Support NeoGAF

[llien]

Vega or not, but what was AMD planning to do vs low end Volta?

It's so complex that it's a contributing reason why Intel gave up on creating a high-end GPU with a revolutionary architecture (read up on Larrabee if you are unfamiliar with it).

Larrabee was a bunch of x86 cores, because, I'd say, Intel felt so much more comfortable with x86 cores... as hardware, foremost.
That idea (hardware, foremost) didn't quite work and if someone tells me it's because of drivers, I'll simply smile.

Seriously, when I say that modern high-performance PC graphics drivers (which need to support everything from OpenGL 1 and DX8 to Vulkan and DX12) are some of the most complex software projects on the planet, what I mean is that they are some of the most complex software projects on the planet.

You don't need to support everything in the same codepath or develop it as a single piece. It is more work, but it isn't more complex. Nothing stops you from using independent project A for Vulkan, B for OpenGL, C for DX<insert version>

The same applies to per game optimizations, which is likely done by rather small teams, well, in parallel. More work? Yep. More complex, because there are many games? Nope.

I simply don't see anything that would challenge monster like Intel.

Let's agree to disagree on this.

 

[dogen]

Larrabee was a bunch of x86 cores, because, I'd say, Intel felt so much more comfortable with x86 cores... as hardware, foremost.
That idea (hardware, foremost) didn't quite work and if someone tells me it's because of drivers, I'll simply smile.

Apparently it actually worked pretty well.

https://tomforsyth1000.github.io/blog.wiki.html

 

[dr_rus]

Vega or not, but what was AMD planning to do vs low end Volta?

You mean, you don't like the newly released mighty RX500 series?

 

[ZOONAMI]

You mean, you don't like the newly released mighty RX500 series?

They will probably just add more CUs/shaders to a Polaris foundation. 40-50 CUs, like what's going in Scorpio, and then clocked higher than Scorpio. That should put them in range with a 1160 etc. 580 was just a rebrand because it's still competing with a 1060 for a good while. 670/680 will be beefed up a bit I imagine.

 

[Locuza]

Nah, Polaris is done, after P12 there will only be Vega.

 

[dr_rus]

They will probably just add more CUs/shaders to a Polaris foundation. 40-50 CUs, like what's going in Scorpio, and then clocked higher than Scorpio. That should put them in range with a 1160 etc. 580 was just a rebrand because it's still competing with a 1060 for a good while. 670/680 will be beefed up a bit I imagine.

Vega 11 should come out somewhere around the launch of middle/low end of Volta. This is what will compete with Volta's midrange and as for low end - who cares? Polaris can't really compete with Pascal there right now but it doesn't stop anyone.

 

[llien]

Dear god, 470 going for $350+ on ebay, now I've seen it all...

Vega 11 should come out somewhere around the launch of middle/low end of Volta. This is what will compete with Volta's midrange and as for low end - who cares? Polaris can't really compete with Pascal there right now but it doesn't stop anyone.

With HBM?

 

[RaijinFY]

Dear god, 470 going for $350+ on ebay, now I've seen it all...




With HBM?

HBM won't do jack shit to help AMD's inefficent µarch.

 

[spyshagg]

HBM won't do jack shit to help AMD's inefficent µarch.

inefficient doing what? gaming? just because it consumes 35% more than nvidia counterparts?

That same inefficient µarch you talk about is currently sold out for months to come because guess what: its more efficient than nvidia at brute forcing digital coins.

I'm not calling you out. You just lack some perspective and should learn a bit more about the things you seem certain to talk about. And yes, nvidia deserves all the props for making their strategy work (stripping their gpu and moving those functions into the cpu). Its great for games no doubt.

 

[ZOONAMI]

Vega 11 should come out somewhere around the launch of middle/low end of Volta. This is what will compete with Volta's midrange and as for low end - who cares? Polaris can't really compete with Pascal there right now but it doesn't stop anyone.

So a 570/580 doesnt compete with 1050/1060. 470 is better than a 1050 or 1050ti, 580 and 1060 are pretty much equal cards.

 

[ethomaz]

Wrong quote... sorry.

 

[dr_rus]

With HBM?

Vega seems to be HBM top to bottom (if you can say that about two chips lineup) so likely yeah. My bet is one 4/8 GB stack for Vega 11.

So a 570/580 doesnt compete with 1050/1060. 470 is better than a 1050 or 1050ti, 580 and 1060 are pretty much equal cards.

What? 580 competes directly with 1060, 570 4GB - with 1060 3GB, 470 is EOL and below that it's almost total NV domination on price/perf or price/watt or perf/watt - however you wanna put it.

https://www.computerbase.de/2017-06/radeon-rx-560-test/2/

Price difference in low end is so small that it doesn't really matter that 560 is $10 less than 1050 to most of people.

 

[ZOONAMI]

I'm contemplating going from a 1070 to a 1080 because I can get a really good deal on one. Would only be about a $100 upgrade after selling the 1070. I honestly don't know that it's worth waiting for Vega.

 

[opticalmace]

I'm contemplating going from a 1070 to a 1080 because I can get a really good deal on one. Would only be about a $100 upgrade after selling the 1070. I honestly don't know that it's worth waiting for Vega.

I dunno if I'd bother. Depends on the specifics. What card is your 1070 and what card is the 1080? Do you already have a buyer for your 1070 or are you gonna have to spend time trying to sell it? How much are you selling it for and how much does he want for the 1080?

Unless you're going from like a reference 1070 and it's a nice AIB 1080 I'd probably not bother. But YMMV.

 

[Unknown Soldier]

I'm staying green no matter what since I have a G-sync monitor, but it would be nice to have some competition to put pressure on pricing.

GPU development is more expensive than ever. Nvidia claims they spent $3 billion on development of Volta, that money needs to be recouped somehow.

On the low and midrange, Nvidia's GPU pricing has roughly followed inflation but otherwise has not increased. However on the top end where AMD is not competitive, Nvidia has been increasing prices for the very small number of enthusiasts to help pay for new development. The new development disproportionately benefits enthusiasts so asking them to pay more is not an unfair proposition.

 

[ZOONAMI]

I dunno if I'd bother. Depends on the specifics. What card is your 1070 and what card is the 1080? Do you already have a buyer for your 1070 or are you gonna have to spend time trying to sell it? How much are you selling it for and how much does he want for the 1080?

Unless you're going from like a reference 1070 and it's a nice AIB 1080 I'd probably not bother. But YMMV.

It is an FE, but it holds over 2000mhz well and 9200 memory at load and temps low 70s with fan maxed at 75% (which really isn't very loud imo, isn't much louder than my 980 ti G1 was on my other rig that I sold. Most framerate vids I see it is matching a stock 1080, so yeah I don't know that it's worth bothering either. Just kind of crazy that 1080s can now be had for same price as launch 1070 FE price.

 

[Durante]

Let's agree to disagree on this.

Ok, but let me leave you with one question: have you actually ever worked on graphics drivers, or at least with graphics drivers, or at the very least o a low-level optimized software product with more than a million lines of code?

Because "you can just develop it independently with no increase in complexity" doesn't sound like it.

 

[Locuza]

Apple announced the new iMac Pro at the WWDC2017 with Vega 10 and 11/22 TF FP32/FP16 performance combined with 16GB HBM2 memory for late 2017:
http://www.anandtech.com/show/11515/the-apple-wwdc-2017-live-blog

 

[Delusibeta]

Apple announced the new iMac Pro at the WWDC2017 with Vega 10 and 11/22 TF FP32/FP16 performance combined with 16GB HBM2 memory for late 2017:
http://www.anandtech.com/show/11515/the-apple-wwdc-2017-live-blog

Also announced that a RX 555 and RX 575 will be in their Macbooks, whatever that means. Something of a surprise, considering Nvidia did manage to get Mac drivers officially released for the Titan Xp.

Worth noting that the quoted flops for the iMac Pros are lower than the known flop rates for the Frontier Edition cards, indicating that these would be cut down (and probably Mac exclusive) variants.

 

[Locuza]

Putting pieces together from Apple's english and german website, there will be two options available for the graphics processor.

One has 64 CUs (4096 ALUs) and 16GB HBM2, the other 56 CUs (3584 ALUs) and only 8GB.
Apple says the bandwidth is up to 400 GB/s.

Radeon Pro Vega 56 Grafikprozessor mit 8 GB HBM2 Grafikspeicher
Optional mit Radeon Pro Vega 64 Grafikprozessor mit 16 GB HBM2 Grafikspeicher

https://www.apple.com/de/imac-pro/specs/

On-package HBM2 replaces external VRAM, so the GPU can fetch data at up to 400GB/s.

https://www.apple.com/imac-pro/

 

[icecold1983]

Ok, but let me leave you with one question: have you actually ever worked on graphics drivers, or at least with graphics drivers, or at the very least o a low-level optimized software product with more than a million lines of code?

Because "you can just develop it independently with no increase in complexity" doesn't sound like it.

i remember some time in the past, nvidia stating their driver has more lines of code than windows. i might be wrong on it being windows, but its an enormous codebase.

 

[BlizzyAzz]

http://www.pcgameshardware.de/Vega-...ntationsfolie-schneller-als-Titan-Xp-1229586/

 

[icecold1983]

http://www.pcgameshardware.de/Vega-...ntationsfolie-schneller-als-Titan-Xp-1229586/

thats very impressive if legit. seems to good to be true tho. also, since when is a titan xp almost 2x as fast as a 1080

 

[DouglasteR]

http://www.pcgameshardware.de/Vega-...ntationsfolie-schneller-als-Titan-Xp-1229586/

Well, its too good to be true honestly.

 

[Dictator93]

The one thing to note about Vulkan Doom is that it uses some 'GPU intristics' vulkan extensions that target AMD hardware (not just async compute) - this, without using the extensions for NV in spite of similar extensions being available for NV hardware. They were made available at a later date and Doom never updated sadly enough.

So a comparison of Vulkan Doom cross-vendor and GPU classes will be a bit disproportionate since the Vulkan code is lower level on GCN than it would be on Maxwell or Pascal. But still, saweet performance at 4K if it is real (who knows).

 

[pulsemyne]

That picture says the NDA ended on May the 8th so we should have seen that picture ages ago.

 

[Captain Chaos]

Well, its too good to be true honestly.

Crossfired Vega?

 

[mephixto]

Crossfired Vega?

Maybe dual GPU?

 

[dogen]

i remember some time in the past, nvidia stating their driver has more lines of code than windows. i might be wrong on it being windows, but its an enormous codebase.

yep lol

https://forums.geforce.com/default/topic/849203/announcing-geforce-hotfix-driver-353-38/

A GeForce driver is an incredibly complex piece of software, with more lines of code than the entire Windows OS. We have an army of software engineers constantly adding features and fixing bugs. These changes are checked into the main driver branches, which are eventually run through a massive QA process and released.

 

[Newboi]

The one thing to note about Vulkan Doom is that it uses some 'GPU intristics' vulkan extensions that target AMD hardware (not just async compute) - this, without using the extensions for NV in spite of similar extensions being available for NV hardware. They were made available at a later date and Doom never updated sadly enough.

So a comparison of Vulkan Doom cross-vendor and GPU classes will be a bit disproportionate since the Vulkan code is lower level on GCN than it would be on Maxwell or Pascal. But still, saweet performance at 4K if it is real (who knows).

So is this why running Doom with the Vulkan renderer on Nvidia GPUs is still a stuttery/artifact prone experience? I have a 1080 and the Vulkan mode still runs really unevenly despite reporting higher framerates than the OpenGL renderer.

Anyway, I honestly wonder if Vega will be a shortlived chip? I'd be surprised if AMD will again leave the high-end of the consumer market with no competition for Nvidia again when Volta high-end chips launch.

 

[napata]

Putting pieces together from Apple's english and german website, there will be two options available for the graphics processor.

One has 64 CUs (4096 ALUs) and 16GB HBM2, the other 56 CUs (3584 ALUs) and only 8GB.
Apple says the bandwidth is up to 400 GB/s.


https://www.apple.com/de/imac-pro/specs/


https://www.apple.com/imac-pro/

It's funny we learn more from Apple about Vega than we learn from AMD.

 

[icecold1983]

So is this why running Doom with the Vulkan renderer on Nvidia GPUs is still a stuttery/artifact prone experience? I have a 1080 and the Vulkan mode still runs really unevenly despite reporting higher framerates than the OpenGL renderer.

Anyway, I honestly wonder if Vega will be a shortlived chip? I'd be surprised if AMD will again leave the high-end of the consumer market with no competition for Nvidia again when Volta high-end chips launch.

Im not aware of their being widespread issues at this point with doom vulkan on nvidia. Theres just no performance improvement when gpu limited

 

[Maybesew]

Also announced that a RX 555 and RX 575 will be in their Macbooks, whatever that means. Something of a surprise, considering Nvidia did manage to get Mac drivers officially released for the Titan Xp.

It's not really a surprise. Apple has exclusively used AMD cards for several years.

 

[tuxfool]

Im not aware of their being widespread issues at this point with doom vulkan on nvidia. Theres just no performance improvement when gpu limited

Performance in terms of frame rates is fine and it does offer a boost over OGL. Frame pacing is atrocious, unplayably so.

 

[PFD]

http://www.pcgameshardware.de/Vega-...ntationsfolie-schneller-als-Titan-Xp-1229586/

lol I don't believe it

 

[Irobot82]

lol I don't believe it

I don't think the Titan Xp is twice as fast as the 1080

 

[Delusibeta]

I don't think the Titan Xp is twice as fast as the 1080

It's not.

Assuming stock speeds, it's 94.2 FPS average for the Titan Xp versus 67.6 FPS average for the 1080. Even overclocked, the Titan Xp can only do 108 FPS. That Vega chart is baloney.

Worth noting that this benchmark is with zero AA, while the Vega benchmark is claiming 8x TSAA, which would definitely be negative on frame rates. So, yeah, definitely garbage.

 

[Seronei]

It's not.

Assuming stock speeds, it's 94.2 FPS average for the Titan Xp versus 67.6 FPS average for the 1080. Even overclocked, the Titan Xp can only do 108 FPS. That Vega chart is baloney.

Worth noting that this benchmark is with zero AA, while the Vega benchmark is claiming 8x TSAA, which would definitely be negative on frame rates. So, yeah, definitely garbage.

We don't know where they tested it so maybe they found some place in DOOM that were significantly less demanding.

Or the more likely alternative that it's fake. Especially since the NDA is supposedly up on it.

 

[LCGeek]

So is this why running Doom with the Vulkan renderer on Nvidia GPUs is still a stuttery/artifact prone experience? I have a 1080 and the Vulkan mode still runs really unevenly despite reporting higher framerates than the OpenGL renderer.

Doom 2016 isn't stuttery. Its the only game on any system I have now that can run the renderer at 500us which is half a millisecond. Most stuttering in windows is due to windows itself, windows being misconfigured, and never disabling shit certain commands or features which cause stuttering globally. I say all that pushing it on a 970 vs the 1060s I have as well.

I say the same at times in the Forza Horizon thread since they fixed it game post launch, doesn't run as low as doom but neither do most other games.

 

[icecold1983]

Performance in terms of frame rates is fine and it does offer a boost over OGL. Frame pacing is atrocious, unplayably so.

Im going to try it again when i get home

 

[thelastword]

Vega seems to be HBM top to bottom (if you can say that about two chips lineup) so likely yeah. My bet is one 4/8 GB stack for Vega 11.


What? 580 competes directly with 1060, 570 4GB - with 1060 3GB, 470 is EOL and below that it's almost total NV domination on price/perf or price/watt or perf/watt - however you wanna put it.

https://www.computerbase.de/2017-06/radeon-rx-560-test/2/

Price difference in low end is so small that it doesn't really matter that 560 is $10 less than 1050 to most of people.

Yes 580 competes with the 1060 6GB, but then both versions do, the 4Gb 580 and the 8GB beats the 1060 6GB in lots of games...funny you have both variations of the 1060, that is 6Gb and 3Gb, yet I'm only seeing the 8GB version of the 580....

The same argument applies to the 570, where is the 8GB version of the 570....The thing about the 570 is that it beats the 1050Ti, the 960 and the 970, so the 570 covers and outperforms a number of NV's cards.....

We then go to the 560, which destroys the 750ti and of course the 550 outperforms the 1030...So I really don't see where NV has the lead in performance here....


Yes, NV has no competition on the higher end, which will soon end, but mid range, it's clear that AMD has the better options...

580 8GB -> 1060 6GB (edges it because of higher vram and better dx12 and vulkan performance) more games coming that will utilize amd hardware.

580 4GB -> 1060 4Gb for the same reasons mentioned above.

570 8GB and 4GB -> 1050Ti, 970, 1050,960

560.-> 750Ti

550 -> 1030

That picture says the NDA ended on May the 8th so we should have seen that picture ages ago.

So it's very likely that performance is even better now through better drivers and better clockspeeds...AMD have delayed Vega for a reason, I imagine it's for them to establish a big enough advantage on the high end. I'm of the opinion that they have pushed Vega much further from where it was in Jan-Mar of 2017......

 

[napata]

...

The problem with AMD cards is that they're only really good for AAA multiplats. When you move to games with a lower budget, especially ones that aren't mulitplat, you see a 1060 match Fury cards. AMD cards just tank in these sorts of games. Developers just don't have the time/money to optimize for AMD. Just take PUBG for example.

I think Nvidia is a company with very shitty business practices and I'd prefer to support AMD but I play mainly PC games where AMD is a terrible choice so I'm stuck with Nvidia.

Chart's already been debunked, in case you haven't read this thread.

Even before the miners cleared out AMD's supply of graphics card, the 4Gb 580 and 8Gb 570 was in so limited supply that they might as well have not existed.

That's also true. 580 vs 1060 doesn't even exist right now because you can't even buy a high end Polaris card. Good for AMD though in the short term.

 

[Delusibeta]

So it's very likely that performance is even better now through better drivers and better clockspeeds...AMD have delayed Vega for a reason, I imagine it's for them to establish a big enough advantage on the high end. I'm of the opinion that they have pushed Vega much further from where it was in Jan-Mar of 2017......

Chart's already been debunked, in case you haven't read this thread.

Even before the miners cleared out AMD's supply of graphics card, the 4Gb 580 and 8Gb 570 was in so limited supply that they might as well have not existed.

 

[dr_rus]

http://www.pcgameshardware.de/Vega-...ntationsfolie-schneller-als-Titan-Xp-1229586/

These are all kinds of wrong for NV cards.
There are no benchmarks of Doom VK/4K where 1080 would be 47% slower than 1080Ti / Titan Xp.

26% slower

24% slower

28% slower

29% slower

29% slower

etc

 

[opticalmace]

^ I agree with you, but just fyi you're reporting percentage points, not percentages, as it's written now.

 

[dr_rus]

^ I agree with you, but just fyi you're reporting percentage points, not percentages, as it's written now.

What's percentages then? I'm reporting how slower a 1080 is relative to 1080Ti which means that I'm using 1080Ti as a 100% here.

 

[Newboi]

Doom 2016 isn't stuttery. Its the only game on any system I have now that can run the renderer at 500us which is half a millisecond. Most stuttering in windows is due to windows itself, windows being misconfigured, and never disabling shit certain commands or features which cause stuttering globally. I say all that pushing it on a 970 vs the 1060s I have as well.

I don't know how to argue with you. It is for me. Telling me it isn't doesn't change that.

By the way, this is only in Vulkan mode for me. The game has weird artifacts at times and has odd stuttering (almost like highly uneven framepacing). The framerates are sky high though. OpenGL mode runs buttery smooth for me on my 1080 at 4K.

 

[Renekton]

Given Fury's mediocre hashrate to price, I'm thinking miners won't be as interested in Vega?

 

[tuxfool]

Die-shot of Vega:

taken from here https://videocardz.com/newz/amd-releases-vega-gpu-die-shot

Still with two stacks. How they managed to get 16GB with two stacks remains a mystery. Maybe this is an 8 die stack?

 

[Buggy Loop]

^^^

Guess i was late..

It's sexy af