AMD Navi Product Reviews and Previews: (5500, 5600 XT, 5700, 5700 XT)

[BRiT]

I'm waiting for SEOB. It's one better in every letter!

 

[Silent_Buddha]

I'm waiting for SEOB. It's one better in every letter!

I just want you to know that I had to load the previous page of posts just to see what the heck you were on about.

Regards,
SB

 

[sniffy]

But it's not, so what's the point?

If your comparative metric is perf/watt then it is matching the competition regardless of node advantage and especially compared to the part it supersedes. We will probably find that AMD have yet again gone for a higher v-core to maintain desired yields and binning. The rest is down to different architecture.

It isn’t matching the competition at all, it is very clearly behind.

I read this vcore excuse everywhere and it is completely unsubstantiated. I can undervolt Nvidia cards just as much, if not more. An RTX 2080 will do 1.5GHz at 700mV. AMD cards don’t even come close to that level of undervolting.

A different, and clearly inferior architecture. I’m not hating, AMDs offerings on the graphics side are just bad. And that’s before you get into the feature deficit.

By the way, I think it’s completely understandable RDNA is what it is. They’ve been so stagnant for so long, how would they possibly match Nvidia, who have been far from stagnant, in one single generation? That’s unrealistic.

 

[PSman1700]

And that’s not good for us, prices are way to high now, besides that consoles are stuck with amd as nvidia would overprice chipsets.

 

[JoeJ]

All this sounds like people want competition only barely good enough so prices drop.
Meanwhile they pay whatever their favorite brand wants.

 

[Deleted member 2197]

AMD Radeon Pro W5700 Professional Graphics Card Reviews

https://www.kitguru.net/components/...-pro-w5700-professional-graphics-card-review/
https://hothardware.com/reviews/amd-radeon-pro-w5700-workstation-gpu-review
https://www.heise.de/newsticker/mel...t-7-nm-Technik-fuer-Workstations-4591352.html
https://aecmag.com/hardware-mainmenu-34/1905-review-amd-radeon-pro-w5700

Interesting to see results from the RTX 2060 Super and RX 5700XT. Also interesting to note the Radeon Pro W5700 is a dual-width card.
https://techgage.com/article/amd-radeon-pro-w5700-performance-review/

 

[DavidGraham]

What's interesting is the much lower compute performance exhibited by the W 5700 (essentially an RX 5700), when compared to the WX 8200 (essentially a Vega 56).
The Vega 56 is actually a 50% to a 100% faster than the RX 5700 in several compute tests.

 

[JoeJ]

What's interesting is the much lower compute performance exhibited by the W 5700 (essentially an RX 5700), when compared to the WX 8200 (essentially a Vega 56).

Interesting: Depending on scene, Navi can both slightly beat or loose badly against GCN in the same Benchmark (luxmark raytracing).
Otherwise i would speculate OpenCL compiler optimizations still not final for Navi yet.
DX12 / VK probably has priority and both use the same compiler which is different from OpenCL. Likely we would get better results looking at this, but no benchmarks exist here that show compute without graphics.
Overall Navi vs. GCN seems just different and mostly Navi has the edge, but some bad losses too.

Personally i'm still worried 7500 could not beat V56 for my own work. If only i knew how close it really is to next gen, i would get me one...
I'm more optimistic for Turing to beat Pascal and i have access now, but not tested yet.

 

[DavidGraham]

but no benchmarks exist here that show compute without graphics.

There are, GPGPU image processing, and Crypto. All showing Vega 56 handily beating RX 5700.
https://hothardware.com/reviews/amd-radeon-pro-w5700-workstation-gpu-review?page=2
Also Vega 56 beats Navi in the most heavy scene in LuxMark.

 

[JoeJ]

I meant a lack of compute benchmarks but using DX or VK, not OpenCL which is used everywhere here.

 

[Kaotik]

There are, GPGPU image processing, and Crypto. All showing Vega 56 handily beating RX 5700.
https://hothardware.com/reviews/amd-radeon-pro-w5700-workstation-gpu-review?page=2
Also Vega 56 beats Navi in the most heavy scene in LuxMark.

There's a reason AMD said GCN still has it's place in compute (and why they still have at least one new GCN chip coming)

 

[Deleted member 2197]

Tough market with a $799 price. Price differential between RTX Quadro 4000 and Radeon W5700 is on average $100, and can find products with a $60 differential.

 

[JoeJ]

There's a reason AMD said GCN still has it's place in compute

But what the hell is this reason?
AMDs says something like 'Navi is optimized for graphics, GCN for compute'. But they don't mention any technical GCN advantage. (IIRC Navi can have 5 cycles of instruction latency vs. 4 on GCN, but that's all i see.)
Reading Navi architecture paper i only see potential advantages, as always with such papers.
Any guesses welcome

 

[DegustatoR]

But what the hell is this reason?
AMDs says something like 'Navi is optimized for graphics, GCN for compute'. But they don't mention any technical GCN advantage. (IIRC Navi can have 5 cycles of instruction latency vs. 4 on GCN, but that's all i see.)
Reading Navi architecture paper i only see potential advantages, as always with such papers.
Any guesses welcome

Higher peak FLOPs per transistor. Possibly per watt also?

 

[JoeJ]

...comparing:

5700XT: 9.7 TF, 10M transistors, TDP 225 W
Vega56: 10.5 TF, 12.5M transistors (= Vega64), TDP 210 W

So not much of a difference here, but there often is in compute benches.
This one the most: https://www.kitguru.net/components/...ro-w5700-professional-graphics-card-review/8/ GCN 2x Navi.
Really confusing because usually Navi is faster with RT.

 

[PSman1700]

GCN is more compute focussed while Navi is more pure gaming performance focus?

 

[Kaotik]

GCN is more compute focussed while Navi is more pure gaming performance focus?

Yes, but what people are wondering is what exactly are the elements that make the difference.

 

[JoeJ]

Exact quote from RDNA architecture whitepaper:

"GCN was compute/throughput focused, RDNA is graphics/latency focused"

ROPs have doubled in comparision to GCN, which explains graphics advantage.
Latency means RDNA can saturate SIMDs more quickly. I think this can benefit both gfx and compute in the same way.
Unsure what they mean with 'throughput'.

 

[PSman1700]

Turing went more compute (and graphics?), Navi the other way around.

 

[DegustatoR]

...comparing:

5700XT: 9.7 TF, 10M transistors, TDP 225 W
Vega56: 10.5 TF, 12.5M transistors (= Vega64), TDP 210 W

So not much of a difference here, but there often is in compute benches.
This one the most: https://www.kitguru.net/components/...ro-w5700-professional-graphics-card-review/8/ GCN 2x Navi.
Really confusing because usually Navi is faster with RT.

Vega is a bad example probably being one of the worse GCN iterations. Try comparing Navi 10 with Polaris 10/20/30 as these are targeted at more or less the same pricing niche.