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

[Malogeek]

Because the resident AMD guys here are saying Vega FE is performing poorly because January drivers, inactive features, and being read/performing like a Fiji card.

Who are you referring to? You probably are mistaken that any of the folks posting here regarding Vega are actually employed by AMD. If they were they certainly would not be making any such statements regarding Vega.

 

[Det0x]

Well now I'm even more confused. Because the resident AMD guys here are saying Vega FE is performing poorly because January drivers, inactive features, and being read/performing like a Fiji card.

But AMD is saying the drivers aren't gimped (I interpreted as features aren't disabled/missing), the drivers are old, but AdoreTV recent video shows they aren't January old, and this is not falling back to Fiji mode.

This can of worms is getting just uglier as the days go by.

I can only speak for myself, but i am just finding it very hard to belive RTG could fail so hard with a hbm2 chip measuring 484mm² (+54%)

If this really turns out to be true, then AMD is more then two whole generations behind the competition in gaming workloads, and i think some heads needs to roll.

Feels like i am forced to choose between a honest fool, and a deceiving genius =/

Oh well, 24 days left and we shall see.

 

[Snarf Snarf]

Well now I'm even more confused. Because the resident AMD guys here are saying Vega FE is performing poorly because January drivers, inactive features, and being read/performing like a Fiji card.

It's likely none of the new features are transparent to the developer minus the new cache hierarchy. This is the worst possible outcome because we know only the biggest players will bother to patch engines for Vega performance. The only hope is if Xbox One X uses enough Vega features that newer titles will have some baseline optimizations that can be built upon.

 

[Glo.]

It's likely none of the new features are transparent to the developer minus the new cache hierarchy. This is the worst possible outcome because we know only the biggest players will bother to patch engines for Vega performance. The only hope is if Xbox One X uses enough Vega features that newer titles will have some baseline optimizations that can be built upon.

I am actually laughing my ass of right now .

Vega in high level layout right now resembles very much Nvidia GPUs structure, however achieves higher throughput compared to previous versions of GCN slightly different way, than Nvidia GPU does. I know it may sound weird at first glance, however, this is best occasion for what I was planning for quite some time.

Vega's geometry, and graphics pipeline pretty much resembles right now Kepler, Maxwell, Pascal, and most likely - Volta GPU architectures. However, it is tied to Programmable Geometry pipeline, Pixel Engine and Primitive Shaders.

None of those features are present on previous versions of GCN. What this means is that if you have specific paths in games, and specific techniques of rendering which were tied with previous versions of GCN - that is what you will get - Fiji performance.

However, because you have similar features used in Vega, that are comparable to Nvidia GPUs, you do not have to do any specific thing to utilize them in your games. GCN has become pretty much Nvidia like, but has higher compute throughput, core for core, clock for clock, even higher than Pascal GP100, and Volta GPUs, which will resemble Pascal GP100 High-Level layout.

Dont worry, about Vega. It is very easy to rewrite the games to be compatible with Vega. You just add few marks in your engine for Vega architecture, and that is all.

Also, bare in mind this: https://en.wikipedia.org/wiki/Feature_levels_in_Direct3D

Tiled resources. Compare GCN, and Maxwell, and Pascal. Now Vega is Tier 3.

Vega is supposedly FL12.1 highest Tier, for everything.

 

[Bacon1]

Well now I'm even more confused. Because the resident AMD guys here are saying Vega FE is performing poorly because January drivers, inactive features, and being read/performing like a Fiji card.

He is just saying that all drivers are irritative, and not separate (which is obvious since they have a single download). What he isn't covering is why none of the new arch changes appear to be working. As shown by gamer's nexus Vega is performing exactly like Fury X when core clocks are the same for games. It is much faster in the professional tasks but performs the same for all the games tested (slightly slower with slightly less memory bandwidth). So yes, they are "Vega" drivers, because there is one set of drivers for all architectures. People are calling it "Fiji" drivers because none of the Vega features appear to be working, so it is running exactly like Fiji does.

Please explain how Vega is slower now than it was when demo'd in Doom 6 months ago if its not drivers causing a problem.

 

[Glo.]

He is just saying that all drivers are irritative, and not separate (which is obvious since they have a single download). What he isn't covering is why none of the new arch changes appear to be working. As shown by gamer's nexus Vega is performing exactly like Fury X when core clocks are the same for games. It is much faster in the professional tasks but performs the same for all the games tested (slightly slower with slightly less memory bandwidth). So yes, they are "Vega" drivers, because there is one set of drivers for all architectures. People are calling it "Fiji" drivers because none of the Vega features appear to be working, so it is running exactly like Fiji does.

Please explain how Vega is slower now than it was when demo'd in Doom 6 months ago if its not drivers causing a problem.

There is no problem with drivers.

 

[railven]

Who are you referring to? You probably are mistaken that any of the folks posting here regarding Vega are actually employed by AMD. If they were they certainly would not be making any such statements regarding Vega.

Never know. We've had AMD affiliated posters before get busted for not disclosing even though AMD requested them to do so. I wouldn't even fault people if they hid it. EDIT: Ooops, I didn't answer your question. I was referring more to the usual gungo pro AMD posters.

I can only speak for myself, but i am just finding it very hard to belive RTG could fail so hard with a hbm2 chip measuring 484mm² (+54%)

If this really turns out to be true, then AMD is more then two whole generations behind the competition in gaming workloads, and i think some heads needs to roll.

Feels like i am forced to choose between a honest fool, and a deceiving genius =/

Oh well, 24 days left and we shall see.

If I had to guess, possible AMD is just focusing on the higher margin HPC market. As long as they produce a product that is faster than their current GPUs, they can still compete in a new market tier/branch/level.

Even possible of a Vega chip slimmed down for specific gaming purpose ala GP102 vs GP100. Who knows. AMD doesn't seem to be sending clear signals for Vega FE at the moment.

It's likely none of the new features are transparent to the developer minus the new cache hierarchy. This is the worst possible outcome because we know only the biggest players will bother to patch engines for Vega performance. The only hope is if Xbox One X uses enough Vega features that newer titles will have some baseline optimizations that can be built upon.

I wouldn't wait for Xbox optimizations, personally. AMD securing the consoles hasn't seemed to alleviate all the port issues.

As for features being transparent, you'd figure AMD would want their latest GPU to be make a big splash, wouldn't you? Either they dropped the ball or as others have said this was rushed to meet promises made to investors.

Please explain how Vega is slower now than it was when demo'd in Doom 6 months ago if its not drivers causing a problem.

How would I know what AMD is doing? I know they have a PR department, surely they'd have answered this by now.

 

[tential]

I am actually laughing my ass of right now .

Vega in high level layout right now resembles very much Nvidia GPUs structure, however achieves higher throughput compared to previous versions of GCN slightly different way, than Nvidia GPU does. I know it may sound weird at first glance, however, this is best occasion for what I was planning for quite some time.

Vega's geometry, and graphics pipeline pretty much resembles right now Kepler, Maxwell, Pascal, and most likely - Volta GPU architectures. However, it is tied to Programmable Geometry pipeline, Pixel Engine and Primitive Shaders.

None of those features are present on previous versions of GCN. What this means is that if you have specific paths in games, and specific techniques of rendering which were tied with previous versions of GCN - that is what you will get - Fiji performance.

However, because you have similar features used in Vega, that are comparable to Nvidia GPUs, you do not have to do any specific thing to utilize them in your games. GCN has become pretty much Nvidia like, but has higher compute throughput, core for core, clock for clock, even higher than Pascal GP100, and Volta GPUs, which will resemble Pascal GP100 High-Level layout.

Dont worry, about Vega. It is very easy to rewrite the games to be compatible with Vega. You just add few marks in your engine for Vega architecture, and that is all.

Also, bare in mind this: https://en.wikipedia.org/wiki/Feature_levels_in_Direct3D

Tiled resources. Compare GCN, and Maxwell, and Pascal. Now Vega is Tier 3.

Vega is supposedly FL12.1 highest Tier, for everything.

This is a LOT of people's worries. This is what MANY people asked AMD numerous times.
Will these features just add performance or will it need dev support.

Given how games are benched, if Vega needs games to be rewritten, this is even more DOA than previously thought.

 

[tential]

https://www.reddit.com/r/Amd/comments/6lh1x6/bits_and_chips_says_that_vega_gaming_performance/

 

[Glo.]

This is a LOT of people's worries. This is what MANY people asked AMD numerous times.
Will these features just add performance or will it need dev support.

Given how games are benched, if Vega needs games to be rewritten, this is even more DOA than previously thought.

Guys, come on. Use your logic...

There is absolutely nothing specific in Vega architecture, that would require special optimization. Yes, Primitive Shaders, are custom, but it is actually fairly easy redevelopment of application to use this feature, and you end with...

I will not say how big improvement in performance.

Vega is DX12.1 FL Tier 3 everywhere.

 

[tential]

Guys, come on. Use your logic...

There is absolutely nothing specific in Vega architecture, that would require special optimization. Yes, Primitive Shaders, are custom, but it is actually fairly easy redevelopment of application to use this feature, and you end with...

I will not say how big improvement in performance.

Vega is DX12.1 FL Tier 3 everywhere.

That assumes the games you play still get updates...
I have no intention of playing any new games anytime soon.

Whatever is after Navi will be out by then. I'm STILL playing Twilight Princess. For the FIRST time.

Edit: Actually just go back to the starting premise.
Absolutely nothing specific in the Vega Architecture would require special optimization?
Seriously?

 

[Snarf Snarf]

You're more optimistic than I am @Glo. It's not even the degree of difficulty I worry about, it's that decisions about what projects get the most attention are made by people who will see a card with less than 5% user base and think the game performs well enough on 95% of systems. The very big devs will likely patch for it eventually, especially the ones with good relationships with AMD like DICE and Bethesda and therein lies the problem. Lots of people play games that aren't AAA titles, some of us want to play them at 1440p 144Hz or 4k 60Hz and if this card performs like a juiced Fiji that consumes 300W it might not be enough to sell to people outside the Freesync market.

 

[Bacon1]

This is a LOT of people's worries. This is what MANY people asked AMD numerous times.
Will these features just add performance or will it need dev support.

Given how games are benched, if Vega needs games to be rewritten, this is even more DOA than previously thought.

We’ll be showing Radeon RX Vega off at Computex, but it won't be on store shelves that week. We know how eager you are to get your hands on Radeon RX Vega, and we’re working extremely hard to bring you a graphics card that you’ll be incredibly proud to own. Developing products with billions of transistors and forward-thinking architecture is extremely difficult -- but extremely rewarding -- work. And some of Vega’s features, like our High Bandwidth Cache Controller, HBM2, Rapid-Packed Math, or the new geometry pipeline, have the potential to really break new ground and fundamentally improve game development. These aren’t things that can be mastered overnight. It takes time for developers to adapt and adopt new techniques that make your gaming experience better than ever. We believe those experiences are worth waiting for and shouldn’t be rushed out the door. We’re working as hard as we can to bring you Radeon RX Vega.

On HBM2, we’re effectively putting a technology that’s been limited to super expensive, out-of-reach GPUs into a consumer product. Right now only insanely priced graphics cards from our competitors that aren’t within reach of any gamer or consumer make use of it. We want to bring all of that goodness to you. And that’s not easy! It’s not like you can run down to the corner store to get HBM2. The good news is that unlike HBM1, HBM2 is offered from multiple memory vendors – including Samsung and Hynix – and production is ramping to meet the level of demand that we believe Radeon Vega products will see in the market.

https://www.reddit.com/r/Amd/commen...n_technologies_group_at_amd_and_were/dhqo7ha/

The new geometry pipeline in Vega was designed for higher throughput per clock cycle, through a combination of better load balancing between the engines and new primitive shaders for faster culling. As a programmer you shouldn't need to do anything special to take advantage of these improvements, but you're most likely to see the effects when rendering geometrically complex scenes that can really push the capabilities of the hardware.

https://www.reddit.com/r/Amd/commen...n_technologies_group_at_amd_and_were/dhqpowf/

Pretty sure he has another quote out that but couldn't figure out what word to search for. But yes, its a mixed bag. Some will require updates and some will be done by the drivers.

 

[Bacon1]

There is absolutely nothing specific in Vega architecture, that would require special optimization.

Developing drivers for new architecture is one of the most complex and difficult engineering tasks for a GPU company

https://www.reddit.com/r/Amd/commen...n_technologies_group_at_amd_and_were/dhqnpo5/

6) Many argue that vega is just a refined polaris gpu, how would you respond to this ?

My software team wishes this was true

Vega is both a new GPU architecture and also completely new SOC architecture. It's our first InfinityFabric GPU as well

https://www.reddit.com/r/Amd/commen...n_technologies_group_at_amd_and_were/dhqoev5/

Those and my previous post's quotes so how much driver work is required

 

[tential]

https://www.reddit.com/r/Amd/commen...n_technologies_group_at_amd_and_were/dhqo7ha/



https://www.reddit.com/r/Amd/commen...n_technologies_group_at_amd_and_were/dhqpowf/

Pretty sure he has another quote out that but couldn't figure out what word to search for. But yes, its a mixed bag. Some will require updates and some will be done by the drivers.

I remember the interviews and the comments.
I think it's just all unclear.
"You shouldn't need to do anything special to take advantage of these improvements"
That basically tells you nothing you realize that? That's a statement that allows you so many outs you can't even take it seriously.

We just don't know until this GPU releases, which is something that with AMD, I won't be surprised if they look forward. I don't even think it's necessarily a bad thing... I just want to know for sure at this point.

Benchmarks sell GPUs... if AMD wants to gamble that it can do better in forward benchmarks and doesn't want to focus on DX11 and backwards looking performance (which will probably be good enough for the majority of gamers), then well....

Anyway, AMD needs a press conference at this point. I wouldn't let this stew.

 

[Glo.]

That assumes the games you play still get updates...
I have no intention of playing any new games anytime soon.

Whatever is after Navi will be out by then. I'm STILL playing Twilight Princess. For the FIRST time.

Edit: Actually just go back to the starting premise.
Absolutely nothing specific in the Vega Architecture would require special optimization?
Seriously?

Nothing hardware specific. I would rather worry about Nvidia doing something that would lock the competitors from optimization, by forcing the industry to optimize for those, specific features .

No there is nothing specific in Vega that requires CUSTOM optimization. Every technique Vega uses is pretty much well known.

Well, not entirely. I think the only thing that "would" require optimization is High Bandwidth cache Controller, however it is supposed to work automatically, without interaction from developers.

I remember the interviews and the comments.
I think it's just all unclear.
"You shouldn't need to do anything special to take advantage of these improvements"
That basically tells you nothing you realize that? That's a statement that allows you so many outs you can't even take it seriously.

We just don't know until this GPU releases, which is something that with AMD, I won't be surprised if they look forward. I don't even think it's necessarily a bad thing... I just want to know for sure at this point.

This is what Raja is basically saying:

Our Vega architecture, has well known techniques, and you do not have to do anything out of standard code to make it work, and optimize for this architecture.

You guys do not know how game development works, and how GPUs, do you?

You still have to redesign the pipeline, for this GCN version, but it is not different than just optimizing it like for Nvidia GK/GM/GP architectures. With, obviously, knowing the boarders, and boundaries of the architecture.

 

[96Firebird]

https://www.reddit.com/r/Amd/comments/6lh1x6/bits_and_chips_says_that_vega_gaming_performance/

They also said Vega was faster than Pascal...

https://twitter.com/BitsAndChipsEng/status/863329100492132352

Oh, right, we just have to "wait and see"... Par for the course.

 

[Bacon1]

Anyway, AMD needs a press conference at this point. I wouldn't let this stew.

Why? They've already said wait for Vega RX for gaming. It comes out the 30th...

 

[Bacon1]

How would I know what AMD is doing?

I'm asking you to make an informed guess. I'm not asking for proof or a fact, just your guess as to why Vega is slower now than 6 months ago.

 

[tential]

I'm asking you to make an informed guess. I'm not asking for proof or a fact, just your guess as to why Vega is slower now than 6 months ago.

You can make informed guesses about this, but for RX Gaming, you have to wait for RX Vega?

 

[Maverick177]

They also said Vega was faster than Pascal...

https://twitter.com/BitsAndChipsEng/status/863329100492132352

Oh, right, we just have to "wait and see"... Par for the course.

I'm not sure how the 2 statements contradict each other?

 

[piesquared]

It's probably important to remember what Raja said in the video interview with PCWorld from January. That they are going to be positioning their product stack in terms of bandwidth, and are going to be working towards that end as an industry wide performance measuring indicator.
Nick said in the recent interview with PCWorld that the higher performing Radeon Pro line would be revealed soon (probably Sigraph), so those will probably be the 4 HBM2 stacks with closer to 1TB/s bandwidth. Perhaps there is an RX Vega in the pipeline with 4 stacks as well.

 

[darkswordsman17]

They also said Vega was faster than Pascal...

https://twitter.com/BitsAndChipsEng/status/863329100492132352

Oh, right, we just have to "wait and see"... Par for the course.

Not sure if serious...

Who am I kidding, I know you're serious, but I have to ask if you actually listen to yourself? I know this isn't the first time I've had to point out how blatantly obvious the statement you post completely abolishes the point you're trying to infer.

You're literally scoffing that it is par for the course for us to wait and see what the performance of an unreleased product is until the product actually releases?

 

[tential]

It's probably important to remember what Raja said in the video interview with PCWorld from January. That they are going to be positioning their product stack in terms of bandwidth, and are going to be working towards that end as an industry wide performance measuring indicator.
Nick said in the recent interview with PCWorld that the higher performing Radeon Pro line would be revealed soon (probably Sigraph), so those will probably be the 4 HBM2 stacks with closer to 1TB/s bandwidth. Perhaps there is an RX Vega in the pipeline with 4 stacks as well.

Like I said, there are a lot of comments from Raja regarding Vega that were about pushing the industry to somewhere rather than delivering performance to current applications.

 

[piesquared]

Like I said, there are a lot of comments from Raja regarding Vega that were about pushing the industry to somewhere rather than delivering performance to current applications.

Yes, i realized that by repeating that part of Raja's statement, it would be low hanging fruit for some to completely gloss over the important part. So what about the higher bandwidth/ higher performance part of his statement (the part of the comment you apparently somehow completely missed)?