AMD Radeon RX Vega 64 and 56 Reviews [*UPDATED* Aug 28]

[itsmydamnation]

So my playing with vega 56 so far is you are completely power limit bound. So its trying to get max stable dynamic clock with max undervolt......

I haven't touched memory OC yet.......

 

[Rasterizer]

AMD Radeon RX Vega 56 gets faster with Vega 64 BIOS

I think this is pretty convincing evidence that Vega is severely bottlenecked at the moment, as the 8 additional CUs of Vega 64 are providing no additional performance whatsoever over Vega 56 with the same BIOS. Considering how much discussion there has been of Fiji being front end bottlenecked over the last two years, and that the core Vega uarch features that would change its front end relative to Fiji haven't been implemented yet, I think it is pretty obvious to infer that Vega is currently front end bottlenecked as well.

I'm working figuring out how to show how much of that front end bottleneck is geometry throughput vs anything else.

 

[guachi]

It isn't just retailers getting in on the Vega64 markups, was in a aisle across from 2 guys stocking stuff, and he goes "Dude, Jerry got 3 vega cards used his discount, and he flipped them at ebay for $750" and they were laughing about it. This was at BestBuy.

I don't get how you could know about Vega being launched and not also know it's not worth $750. I mean, I like AMD, I'd like to get a Vega 64. But I don't $750 like it.

 

[Muhammed]

-Vega is going to be awesome, poor Volta, wait for Vega!
-Vega demo is slightly faster than 1080? No that's Fiji drivers
-Vega FE is slower than 1080? No that's pro drivers, gimped in gaming
-RX Vega leaks slower thab 1080? Early gimped drivers missing features
-RX Vega 64 slower than 1080, liquid cooled 64 trading blows with 1080? Shock, disbelief, denial. Missing features, early drivers, wait for fearure X, Vega will be awesome! Wait for proper Vega drivers!

And the myth goes on and on.

 

[krumme]

So my playing with vega 56 so far is you are completely power limit bound. So its trying to get max stable dynamic clock with max undervolt......

I haven't touched memory OC yet.......

Same here on rx64. Have to apply 25% power.
Avfs and wattman seems buggy. Selecting lower voltage makes it revert back to 1.2 i think. Tries other voltage tools as well. Cant crash it from selecting lower voltage lol.

Its pretty apparent when you have the card they need to adress power usage. Its power limited like crazy.

 

[Krteq]

-Vega is going to be awesome, poor Volta, wait for Vega!
-Vega demo is slightly faster than 1080? No that's Fiji drivers
-Vega FE is slower than 1080? No that's pro drivers, gimped in gaming
-RX Vega leaks slower thab 1080? Early gimped drivers missing features
-RX Vega 64 slower than 1080, liquid cooled 64 trading blows with 1080? Shock, disbelief, denial. Missing features, early drivers, wait for fearure X, Vega will be awesome! Wait for proper Vega drivers!

And the myth goes on and on.

Yep, it's been confirmed from many sources that new uarch features are not yet enabled in current public drivers

Myth busted

//@krumme: I'm not the one derailing a thread again

 

[krumme]

Yep, it's been confirmed from many sources that new uarch features are not yet enabled in current public drivers

Myth busted

Stop. I am going to report him. We are not going down that road again. Trash discussion.

 

[Veradun]

Some of you guys are hilarious with your expectations of this new magical driver.

When in the history of GPU's has there ever been a driver release that boosted performance over 20%?(other than when nvidia Multithreaded their drivers a few years ago)

Of course it never having happened before doesnt mean it cant, but it does mean it is extremely unlikely.

https://blogs.nvidia.com/blog/2017/07/31/titan-xp-drivers-new-levels-of-performance-for-creatives/

Our latest driver — available today — delivers 3x more performance in applications like Maya to help you create and design faster than ever.

 

[deathBOB]

https://blogs.nvidia.com/blog/2017/07/31/titan-xp-drivers-new-levels-of-performance-for-creatives/

But Titan was intentionally gimped vs the real pro cards. The Vega situation is not similar.

 

[Despoiler]

https://blogs.nvidia.com/blog/2017/07/31/titan-xp-drivers-new-levels-of-performance-for-creatives/

I pointed out somethings in another thread. This was a straight up fabrication by Nvidia. Testing with these drivers showed Maya to have REGRESSED significantly in performance. There was only one app that met or exceeded the 3x claim, snx. Also, Nvidia provided zero notes as to how they tested to achieve their claims. AMD might massage a scenario for good numbers, but at least they are transparent as to how they achieved the numbers.

 

[Konan]

Flashing 56 bios with 64 - https://videocardz.com/72299/amd-radeon-rx-vega-56-gets-faster-with-vega-64-bios

though it was expected, the sample used by KDtree did not unlock to full 4096-core Vega. This means that flashing 64 BIOS into 56 will not give you any extra boost thanks to more cores. However, the BIOS will change clock speeds which apparently have a much greater effect on the performance.

The RX Vega 64 BIOS has 1545 MHz boost clock and 945 MHz HBM2 clock. The RX Vega 56 is slower with boost clock at 1471 MHz and memory clock at 800 MHz. The BIOS mod gives you the same clocks on your 56, which practically improves performance.

By increasing the frequency by 75 MHz the modder noticed that the card was only 2% slower than reference RX Vega 64.

Furthermore, it was proven that overclocked RX Vega 56 at 1650 MHz (core) 2200 MHz (memory) actually surpasses 64 at default clocks.

 

[Veradun]

But Titan was intentionally gimped vs the real pro cards. The Vega situation is not similar.

I sure hope they've intentionally disabled features in Vega drivers and it wasn't an accident

 

[Rifter]

For me, Vega was a disappointment because I was hoping for a 1080Ti competitor. It isn't a failure though IMO. Both Vega cards compete well with GP104.

If you believe some people here it will hit 1080ti perf soon.

And they do compete well with the 1080 in perf, but certainly not in perf/watt.

Like you though i was expecting a 1080ti competitor not a 1080, but hey who knows if the super optimistic crew is right we may get there in the coming weeks. Then all that will be left is seeing if prices stabilize.

 

[Rifter]

https://blogs.nvidia.com/blog/2017/07/31/titan-xp-drivers-new-levels-of-performance-for-creatives/

Maya is a productivity program not a game is it not?

 

[Despoiler]

If you believe some people here it will hit 1080ti perf soon.

And they do compete well with the 1080 in perf, but certainly not in perf/watt.

Like you though i was expecting a 1080ti competitor not a 1080, but hey who knows if the super optimistic crew is right we may get there in the coming weeks. Then all that will be left is seeing if prices stabilize.

I guess you missed that both the 56 and 64 exceeded an overclocked 1080ti in PUBG @ 1080p no less. How is that possible if it's not?

 

[Veradun]

Maya is a productivity program not a game is it not?

I'll highlight for you the part of your post in which you said you meant the uplift had to be in games:

Some of you guys are hilarious with your expectations of this new magical driver.

When in the history of GPU's has there ever been a driver release that boosted performance over 20%?(other than when nvidia Multithreaded their drivers a few years ago)

Of course it never having happened before doesnt mean it cant, but it does mean it is extremely unlikely.

Note: there is no highlightning.

Care to explain how improving performance in a pro program through an updated driver is different from improving performance in a game through an updated driver?

 

[zinfamous]

I guess you missed that both the 56 and 64 exceeded an overclocked 1080ti in PUBG @ 1080p no less. How is that possible if it's not?

At this point, it's nothing more than an outlier, especially considering the ~beta? state of whatever PUBG is. Interesting and strange, of course, but a bizarre outlier until more data arrives.

 

[Rifter]

I'll highlight for you the part of your post in which you said you meant the uplift had to be in games:



Note: there is no highlightning.

Care to explain how improving performance in a pro program through an updated driver is different from improving performance in a game through an updated driver?

Then i guess i miss your point, unless you are trying to say that AMD is intentionally gimping compute performance like Nvidia did for the titan compute drivers?

If AMD could flip a switch and unleash this performance like Nvidia did why would they have not done so for launch?

Nvidia could have done this at any time, they just had to flip a switch in the drivers, i dont see how this compares to the current vega situation unless AMD is truly trolling all of us.

 

[Muhammed]

I guess you missed that both the 56 and 64 exceeded an overclocked 1080ti in PUBG @ 1080p no less. How is that possible if it's not?

Lol I guess you are referencing the PCGH test? Typical case of grapsing at straws.

The test was done at high settings, not Ultra, and the 1060 had the same performance of the 1080Ti. It was a typical case of cpu limitation! At 1440p and 4k Vega dropped behind a 1070, at Ultra Vega is so far behind as well in every resolution.

 

[Rifter]

I guess you missed that both the 56 and 64 exceeded an overclocked 1080ti in PUBG @ 1080p no less. How is that possible if it's not?

Thats one game, one, uno, singular. I could post dozen upon dozen in which the 1080Ti destroys vega, you know like in 99.99% of the benchmarks posted in post one of this very thread. But. i still havent given up completely, you guys may afterall be right, there could be a major driver release in few weeks fixing performance, ill need benchmarks to believe it though. Time will tell, if AMD can turn Vega into a Volta competitor with pure software then i have truly underestimated them.

 

[Despoiler]

Lol I guess you are referencing the PCGH test? Typical case of grapsing at straws.

The test was done at high settings, not Ultra, and the 1060 had the same performance of the 1080Ti. It was a typical case of cpu limitation! At 1440p and 4k Vega dropped behind a 1070, at Ultra Vega is so far behind as well in every resolution.

Literally no one runs PUBG on ultra. This the test PCGH did was of the "real world" type testing. That was their goal. It wasn't to sus out GPU limited data. The most common real world settings are ultra distance, high textures(or there abouts), and low everything else. Everyone I know uses though. AMD or Nvidia it doesn't matter. Also, Nvidia has CPU utilization issues now? I thought that was AMD? The point is that AMD at no point should be able to best a 1080ti if it's not possible. When the settings are cranked up towards GPU limited Vega drops off suggesting a bottleneck. You know the kind that happen when Vega features are disabled.

Thats one game, one, uno, singular. I could post dozen upon dozen in which the 1080Ti destroys vega, you know like in 99.99% of the benchmarks posted in post one of this very thread. But. i still havent given up completely, you guys may afterall be right, there could be a major driver release in few weeks fixing performance, ill need benchmarks to believe it though. Time will tell, if AMD can turn Vega into a Volta competitor with pure software then i have truly underestimated them.

Outlier...we now have two outliers that show the exact trend. 1080p wins but then drop off at higher res. So much for Vega is just another Fiji. Fiji being the exact opposite.

 

[Muhammed]

Literally no one runs PUBG on ultra. This the test PCGH did was of the "real world" type testing. That was their goal. It wasn't to sus out GPU limited data. The most common real world settings are ultra distance, high textures(or there abouts), and low everything else. Everyone I know uses though. AMD or Nvidia it doesn't matter. Also, Nvidia has CPU utilization issues now? I thought that was AMD? The point is that AMD at no point should be able to best a 1080ti if it's not possible. When the settings are cranked up towards GPU limited Vega drops off suggesting a bottleneck. You know the kind that happen when Vega features are disabled.

Lol, this is painful to watch, So now Vega is good only at low settings, 1080p and when NVIDIA IS CPU limited in one instance! But when Vega is destroyed in all other benchmarks it has to be a bottleneck! The level of contradictions in your post is alarming to the amount of mental gymnastics and straw grasping you had to do.

 

[Veradun]

Then i guess i miss your point, unless you are trying to say that AMD is intentionally gimping compute performance like Nvidia did for the titan compute drivers?

The point is: big performance boosts from drivers are possible. No. Big performance boosts through drivers are the norm: see drivers release notes.

 

[Rifter]

Lol, this is painful to watch, So now Vega is good only at low settings, 1080p and when NVIDIA IS CPU limited! But when Vega is destroyed. In all other benchmarks it has to be a bottleneck! The level of contradictions in your post is alarming to the amount of mental gymnastics and straw grasping you had to do.

It is sad to see what some of these guys are thinking, especially the guy saying he thinks 50% boost is possible a few pages back.

However all we need to do is wait to prove ourselves right, or to prove them right, its really that simple. In a short amount of time this whole issue will be put to bed definitively with real world benchmarks to back it up.

I was thinking about this last night, this thread is going off the rails. Right now current performance is dismal especially if you take efficiency into consideration, and expectations built up on the poor volta marketing blunder. There is a pile of benchmarks to support this, now, at this time.

However, there is also a real possibility that they may improve performance greatly, the fact that myself and i think most realists do not believe that doesn't make it impossible.

So personally after giving this much thought ill give it a few weeks to see what the outcome is, and we will know which camp was right in their expectations.

We can come back to this thread in a few weeks time and have our answers, one way or the other.

 

[Rifter]

The point is: big performance boosts from drivers are possible. No. Big performance boosts through drivers are the norm: see drivers release notes.

I just dont see how it compares to vega, the titan could have launched with that performance, they didnt add it with drivers as it was always there to be had. As far as i know this is not the case with vega.