are disappointed because they expected moar. Perhaps unfounded, but they expected more. For AMD's next series of graphics cards, there are already rumors that it will be short of high-end offerings. This is the RX 8000 series. What's next. Well apparently the RX 9000 series in 2026 or 2027. The series number corresponds to ATI's legendary Radeon 9000 series. If AMD fails to deliver something worthwhile with this, in a sense, anniversary series, it is unlikely that it ever will. Still around 3(+?) whole years left to work towards this goal and hopefully it will succeed, otherwise it will kill all my hope for a bright digital red future. Huh.

 

[ratirt]

I think the 7900 GRE would've been a closer bet. 7900 XT with its 320-bit memory bus and 192 ROPs puts it way ahead of any x800 XT GPU.

Perhaps but this one is not in Europe where I reside currently (Norway) or is it? I thought it is China exclusive.

 

[ARF]

Maybe I read it wrong but I see the RT somewhat improved. 7600 100% 6600XT 89%. That is an improvement.

And yet there are the RX 6600 and RX 6650 XT that are better than RX 6600 XT in the first case, and RX 7600 in the second.
It is not an improvement because you didn't consider the RX 6650 XT. Why did you skip it?

I can see a very heated argument over an obscenely small percentage difference number. In any case, many gamers are disappointed because they expected moar. Perhaps unfounded, but they expected more. For AMD's next series of graphics cards, there are already rumors that it will be short of high-end offerings. This is the RX 8000 series. What's next. Well apparently the RX 9000 series in 2026 or 2027. The series number corresponds to ATI's legendary Radeon 9000 series. If AMD fails to deliver something worthwhile with this, in a sense, anniversary series, it is unlikely that it ever will. Still around 3(+?) whole years left to work towards this goal and hopefully it will succeed, otherwise it will kill all my hope for a bright digital red future. Huh.

Many gamers will be disappointed because the upgrade cycle is ruined. They will have nothing to upgrade with. No one upgrades for 2-3% performance, or even 20-30% performance.

 

[ratirt]

And yet there are the RX 6600 and RX 6650 XT that are better than RX 6600 XT in the first case, and RX 7600 in the second.
It is not an improvement because you didn't consider the RX 6650 XT. Why did you skip it?

it is the same exact chip but with boost isn't it? You can argue about the XT not being there or being there. You can argue that the 6650 was released later on with matured silicon and boost in clocks as it comes with the maturity. There is a lot you can argue about but I'd rather focus on here an now which means we look at what the 7600 is tackling from the competition and how does it look in terms of price. Which is $269 vs $399 you can try guessing which price tag belongs to which graphics. I leave space for opinions regarding is it an improvement or not.
In my eyes it is but I can argue.

 

[Chrispy_]

Maybe I read it wrong but I see the RT somewhat improved. 7600 100% 6600XT 89%. That is an improvement.


The price is good for sure but the performance? Well, for the price is good but it is a 7800xt. I feel like all cards nowadays are slower even though the name would suggest otherwise.
I feel like the 7900 xt is the 7800 xt and AMD played a charade a bit with the naming.

I think that whole insane argument for the first two pages of comments was because of a simple inability by some people to grasp the distinction between "raytracing performance" and "relative raytracing performance".

I thought IQs on TPU forums were higher than that but here we are.

Yes, the RX7600 raytraces ~11% faster than a 6600XT, so that's an indisputable 11% RT improvement.
However, the RX7600 is also ~11% faster at everything else, so there's no relative improvement; Turning on RT effects hurts RDNA3 just as much as it hurts RDNA2.
Effectively, RDNA3 raytraces faster than RDNA2 when it's clocked higher, but clock for clock, it adds nothing.

Today's reviews are going to be interesting, because the 6800XT has 20% more compute units and ray accelerators than the 7800XT, so it's going to be about whether the clocks and bandwidth can make up the difference. Early leaks say 'not quite' which is intuitive really, but the 'real' reviews should be released from their embargo in a matter of hours.

 

[ARF]

it is the same exact chip but with boost isn't it? You can argue about the XT not being there or being there. You can argue that the 6650 was released later on with matured silicon and boost in clocks as it comes with the maturity. There is a lot you can argue about but I'd rather focus on here an now which means we look at what the 7600 is tackling from the competition and how does it look in terms of price. Which is $269 vs $399 you can try guessing which price tag belongs to which graphics. I leave space for opinions regarding is it an improvement or not.
In my eyes it is but I can argue .

You have no proof that RX 6650 XT uses a different silicon. It's simply the same as in RX 6600 XT. The card was rebranded.
If you argue about the binning, then maybe RX 6650 XT gets a notch higher bin chips but maybe it gets a notch lower. You simply don't know.

 

[Chrispy_]

You have no proof that RX 6650 XT uses a different silicon. It's simply the same as in RX 6600 XT. The card was rebranded.
If you argue about the binning, then maybe RX 6650 XT gets a notch higher bin chips but maybe it gets a notch lower. You simply don't know.

Good god man, learn to read.
He never said different silicon, he said matured silicon - ie, the same silicon on the same, more mature process.

The 6650XT is more than a rebrand, btw - it has more memory bandwidth with 18Gbps GDDR6, and in certain games (especially at higher resolutions) the 6650XT is faster than a 6600XT even when both are locked to the exact same clock speeds.

 

[ZoneDymo]

I'm going ahead and say these will be super meh like the rest of them, just fine for the price, no real shake up

 

[Frick]

I thought IQs on TPU forums were higher than that but here we are.

IQ has nothing to do with it.

Anyway, I jumped the gun and ordered a 6950xt. We'll see if I return it and get a 7800xt instead, but I suspect I will not.

 

[Chrispy_]

IQ has nothing to do with it.

Anyway, I jumped the gun and ordered a 6950xt. We'll see if I return it and get a 7800xt instead, but I suspect I will not.

You won't want a 7800XT after ordering a 6950XT. It's likely to be a 10%+ downgrade and all you get in return for that is an AV1 encoder. 3 hours to go until the review embargo ends....

 

[Macro Device]

a simple inability by some people to grasp the distinction between "raytracing performance" and "relative raytracing performance".

A million times this.

I tried hard to explain it but apparently no matter how hard you try there always will be someone who calls you a jerk instead of trying to understand you.

 

[ARF]

Good god man, learn to read.
He never said different silicon, he said matured silicon - ie, the same silicon on the same, more mature process.

The problem that I see is that I have seen regression over time, not more mature process. Later batches get worse bins.

 

[ratirt]

The problem that I see is that I have seen regression over time, not more mature process. Later batches get worse bins.

I doubt that is true and that you see the bigger picture here. The silicon has different quality in a wafer. Some chips are better some are slightly worse and/or have defects. You can use the worse one or with a defect for a different product assuming it meets the requirements of the end product. Maybe, later in development, a company possesses more of those lower quality chips and they want to do something with it so they release lower tier product. I'd rather go with this than saying that maturity degrades the silicon and the quality of the silicon gets worse over time.
7800XT can be that example. Less shading units than the 7900xt and xtx so AMD can make 7800xt's and lower tier cards instead of getting a dedicated silicon just for that purpose. Then you check the defects on a silicon if it can be used for 7800xt or lower and you just use it. The maturity of the process will definitely make the silicon better no doubt and the number of defects is less.

 

[Chrispy_]

The problem that I see is that I have seen regression over time, not more mature process. Later batches get worse bins.

Cite a single source please; That's not how semiconductor manufacturing works.

If you're talking about a sample size of one where you bought a new GPU with the same die as an older GPU, then that's the silicon lottery, as Ratirt says - but also there's no way with AMD to know when an actual die was made. Navi23 in a brand-new 6650XT could have been manufactured before or after the 6600XT, only it's been sat in inventory with AMD, the AIB partner, or at the retailer you bought it from.

As a rule of thumb, yields increase over time as the foundry gets better experience and refines the process. The overall trend is that the average ASIC quality increases over the course of production. You can still obviously get a new runt die having previously got lucky on an early golden sample, but the average improvements over time are precisely why rebrands often come with a mild clock speed increase.

 

[evolucion8]

A million times this.

I tried hard to explain it but apparently no matter how hard you try there always will be someone who calls you a jerk instead of trying to understand you.

Like you for example. Relative RT performance is a different metric from pure ray tracing performance. As you noticed by now, the 7800XT has a 1,280 core deficit, over 33% less bandwidth, half the Infinity Cache, 25% less Ray Accelerators and 33% less ROPs and yet it can match the 6900XT in nearly al scenarios, exceeding it constantly in heavy GPU workloads and RT by a small margin, and losing against it in more CPU bound/high FPS scenarios. Showing that the 17% IPC increase per CU at the same clockspeed and the improvements in the RT like early curling, offloading more of the BVH workload in fixed hardware and better utilization of the existing CU engines for other RT related tasks like Denoising and shading, is actually real. Regardless if both loses almost the same amount of performance or not when using RT.