AMD Vega Discussion Thread

[FordGT90Concept]

I haven't looked into the die size.
That is pretty impressive on NVIDIA side twice the performance for 50% larger die size on 16nm vs 14nm for AMD side.
Looking at it like that Vega better be a huge step up for AMD.
My guess is Vega will match gtx 1080 for less they not gonna be able to touch 1080ti unless they pull a rabbit of their hat.
As long is not a power hungry monster and has good value I'm good.

Kepler was shit yet Maxwell was great. What was the difference? Tiled rasterization. What does Vega add that Polaris doesn't have? Tiled rasterization. Why does it matter? Tile-based rasterization can keep the data inside of the GPU's cache structure (e.g. Vega's fancy new L2).

More info:

Vega could be amazing but it also might not be. Time will tell.

 

[hapkiman]

Kepler was shit yet Maxwell was great. What was the difference? Tiled rasterization. What does Vega add that Polaris doesn't have? Tiled rasterization.

More info:

Yes Kepler was little more than optimized Fermi, but shit it was not. I remember when the GTX 680 first came out, and it was freaking awesome for the time.

 

[FordGT90Concept]

GTX 680 only beat the 7970 in games that used GameWorks and even then, by the slimest margins at the highest resolutions. Source.

AMD responded a few months later with the 7970 GHz, the places where GTX 680 managed to pull ahead were again erased. Source.

GCN continued to shame NVIDIA until Maxwell.

 

[the54thvoid]

GTX 680 only beat the 7970 in games that used GameWorks and even then, by the slimest margins at the highest resolutions. Source.

AMD responded a few months later with the 7970 GHz, the places where GTX 680 managed to pull ahead were again erased. Source.

GCN continued to shame NVIDIA until Maxwell.

GK104 (GTX 680) was released exactly because the 7970 was not as powerful as expected. The 680 was a mid range chip product. Yes the 7970Ghz beat it but it was at a price premium as well. I know - I bought two 7970Ghz cards and both were more expensive than the GTX680. The irony was that despite pulling an awesome (arguably one of their best 'recent') cards out of the hat, it was beaten initially by what was not designed to be Nvidia's top card. You say Kepler was shit but it was the back bone of the GTX780/GTX Titan and GTX 780ti, which all performed well at the time.

I think you're being slightly disingenuous when you talk about the graphics past. People were up in arms that Nvidia didn't initially bother to 'smack' down the 7970. It was almost a scandal that Nvidia priced what was a mid range chip at top end price because it matched the best AMD could throw at it. It was also the crappy crossfire performance back then in dx9 that made me swap out the two 7970's for a single GTX Titan.

Don't sugar coat history, it's still there to be read.

http://www.techradar.com/news/compu...-nvidia-s-geforce-gtx-680-an-imposter-1074235

 

[FordGT90Concept]

R9 2## was AMD's response to GTX 7##.

R9 290X cards are still a great card today. GTX 780 Ti, not so much.

Maxwell was an earthquake. AMD had to go full retard with Fiji to make a card that could climb on top only to be shamed again by Pascal.


NVIDIA had really crappy yields with Kepler which is why they priced GTX 680 at a premium. By GTX 780, they fixed the yield problems.

 

[Solaris17]

how are they getting those HBM2 numbers?

didnt HBM have a 4GB limit? can HBM2 now get passed that?

 

[FordGT90Concept]

HBM was limited to 1 GiB per stack and you can only reasonably fit 4 stacks on the interposer. HBM2 is up to 8 GiB per stack. The Vega cards that have been demo'd have two stacks of 4 GiB each (8 GiB). MI25 likely has four stacks of 4 GiB each (16 GiB). 8 GiB stacks seem unlikely due to the height of the chip itself although it is not impossible (using pliable TIM) that we could see 32 GiB chips.

HBM3 is apparently coming in 2020 with even higher densities.

 

[Kanan]

Because it upsets the zealots. It means they can't gloat about how superior they are. I'm obviously not serious about wanting it to be a train wreck, I just wish they'd released it already. I've gambled on it being not as fast as the 1080ti. If it is, then my gamble has not paid off and I'll be like, 'meh, win some lose some'. I can see it being £599 if it is as good as the 1080ti though so it's not going to be a cheap card by any means. It might even walk up to the 1080ti price point.

Even if Vega is faster than 1080 Ti, what I'd call unlikely too, you're still good, GeForce has some great features Radeon still hasn't. Performance isn't everything.

GTX 680 only beat the 7970 in games that used GameWorks and even then, by the slimest margins at the highest resolutions. Source.

AMD responded a few months later with the 7970 GHz, the places where GTX 680 managed to pull ahead were again erased. Source.

GCN continued to shame NVIDIA until Maxwell.

GCN fanboy huh? Not really true what you say. First of all the GTX 680 was a much smaller chip and had way lower power consumption too, despite that it still beat the HD 7970. Win-win for Nvidia. Go and read up old reviews too if you want.

R9 290X cards are still a great card today. GTX 780 Ti, not so much.
NVIDIA had really crappy yields with Kepler which is why they priced GTX 680 at a premium. By GTX 780, they fixed the yield problems.

780 Ti is still nice, I would know. R9 290X on the other hand is hot and a power eater, I wouldn't want to swap my GPU with that one, never ever. I enjoy my low 15 W power consumption on multi monitor every day, something that's not possible with almost any Radeon card, a major flaw still. Even the Bluray / Web youtube power consumption is bad, and I don't like that either. I want a PC that runs quiet and has low power consumption when I'm not using it much, and exactly that ca. prohibits me from most Radeon cards - only Fury/Nano is kinda okay in that regard, but it's only solved via HBM, because HBM has low power consumption and not because AMD fixed it. Vega will not have the issue either (high memory clocks = high power consumption) because of HBM2, but I hope they fix the problems for the regular GPUs with GDDR5 as well, it's just bad.

Yield problems with GK104? I'd call that rumour, or give me some evidence. I never heard of such problems and seing how well GTX 680 sold and never really was out of stock, I'd simply say it's not true.

 

[FordGT90Concept]

http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review

NVIDIA has been fairly forthcoming to their investors about how 28nm production is going, and while yields have been acceptable TSMC doesn’t have enough wafers to satisfy all of their customers at once, so NVIDIA is still getting fewer wafers than they’d like. Until very recently AMD’s partners have had a difficult time keeping the 7970 in stock, and it’s likely it will be the same story for NVIDIA’s partners.

 

[cdawall]

So on a different note does anyone have any rumors pointing towards pcb length on these? The 1080/1080Ti is too long for my current case setup.

 

[TheGuruStud]

So on a different note does anyone have any rumors pointing towards pcb length on these? The 1080/1080Ti is too long for my current case setup.

Wouldn't Fury X be the max size?

Intended as a hype train or not, I can't get into anything AMD regarding GPUs lately, because their drivers seem to be getting worse and worse. At first I thought it was the old (7970) model I'm using. Then I saw on an AMD forum thread, that many have been reporting Crimson scaling not working, Radeon settings not installing, ReLive not working properly, etc, etc, even on much more current models. And the idiot Ray moderating the forum labeled the thread as solved, even though many are still reporting the problems.

Whatever. I had 4890s, 6950s, 7970s and a 290x....I had so many problems...wait, no, I didn't. They were all game issues that were experienced on Nvidia, too.

And then we have fruitcake Mr. 2x470 boy. "Hurr durr, I ignored that it's a complete redesign lolz, let me smoke more crack"

 

[cdawall]

Wouldn't Fury X be the max size?

I am hoping that is the length, but the pics shown didn't really show the pcb and they had a standard cooler.

 

[Vayra86]

Yes Kepler was little more than optimized Fermi, but shit it was not. I remember when the GTX 680 first came out, and it was freaking awesome for the time.

Exactly my thoughts, and everyone jumped on the 670 instead and clocked it to 680 stock performance for a lot less money.

7970 was strong as well, but 680/670 was Nvidia's money maker and they sold shitloads.

Now the 780... was topped by the 780ti and it performed as good as Hawaii but without being an oven. The 290 was a popular card for AMD but they sold most of those in the Bitcoin mining craze, and shortly after that the internet was overran by second hand 290's with stock coolers and Sapphire Vapor X's.

Within the 'Kepler' era the only really strong GCN release was the 7970 and its smaller brother 7950 for a cheap alternative on that. Below that, the 7870 and the XT of that card sold very well and probably did better than Nvidia's midrange at the time, also performing better than it at similar price. But 290(x) was a failure, plagued by launch issues, high temps, and the only card anyone could really recommend was Sapphire's Tri-X version. All the others were airplanes.

And below Hawaii we had only rebrands... most of which didn't improve on their originals either, while Nvidia launched its 770 with higher clocked memory.

 

[FordGT90Concept]

285 was not a rebrand. It was GCN 1.2.

 

[Frag_Maniac]

Whatever. I had 4890s, 6950s, 7970s and a 290x....I had so many problems...wait, no, I didn't. They were all game issues that were experienced on Nvidia, too.

Let me clarify that I'm not just talking about the display driver itself. Granted there are a few games that are more likely to demonstrate artifacts when you're having issues with the driver installer (which happens WAY more often on AMD, especially since Crimson and ReLive), but there are FAR more things problematic with AMD's software packages than the display driver.

No, I'm talking about the tons of complaints I and many others have voiced while AMD lives in denial land, about ongoing problems with features not installing or working in their latest driver packages. Whether it's scaling not being functional, or additional Radeon settings not even installing, or other such problems, they never seem to have wide spread compatibility with their software package.

I never had these problems on my Nvidia GPUs. You would think that the fact that ever since Angel Trinidad (Omega Man) having specifically written modified drivers mostly for ATI, people wouldn't be in denial about ATI/AMD bearing the brunt of software issues in the GPU industry. Angel himself quite often said the main reasons their drivers needed modifying was due to horrible installers, and unnecessary bloat.

Some things never seem to change. Even with the new Ryzen CPUs, once again, whether it be AMD drivers, W10 updates, or AM4 BIOS issues, or RAM frequency glitches, these chips launched with crippled performance because of it. Does this happen with Intel, well, once in a while, but not like clockwork with every new architecture like with AMD. I mean, can AMD fanatics not see the writing on the wall by now?

 

[xkm1948]

Once again it is from Chiphell. And this is from the same guy who leaked RyZen 1800X running at 3.6GHz around January. He also leaked that Polaris based GPU will be rebranded for mainstream/low end this year.

To summarize it up:

There will be two VEGA: VEGA10 and VEGA11.
1. Vega11 will use GDDR5, this will be the first one to come out.

2. Vega10, aka BigVega, will use HBM2. It is VERY LIKELY to be delayed. As production has only just began for Vega10

3. And performance. Don't even hope for 1080Ti level performance. So somewhere between 1080Ti and 1080. Really disappointing. AMD is doing all they can to "optimize" the driver. But falling short of 1080Ti is pretty much set.


In short. AMD f*ucked up big time again at high end this year.

source:

https://chiphell.com/thread-1729030-1-1.html

 

[P4-630]

AMD hasn't even an answer to the GTX1070....
Let alone an answer to a 1080Ti...

 

[Vya Domus]

They didn't fuck up , at least I don't think so. A lot of people are missing out the fact that they already held a major release with Ryzen , AMD doesn't have as much resources as Nvidia and obviously no where near what Intel has. As a result you can't expect them to deliver big products one after the other with stellar attributes.

If there is one thing that I find disappointing is that they do not use HBM for Vega 10.

3. And performance. Don't even hope for 1080Ti level performance. So somewhere between 1080Ti and 1080. Really disappointing. AMD is doing all they can to "optimize" the driver. But falling short of 1080Ti is pretty much set..

Why would they try to beat the 1080ti in that 700$ price bracket which accounts for a small portion of what people buy ? I wouldn't see that as a smart move. Performance between 1080 and 1080ti is great for 500$ or under.

 

[PerfectWave]

so why they cant use gddr5x? so vegao 10 will be between 1080 and 1080ti.. but what about vega 11?

 

[Aenra]

One question and keeping in mind this is all just rumormill.

When you say performance between 1080 and Ti, is that for 11, for 10, or for both?

(and speaking of which, how come the big boy is numbered 10 and the little one 11? Counter-intuitive, isn't it?)

 

[the54thvoid]

Performance between 1080 and 1080ti isn't fail.

But if the OP is correct about a delay? Even though the AMD Twitter thing said this quarter?

Too much rumour. Not enough fact. FWIW, Ryzen is pretty neat - I'm liking mine.

 

[Xzibit]

Once again it is from Chiphell. And this is from the same guy who leaked RyZen 1800X running at 3.6GHz around January. He also leaked that Polaris based GPU will be rebranded for mainstream/low end this year.

To summarize it up:

There will be two VEGA: VEGA10 and VEGA11.
1. Vega11 will use GDDR5, this will be the first one to come out.

2. Vega10, aka BigVega, will use HBM2. It is VERY LIKELY to be delayed. As production has only just began for Vega10

3. And performance. Don't even hope for 1080Ti level performance. So somewhere between 1080Ti and 1080. Really disappointing. AMD is doing all they can to "optimize" the driver. But falling short of 1080Ti is pretty much set.


In short. AMD f*ucked up big time again at high end this year.

source:

https://chiphell.com/thread-1729030-1-1.html

If you translate his post

Vega11 ok gddr5
vega10 great chance will delay, because this month only try to produce
computex will have the exact message
performance is still in the adjustment 1080ti by the original is not to think

So your speculating on his speculation

Performance between 1080 and 1080ti isn't fail.

But if the OP is correct about a delay? Even though the AMD Twitter thing said this quarter?

Too much rumour. Not enough fact. FWIW, Ryzen is pretty neat - I'm liking mine.

The graph they put up is fake even the OP on Chiphell pointed it out

Too much rumors and drum beating



PC Games News Reported this today

Update, April 25, 2017: Two months. That's the most you're going to have to wait because AMD have confirmed that the Radeon RX Vega cards are still on track for a release this quarter.

 

[xkm1948]

One question and keeping in mind this is all just rumormill.

When you say performance between 1080 and Ti, is that for 11, for 10, or for both?

(and speaking of which, how come the big boy is numbered 10 and the little one 11? Counter-intuitive, isn't it?)

Polaris10=2304, Polaris11=1024. Need much explanation.

From the rumor, it seems both VEGA will be beaten by 1080Ti

If you translate his post



So your speculating on his speculation




The graph they put up is fake even the OP on Chiphell pointed it out

Too much rumors and drum beating



PC Games News Reported this today AMD have confirmed that the Radeon RX Vega cards are still on track for a release this quarter.

Yes. The Vega11 with GDDR5 is still on track.

I am thinking:

TitanXp > 1080Ti >> Vega10(HBM2) > Vega11(GDDR5) >= 1080

 

[HD64G]

Some of you forget that ALWAYS there are 2 iterations for each chip. So, Vega 10=2 products, Vega 11=another 2 products. At least. Too difficult to predict where each of those 4 will land on performance and price without hard numbers to base that predictions.

 

[Aenra]

Either way, like TheVoid said above, i still don't see this as a fail, even if it turns out so.
On a 1080 res, way it's going? I don't see me needing a replacement anytime soon. If i did though, would i not be happy having something near or equal to my GPU, but at half the price? Yessir i would.