Sapphire RX 7900 GRE Nitro+

[W1zzard]

Sapphire's RX 7900 GRE flagship card, the Nitro+, impresses with its superb cooler featuring ARGB illumination and unparalleled quietness, topping tests conducted today. Its gaming performance beats the RTX 4070 Super, but ends up slightly more expensive than that.

Show full review

 

[AnarchoPrimitiv]

Saw a 7900XT on Newegg for $699 last night....at that price, does that make the 7900GRE worse as a value proposition?

 

[Denver]

It would be interesting how much drivers are influencing the difference in performance between the 7900XT/7800XT vs GRE. AMD was working hard to improve performance in games like "Allan Woke".

 

[Nostras]

It seems W1zzard has tested 4 cards who are all equally mid really.
If HUB is to be believed all Sapphire models are kinda poor (although you could argue that the Pulse at least could be ok value) but flop compared to the hellhound (and probably TUF).

(I timestamped it)
Sapphire really needs to fix it as their coolers as they are kinda poor and not worth recommending.

 

[W1zzard]

It would be interesting how much drivers are influencing the difference in performance between the 7900XT/7800XT vs GRE. AMD was working hard to improve performance in games like "Allan Woke".

I tested 24.1.1 vs the AMD press driver, no differnce

 

[AnotherReader]

Saw a 7900XT on Newegg for $699 last night....at that price, does that make the 7900GRE worse as a value proposition?

I think the cheaper models are still fine, but the premium variants like the Nitro look less appealing.

 

[Denver]

I tested 24.1.1 vs the AMD press driver, no difference

It's strange that the HUB review shows almost no difference between the 7800XT and GRE, I assumed it was the driver effect. So, GRE aside, all AMD GPUs are running on 24.1.1 driver and not 23.11.1 WHQL?

 

[AnotherReader]

It's strange that the HUB review shows almost no difference between the 7800XT and GRE, I assumed it was the driver effect. So, GRE aside, all AMD GPUs are running on 24.1.1 driver and not 23.11.1 WHQL?

It's probably a consequence of @W1zzard 's wider test suite: 25 games vs 12 games used by Hardware Unboxed.

 

[Denver]

It's probably a consequence of @W1zzard 's wider test suite: 25 games vs 12 games used by Hardware Unboxed.

If you only check the games that are present in both, you'll notice that the difference between the 7800XT and GRE on HUB is almost nil, in contrast to the 10-15% of TPU.

 

[W1zzard]

It's strange that the HUB review shows almost no difference between the 7800XT and GRE, I assumed it was the driver effect. So, GRE aside, all AMD GPUs are running on 24.1.1 driver and not 23.11.1 WHQL?

as listed, all AMD cards except GRE are running 23.11.1. For your specific query "does the reviewer driver gain big perf in Alan Wake?" I posted the screenshot above

 

[Dr. Dro]

It's strange that the HUB review shows almost no difference between the 7800XT and GRE, I assumed it was the driver effect. So, GRE aside, all AMD GPUs are running on 24.1.1 driver and not 23.11.1 WHQL?

They retain the same level of performance in a wide swath of games. Unfortunately, the GRE exemplifies the extremely poor scaling of RDNA 3 architecture at the high end, and makes of this a pretty lame product overall. IMO, anyone with in the market for an AMD GPU with their sights set on this segment should just do themselves a favor and either buy a 7800 XT or shell out a bit more for the 7900 XT.

 

[BoGGe]

i think Catalyst 24.1.1 has some bug with 7900 cards and power usage. idle wattage is a little lower than 23.12.1 but browser/videoplayback is much worse. my 7900 XT card peaks about 100-110W while watching youtube clips. and hover around 60-70W otherwise. with 23.12.1 it between 40-60W max.

 

[AnotherReader]

They retain the same level of performance in a wide swath of games. Unfortunately, the GRE exemplifies the extremely poor scaling of RDNA 3 architecture at the high end, and makes of this a pretty lame product overall. IMO, anyone with in the market for an AMD GPU with their sights set on this segment should just do themselves a favor and either buy a 7800 XT or shell out a bit more for the 7900 XT.

The RDNA3 high end actually scales pretty well with increasing resources, but the GRE is fairly cut back. In addition to losing 5% of the 7900 XT's CUs, it has lower clock speed and lower bandwidth to both L3 and DRAM. I suspect that besides the lower clock speed, the lower capacity and bandwidth L3 is also impacting performance.

 

[Dr. Dro]

The RDNA3 high end actually scales pretty well with increasing resources, but the GRE is fairly cut back. In addition to losing 5% of the 7900 XT's CUs, it has lower clock speed and lower bandwidth to both L3 and DRAM. I suspect that besides the lower clock speed, the lower capacity and bandwidth L3 is also impacting performance.

What I mean is, if you look at the theoretical maximums and how this card should be performing, it should still come off at least as beefy as a 4080...

 

[AnotherReader]

What I mean is, if you look at the theoretical maximums and how this card should be performing, it should still come off at least as beefy as a 4080...

I see what you mean. The 4080 clocks substantially higher than this card so that explains part of the loss. The lower capacity and lower bandwidth L3 also hampers it. The rest is explained by AMD's dual issue feature not enhancing average performance in graphics workloads by as much as Ampere and Ada's doubled FP32 throughput per SMX.

 

[RandallFlagg]

These are pretty attractive at the stated price point. More and more I favor AMDs GPUs, by all rights this should really ramp up pressure on Nvidia's 4070 Super pricing, but probably won't which is fine by me. If I were looking at upper midrange \ lower high end GPUs right now, I'd probably go with this.

In fact, if prices drop a bit below MSRP late this year with the newer generation GPUs coming out, I'll probably go for this one.

My only reservation is the multi-monitor power draw, annoying on AMD cards, though obviously not a deal breaker for me.

 

[beautyless]

My area still selling 7900xt at 900$.

 

[Macro Device]

it should still come off at least as beefy as a 4080...

7800 XT (60 CUs, 75%) is 89% as fast as 7900 GRE at 4K (or 7900 GRE is 112% as fast as 7800 XT). Meaning if it scaled linearly then 80 CUs should've been 133% as fast as 7800 XT's 60.

However, 4080 is 50% faster than 7800 XT. You don't expect a 33% more advanced GPU to compete with a 50% more advanced GPU.

Overall, "but expensive" alone seals the deal. 7900 GRE is criminally VRAM bandwidth limited which makes it a bad deal for its money. 7800 XT has faster VRAM, that's why it's occasionally providing more FPS. If one could hack it and make VRAM run at 2600+ MHz it would've been a very interesting purchase for enthusiasts but realistically, I'd much prefer a 7900 XT at this point. Much more reliable.

 

[Rabit]

AMD announced that 7900 GRE memory OC limit at 3% is bug and will be resolved soon. If we look at 18 Gbps on Rx 7700Xt some models OC can reach up to 2540Mhz we can hope for decent uplift in performance.

Info in

 

[mahoney]

Is this the most expensive GRE model?

 

[nikoya]

- bought 7900XTX nitro+ new Jul23 7000RMB(972US$) in 4k it achieves 137% = 51RMB (7.09US$) per %
- 7900GRE nitro+ in China now cost 5000RMB(695US$) in 4k achieves 100% = 50RMB (6.95 US$) per %

same
happy

 

[Taisho]

64W power draw when watching cat videos on YT, sounds fair.

 

[ViperXZ]

Two things: why is the 20GBits vram down clocked to just 18 again? The gpu could surely use more bandwidth. Edit: it seems AMD buys these in bulk and limits them to 18Gbits thus saving money.

If two dies are dummy dies the combined transistor count is far less than 58B. Just saying.

edit:
It’s sad how limited OC is on this GPU by the software, it has way higher potential.

 

[W1zzard]

If two dies are dummy dies the combined transistor count is far less than 58B. Just saying.

That's a great point and I lean towards adjusting the numbers in the review/GPU-Z/GPU database, but will that lead to confusion because people see different counts for the same GPU?

 

[ViperXZ]

That's a great point and I lean towards adjusting the numbers in the review/GPU-Z/GPU database, but will that lead to confusion because people see different counts for the same GPU?

I think it will lead to people thinking you have (more) attention to detail, so I only see the positive side of it, I don’t think negative sides are really a important thing to consider. So I’m all for it, I think in the future this becomes more and more important as more such designs are released.