videocardzAMD’s Greenland/Vega 10 GPU to feature 4096 Stream Processors

[Maverick177]

At least Cat guy posts really nice infos now and then. Now if only the other 2 could stop their blatant shilling.

 

[tenks]

Polaris SPs =/= Vega SPs, so those jumping to conclusions by comparing Polaris shader counts to justify what this mysterious Vega chip is (its Greenland Vega 10)..just stop. Polaris is closer to Fiji, Vega is a brand new uARCH. Saying Vega has to be double Polaris is a logical fallacy, they're different, apples and oranges.

 

[Madpacket]

We really don't have any idea of performance for Vega yet. I would like to think we're going to see double the performance of a Fury X but that's a bit of a stretch. I'm really more interested in async computing as that's what AMD really needs to push to separate themselves from team green.

 

[garagisti]

I can see this going as HD7970 and R9 290x went. At first we will be lucky to get anything close to 300mm2, but i wouldn't deny any such gifts that may come our way Then, the second chip will be bigger, i would think 400m2 if not bigger.

Nvidia, i think will launch something big closer to 350mm2-400mm2. They got to do something for compute.

 

[EightySix Four]

they're different, apples and oranges.

That makes no sense, why can't fruit be compared?

 

[swilli89]

Sorry Rvenger. But in the pic below, you'll notice why I don't post here anymore. Can't stand trash on a forum. When a pessimistic person post 25% of the post count on a thread to bash a company and kill the hype of the people, I call that a troll and a shill. I have no pleasure in participating on a forums with behavior like this.

It does not matter that you edited your post, member callouts are not allowed. Just to remind you, below is your original text that you got an infraction from:

"Spot the troll who constantly find a way to bash at AMD and post negativity in AMD rumor threads.

That troll is a reason why I don't post here anymore and this troll is garbage to the community... Constantly igniting wars.

The funny thing is that troll hates AMD and has no interest in their products...yet he has the biggest nunber of posts in the thread. Garbage needs to go at the dump."

Markfw900

So much this! ShintaiDK I really try to give you the benefit of the doubt but you constantly crap all over any AMD thread that would have been positive and exciting for hardware enthusiasts like myself. I've been a reader of these forums since 2002 and the forum has definitely had its fanboys on one side or the other, its part of how it is; but you have taken it to an entirely new level of consistent and abundant threadcrapping with almost autism-level shots at AMD and your little performance/watts crusade. Just get the hell out dude!! you ruin every. single. thread that would have otherwise been completely different and more enthusiastic. I can promise this forum gets less hits just because of you.


insulting other members is not allowed. You can't learn from me infracting him ?
Markfw900

 

[3DVagabond]

When do they not reflect it? You mean when people kept buying gtx970? Perhaps the people who bought it were not aware they could potentially get better bang for their money with comparable AMD card? So because they actually are not HW enthusiasts, not that much because they somehow care about their card to have say 150 instead of 200W consumption?

OFC they do the right thing. But only cause more performance per watt means more absolute performance. At the same wattage levels as before per usual tier of peformance.

Because all of the reviews were showing benchmarks from Hawaii reference cards in comparison. Reduced the scores by at least 10%. More on the 290X when they used quiet mode. Even though reference cards were near impossible to get. Now you routinely see O/C'd aftermarket cards in the comparisons.

 

[3DVagabond]

Sorry Silver, I agree with you. And I read the thread you linked and I have no doubt some members here like to help. You know I'm not pointing the finger at you bud. And you know who the usual bashers are .

Seriously Karl, use the ignore function. Then sit there and wish everyone would quit quoting him.

 

[MrTeal]

Polaris SPs =/= Vega SPs, so those jumping to conclusions by comparing Polaris shader counts to justify what this mysterious Vega chip is (its Greenland Vega 10)..just stop. Polaris is closer to Fiji, Vega is a brand new uARCH. Saying Vega has to be double Polaris is a logical fallacy, they're different, apples and oranges.

Source? I have seen no evidence of a major uArch change between Polaris and Vega. 4th Gen GCN is part of Polaris, but only part of it.

Thankfully for Polaris, RTG is revising their naming policies in order to present a clearer technical message about the architecture. Beginning with Polaris, RTG will be using Polaris as something of an umbrella architecture name – what RTG calls a macro-architecture – meant to encompass several aspects of the GPU. The end result is that the Polaris architecture name isn’t all that far removed from what would traditionally be the development family codenames (e.g. Evergreen, Southern Islands, etc), but with any luck we should be seeing more consistent messaging from RTG and we can avoid needing to create unofficial version numbers to try to communicate the architecture.

To that end the Polaris architecture will encompass a few things: the fourth generation Graphics Core Next core architecture, RTG's updated display and video encode/blocks, and the next generation of RTG's memory and power controllers. Each of these blocks is considered a seperate IP by RTG, and as a result they can and do mix and match various versions of these blocks across different GPUs, such as the GCN 1.2 based Fiji containing an HEVC decoder but not the GCN 1.2 based Tonga. This, consequently, is part of the reason why AMD has always been slightly uneasy about our unofficial naming. What remains to be seen then is how (if at all) RTG goes about communicating any changes should they update any of these blocks on future parts, and whether say a smaller update like a new video decoder would warrant a new architecture name.

Vega is a pretty major departure from Polaris as far as their naming scheme is concerned. We don't know all the details on Vega, but we do know the memory controller is radically different. Whether Vega would also include 5th Gen GCN is unknown, but given that 2nd Gen GCN appeared 15 months after Tahiti (for Bonaire, 22 months for Hawaii), and 3rd Gen GCN appeared 18 months after that, it seems unlikely that there would be major uArch changes ~6 months after Polaris launches.

 

[William Gaatjes]

Source? I have seen no evidence of a major uArch change between Polaris and Vega. 4th Gen GCN is part of Polaris, but only part of it.



Vega is a pretty major departure from Polaris as far as their naming scheme is concerned. We don't know all the details on Vega, but we do know the memory controller is radically different. Whether Vega would also include 5th Gen GCN is unknown, but given that 2nd Gen GCN appeared 15 months after Tahiti (for Bonaire, 22 months for Hawaii), and 3rd Gen GCN appeared 18 months after that, it seems unlikely that there would be major uArch changes ~6 months after Polaris launches.

I would not be surprised if vega will be polaris with architecture tweaks to remove the possible bottlenecks that will arise in the upcoming 18 months. Or maybe it means adding some extra instructions that make life easier for the driver or to enhance some dx12 features. If polaris is a s good as AMD /RTG says, they do not need to redesign from scratch. It could be just an iteration.

 

[Silverforce11]

Where's the evidence that Vega is a different uarch? Come on, do you guys think AMD is rolling in $$ to do multiple uarch in parallel?

It's just Polaris GCN without the GDDR5 MC and adapter for HBM2. There's no other indication it's anything else. On the time-table, it's 6 months apart. Essentially the same stack, just staggered in launch with smaller Polaris SKUs going first.

 

[Borealis7]

even so, what is the reason to buy a Polaris based card when Vega based cards will be available shortly after? unless high end will only be on one architecture and the mid-range/low-end will be on the other.

 

[MrTeal]

I would not be surprised if vega will be polaris with architecture tweaks to remove the possible bottlenecks that will arise in the upcoming 18 months. Or maybe it means adding some extra instructions that make life easier for the driver or to enhance some dx12 features. If polaris is a s good as AMD /RTG says, they do not need to redesign from scratch. It could be just an iteration.

Why do you think Vega will be arriving a full year or more after Polaris? It's possible given the ambiguity of what we've seen so far, but most people (including the professional press) seem to be looking at late 2016/early 2017, not Q4 2017.

 

[Silverforce11]

even so, what is the reason to buy a Polaris based card when Vega based cards will be available shortly after? unless high end will only be on one architecture and the mid-range/low-end will be on the other.

Money.

Not every gamer can afford high-end GPUs.

Obvious?

 

[Qwertilot]

A bit yes Not just money of course - there's people who just don't want the high end cards.

Higher power draws/larger cards etc.

Actually anyone who's got a 1080p monitor/isn't planning to upgrade - there is unlikely to be significant purpose in having anything past the low end cards of the next generation at that resolution.

This low/mid end is where the considerable bulk of the market is, and it'll be very good for competition (especially in laptops) to have AMD properly addressing it again.

 

[Mahigan]

Where's the evidence that Vega is a different uarch? Come on, do you guys think AMD is rolling in $$ to do multiple uarch in parallel?

It's just Polaris GCN without the GDDR5 MC and adapter for HBM2. There's no other indication it's anything else. On the time-table, it's 6 months apart. Essentially the same stack, just staggered in launch with smaller Polaris SKUs going first.

I think that a lot of the performance we're going to see will come from AMD and NVIDIAs new memory controller's.

Maxwell contains a gimped memory controller, to save on die space, as compared to Kepler. Color/Delta compression was used to make up for it (as well as doubling the ROp to memory controller ratio from 8:1 to 16:1). So Maxwell contains 16 ROps per 64-bit memory controller increment. The end result of this is bandwidth starved ROps.

We can see this here:

Kepler is able to hit its theoretical Pixel fillrate of 37GPixels/s whereas Maxwell cannot. This test only stresses ROps so the fact that with only the ROps occupying memory bandwidth leads to this sort of result doesn't bode well for gaming.

If we look at the memory bandwidth for texture operations we see the same result (glance at the random texture results as they're not affected by color compression):

We also see a similar behavior from Fiji, though Hawaii is affected to a lesser extent.

A scientific research paper confirms these findings here: http://arxiv.org/pdf/1509.02308&ved...qHEz78QnmcIVCSXvg&sig2=IdzxfrzQgNv8yq7e1mkeVg

This is the table, from the research paper, worth considering:

We also see Maxwell running out of L1 Instruction cache in its SMMs and spilling into L2 cache in between 16 and 32 Warps of occupation:

Shader occupancy is up in Maxwell, however, which makes up for that particular deficiency but it may explain why NVIDIA is not interested in Asynchronous compute + graphics because too many compute threads causes a performance penalty in Maxwell.

AMD, on the other hand, has nearly the same horrendous memory bandwidth efficiency figures as Maxwell with their Fiji (69%) when color compression isn't used and 76% when it is used.

It seems to me that both AMD and NVIDIA had made some large sacrifices in their 28nm designs in order to keep power usage down. Maxwell makes a ton of sacrifices throughout its caching hierarchy whereas Fiji makes its largest sacrifice in terms of memory bandwidth efficiency.

Most likely both AMD and NVIDIA shrank their memory controllers to save on logic and thus die space.

This bodes well for 16nm and 14LPPE and Finfet. As this allows both to reintroduce memory controller logic.

This may be why Polaris 10 is set to surprise us. For NVIDIA, the biggest surprise is set for GP100 but GP104 should also be interesting.

Anyway, my 2 cents.

 

[Silverforce11]

@Mahigan

You raise some good points. I always thought it was odd that Fiji performance scales so well with vram OC, because on paper, we all assume it's got lots of bandwidth and shouldn't be bottlenecked.

I'm sure next-gen is going to raise the bar, we're all pretty bored with 28nm and could use some new GPU/toys.

 

[zlejedi]

Meh - looks like GP100 is my only hope for worthy 980ti replacement.

Even if they have solid architecture improvements and remove bottlenecks it will most likely still stay in the same performance bracket as overclocked 980 ti with only 4096sp.

Of course there's also a matter of clock speeds - we have basically zero information how this will compare to current 28nm. If they are 30-40% higher it would make those gpus solid. On the other hand if it goes like Intel die shrinks we could see stagnation or small regression in clocks.

 

[Mahigan]

@Mahigan

You raise some good points. I always thought it was odd that Fiji performance scales so well with vram OC, because on paper, we all assume it's got lots of bandwidth and shouldn't be bottlenecked.

I'm sure next-gen is going to raise the bar, we're all pretty bored with 28nm and could use some new GPU/toys.

Yep, I thought it was weird as well.

Just like I thought it was weird that a GTX 980 Ti is unable, less overclocked, to keep up with a FuryX at 4K resolution. You'd think that 96ROps would dominate 64ROps but that's not the case.

Spoiler

Even at stock clocks, the GTX 980 Ti clocks its ROps higher than those of a FuryX. The GTX 980 Ti's caches are also clocked higher than those of a FuryX (as ROps also tap into their own caches and/or the available L2 cache).

People who looked at Fiji's 64 ROps and claimed they were weak simply weren't taking any of the pipelining, dedicated caches and available memory bandwidth into consideration.

If we do take these elements into consideration then we can see how a die shrink, from 28nm to 14LPP, can result in a dramatic increase in Global Memory usage efficiency due to there being more room for more robust memory controller logic.

 

[ShintaiDK]

Where's the evidence that Vega is a different uarch? Come on, do you guys think AMD is rolling in $$ to do multiple uarch in parallel?

It's just Polaris GCN without the GDDR5 MC and adapter for HBM2. There's no other indication it's anything else. On the time-table, it's 6 months apart. Essentially the same stack, just staggered in launch with smaller Polaris SKUs going first.

Agree. The delayed launch seems to be triggered mainly due to bigger size(yield) and HBM2 supply. I could also see a Polaris refresh in 2017 with GDDR5X, unless it pulls a joker and launches with it.

 

[ShintaiDK]

even so, what is the reason to buy a Polaris based card when Vega based cards will be available shortly after? unless high end will only be on one architecture and the mid-range/low-end will be on the other.

Its low to high. Polaris 11->Polaris 10->Vega 11->Vega 10.

 

[Vesku]

Agree. The delayed launch seems to be triggered mainly due to bigger size(yield) and HBM2 supply. I could also see a Polaris refresh in 2017 with GDDR5X, unless it pulls a joker and launches with it.

Am hesitant about buying mid range next generation but 3-4 month wait vs 6-9 is a big deal for those like me who want to go straight from 7950/70, 670/80 to new node. GDDR5X would seal a mid-range purchase for me, especially if AMD Raja's budget VR comments means a roughly $300 card that gets mostly 60fps+ mins at 1440p.

 

[Arachnotronic]

Just reread all your posts and the cat's posts and We will all know in which boat you float. And it is funny how you say the AMD posters make the forums unpleasant because the Nvidia posters are doing the same or worse...you know, the other boat I am referring to. That is why I tend to avoird this UNPLEASANT forum.

OK, thanks for sharing.

And when I come in a news thread expecting to learn something and instead I see bashing all around, it pisses me off. Just notice my post count since a year... This forums is now a POS.

Tell us how you really feel.

And your poll thread was all about what? Igniting flame wars? This is another reason why this forum is a POS. This forum is filled with flame threads...look the threads with OPs asking for help...nobody helps...because members here want to bash, not help.

Nope, not about "igniting flame wars." There are a few posters around here who claimed that these forums were "pro-NVIDIA" because NVIDIA now has dominant market share. I wanted to get to the truth and the truth is these forums are not as "NVIDIA-biased" as some people loudly proclaim anytime somebody has the gall to suggest buying/using an NVIDIA product.


Thread crapping and off-topic is not allowed here. This was only a warning, but you are on thin ice.
Markfw900

 

[gamervivek]

Source? I have seen no evidence of a major uArch change between Polaris and Vega. 4th Gen GCN is part of Polaris, but only part of it.



Vega is a pretty major departure from Polaris as far as their naming scheme is concerned. We don't know all the details on Vega, but we do know the memory controller is radically different. Whether Vega would also include 5th Gen GCN is unknown, but given that 2nd Gen GCN appeared 15 months after Tahiti (for Bonaire, 22 months for Hawaii), and 3rd Gen GCN appeared 18 months after that, it seems unlikely that there would be major uArch changes ~6 months after Polaris launches.

The linkedin profile of AMD employee had info of Project 'Greenland' with graphics ip version 9. Tonga/Fiji and the Polaris chips have it at 8. So Vega with 9 would be more than two years since Tonga's release.

As for GCN generations, AMD probably checks them on smaller chips before using them. GCN 2nd gen chips were in mobile pretty much within a year after Tahiti's release.

http://forums.anandtech.com/showpost.php?p=37483897&postcount=158

http://www.blogcdn.com/www.engadget.com/media/2012/12/amd8000-1.jpg

 

[MrTeal]

The linkedin profile of AMD employee had info of Project 'Greenland' with graphics ip version 9. Tonga/Fiji and the Polaris chips have it at 8. So Vega with 9 would be more than two years since Tonga's release.

As for GCN generations, AMD probably checks them on smaller chips before using them. GCN 2nd gen chips were in mobile pretty much within a year after Tahiti's release.

http://forums.anandtech.com/showpost.php?p=37483897&postcount=158

http://www.blogcdn.com/www.engadget.com/media/2012/12/amd8000-1.jpg

The 8800M series was still the GCN1(.0)-based Oland along with Cape Verde and Pitcairn. GCN2(1.1) didn't show up until Bonaire.

Do you have a source stating that Polaris uses the same graphics IP as Tonga/Fiji. AMD seemed pretty explicit in their launch event that Polaris was GCN 4 vs the 3rd gen used in Fiji.