• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

AMD Vega (FE and RX) Benchmarks [Updated Aug 10 - RX Vega 64 Unboxing]

Page 11 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.
DiogoDX said:
Mine is default HUD.

EDIT: Curious to see 4K performance since is my native resolution and the 980Ti get destroyed on the same settings. I have to turn down a lot of settings and play with 45~50 fps.
Click to expand...

From Tom's review of a 1080 Ti AM card (http://www.tomshardware.com/reviews/evga-gtx-1080-ti-ftw3-gaming,5061-4.html):

1080 Ti (300W): 73.7
1080 Ti FE: 68.3
Titan X (Pascal): 67.5
1080 FE: 49.6
1070 FE: 39.2
Fury X: 39.1
980 Ti Ref: 37.9

For Vega 64, I'll give the O/U of 58.
 
Head1985 said:
I think GTX1070 can only manage 3triangles/clock with 3x GPC and GTX2070 will be probably again 3x gpc with 3x triangles/clock.
But yeah 4x SE will be probably again bottleneck.AMD need 6x or better 8x shader engines.
Click to expand...

Actually, I believe that the equivalent of AMD's Geometry Processor is the nVidia's Polymorph Engine. A Geometry Processor can do one triangle per clock and recent AMD designs have four Geometry Processors for four triangles per clock. nVidia designs have 15, 20 or 30 Polymorph Engines for the 1070, 1080 and 1080 Ti respectively, so even at one triangle every two clocks, you end up with 7.5, 10 and 15 triangles per clock respectively, although I don't believe that any nVidia cards can achieve this throughput (they just aren't bottlenecked by triangle setup).
 
Bouowmx said:
Live testing of AMD Vega Frontier Edition

One result already obtained: TIme Spy 6785 graphics score
Click to expand...
1650MHz and only 6785 graphics score? That's actually pretty bad. Even slower than a 1080 and we're talking about Time Spy and DX12.
 
Karnak said:
1650MHz and only 6785 graphics score? That's actually pretty bad. Even slower than a 1080 and we're talking about Time Spy and DX12.
Click to expand...

I was just thinking the same thing... that's really really disappointing. Hopefully it's just old drivers that are holding it back.
 
If they can release the vega gaming version and it it is not crippled in any way, is not a paper launch, has drivers working on day 1 and is priced at $399 there might be something left of my stock by next month >.<
 
Wall Street said:
Actually, I believe that the equivalent of AMD's Geometry Processor is the nVidia's Polymorph Engine. A Geometry Processor can do one triangle per clock and recent AMD designs have four Geometry Processors for four triangles per clock. nVidia designs have 15, 20 or 30 Polymorph Engines for the 1070, 1080 and 1080 Ti respectively, so even at one triangle every two clocks, you end up with 7.5, 10 and 15 triangles per clock respectively, although I don't believe that any nVidia cards can achieve this throughput (they just aren't bottlenecked by triangle setup).
Click to expand...

Actually it's equivalent to the GPCs ...

Rasterizer rate is tied to the GPCs ...
 
I get just over 100 fps at 1080p Ultra in that location in Witcher 3 with a GTX 1080 Ti. The problem is it is 1080p and in the city, so it is CPU bound.

Running 4K ultra + hairworks off would give a better indication of GPU-bound performance.
 
utahraptor said:
If they can release the vega gaming version and it it is not crippled in any way, is not a paper launch, has drivers working on day 1 and is priced at $399 there might be something left of my stock by next month >.<
Click to expand...

To put it bluntly, gaming Vega is almost irrelevant to the stock price at this point - at least if the market participants are paying attention. The real profits are to be found in the server market - the EPYC CPU looks to be very competitive. This is where the lion's share of AMD's profits will come from in the next couple of quarters. To the extent Vega matters at all, it's the high-margin workstation and Instinct parts that will be bringing in the dollars.

It's a technical disappointment and yet another epic marketing fail (seriously, fire all those guys already), but it shouldn't hurt the stock much.
 
oh man i am not gonna be able to upgrade unless the RX Vega's are a serious bargain... between the seemingly underwhelming FE performance and the rumors of the RX series being power hungry beasts I am becoming a bit more complacent with Polaris. i didn't expect 1080ti or titan x performance from the Frontier card, but considering the green team is still pulling ahead with Volta, I feel like Vega is too little and much too late. hopefully i'm dead wrong but now I'm more worried than excited.
 
Disappointing to the point of unreasonableness, to the point where I'd say the smart money is on a significant performance increase with gaming drivers next month for RX Vega... but I wouldn't assign a high amount of confidence to that either.

Truly a bizarre launch in any case.
 
jpiniero said:
You really think so? The 1080 Ti FE is 70-85% faster in most games than the Fury X, and sometimes more. That's a lot to ground to cover with no extra cores or ROPs.
Click to expand...

Fury X was a severely bottlenecked card. In terms of perf/TFlop it was substantially worse than Polaris. Many of us expected (or at least hoped) that these bottlenecks would be eliminated (now that reticle size was no longer a limiting factor) and that the new hardware features would increase perf/TFlop to Maxwell/Pascal levels. That would have resulted in a competitive product.

Instead, we get what appears so far to be an epic disaster. The gaming benchmarks look very much like what you'd expect if you took Fury X and were able to boost the clock speeds to 1400-1600 MHz. Architectural improvements since then don't seem to have made any difference at all in ordinary gaming. The only thing I can think of that could possibly excuse this is if the Draw Stream Binning Rasterizer (tiled rendering) isn't active in the current drivers. Otherwise, it indicates that whatever bottlenecks held back Fury X are still present. The most likely culprit is the front end; we know from Linux drivers that there are still only 4 shader engines. And, as you noted, there are only 64 ROPs (GP102 has 96). It looks like the cut-rate Chinese design team that put together Vega (and Polaris) wasn't skilled enough to work around the limitations of GCN. Vega was supposed to remove the 4 shader engine limitation but it doesn't appear that they were able to pull this off.

Another disappointment is clock rates. Pascal can easily do 1600-1700 MHz without substantially sacrificing efficiency, and easily overclocks to nearly 2000 MHz. In contrast, it looks like Vega barely clocks higher than Polaris. We were expecting 1600 MHz as the sustained speed, but it looks like this will only be obtainable on watercooled cards, if that. And the power budget is going to be obscene. "Typical Clock" for aircooled Vega FE is only 1382 MHz, which is barely better than Polaris 20 and far worse than the Pascal competition.

It's time for Raja to go. No longer can we make the excuse that he's just inheriting someone else's problems. Vega is his legacy, and it looks like it's going to cripple RTG for years to come. Enough. Get rid of Raja, get rid of the cheap Chinese design team, and reboot RTG in America where it belongs.
 
JDG1980 said:
Fury X was a severely bottlenecked card. In terms of perf/TFlop it was substantially worse than Polaris. Many of us expected (or at least hoped) that these bottlenecks would be eliminated (now that reticle size was no longer a limiting factor) and that the new hardware features would increase perf/TFlop to Maxwell/Pascal levels. That would have resulted in a competitive product.

Instead, we get what appears so far to be an epic disaster. The gaming benchmarks look very much like what you'd expect if you took Fury X and were able to boost the clock speeds to 1400-1600 MHz. Architectural improvements since then don't seem to have made any difference at all in ordinary gaming. The only thing I can think of that could possibly excuse this is if the Draw Stream Binning Rasterizer (tiled rendering) isn't active in the current drivers. Otherwise, it indicates that whatever bottlenecks held back Fury X are still present. The most likely culprit is the front end; we know from Linux drivers that there are still only 4 shader engines. And, as you noted, there are only 64 ROPs (GP102 has 96). It looks like the cut-rate Chinese design team that put together Vega (and Polaris) wasn't skilled enough to work around the limitations of GCN. Vega was supposed to remove the 4 shader engine limitation but it doesn't appear that they were able to pull this off.

Another disappointment is clock rates. Pascal can easily do 1600-1700 MHz without substantially sacrificing efficiency, and easily overclocks to nearly 2000 MHz. In contrast, it looks like Vega barely clocks higher than Polaris. We were expecting 1600 MHz as the sustained speed, but it looks like this will only be obtainable on watercooled cards, if that. And the power budget is going to be obscene. "Typical Clock" for aircooled Vega FE is only 1382 MHz, which is barely better than Polaris 20 and far worse than the Pascal competition.

It's time for Raja to go. No longer can we make the excuse that he's just inheriting someone else's problems. Vega is his legacy, and it looks like it's going to cripple RTG for years to come. Enough. Get rid of Raja, get rid of the cheap Chinese design team, and reboot RTG in America where it belongs.
Click to expand...
They cant do anything without money and AMD didnt give them.All money after 7970/290x went to ryzen.
 
Malogeek said:
Yeah then he ran Witcher 3 again to test (with no difference) and I noticed he had vsync on as well. It must have been running triple buffering.
Click to expand...

Was there no difference? The card was reportedly running between 1350 and 1520 in gaming mode, while running the same test in pro mode at 1650 MHz. With no difference in the frame rate.
 
Maverick177 said:
Imagine the humiliation when Vega comes out and it's barely faster than a GTX 1080.
Click to expand...

It's not hard to imagine the failure of Vega. The chances of Firestrike not being optimized in drivers is pretty low at this point. That's probably one of the 1st thing you optimize for. Vega ~ 1080 right now. It might be slightly faster later. Vega is going to be the biggest flop for AMD since Bulldozer when compared to its competitors. It's such a horrible card. High TDP, huge die size, expensive HBM2, and very, very hard to produce. This is a pretty big monumental failure.
 
JDG1980 said:
It's time for Raja to go. No longer can we make the excuse that he's just inheriting someone else's problems. Vega is his legacy, and it looks like it's going to cripple RTG for years to come. Enough. Get rid of Raja, get rid of the cheap Chinese design team, and reboot RTG in America where it belongs.
Click to expand...
Think you said this before?
Those guys made Polaris, and for most things, it does a pretty good job.
Vega is a brand new design, and not a re-spin of Polaris.

We just don't know what the story is on Vega, could be HBM2 issues/costs, could be GloFlo issues, could be bad yields (which is why they are going "high end" first) could be a number of things.

Most people think it is late, should have used GDDR5(X), and be priced around $500 level.
No arguments here.

Yeah, (most) everyone was hoping they will pull off another Ryzen type of 'win', and early results aren't particularly encouraging.

Just wait a bit, and things will become clearer in the next AMD conference call.
Maybe heads will roll, but, we just don't know the facts yet.
 
Status
Not open for further replies.

TRENDING THREADS

Back
Top