Vega/Navi Rumors (Updated)

[zinfamous]

I was using common sense. I was letting everyone know that the same card they used to test AOTS 3 weeks ago, was the same card they used for the Doom benchmarks in todays article. I was not arguing, its a fact.

We all know big Vega to be Vega 10, correct?

Quote:
"AMD Vega 10 With 8 GB HBM2 – Device ID “687F:C1” Confirmed"

quote:
Vega’s Confirmed Specs
Members of the press inside the demo room were able to spot some key specifications pertaining to Vega by taking a look at the expanded statistics in Doom. 8GB of HBM2 memory for the consumer version of Vega was confirmed.

So in conclusion what we are seeing IS big Vega 10, correct? Yes it is.
Next conclusion we come to after seeing the 2 leaked benchmarks is that this card is Vega 10, otherwise known as big Vega and is a little faster than a gtx1080 in AMD favored games.

It seems rather simple.

I don't know how we know that Vega 10 is big Vega simply because some confirmed serial number is Vega 10.

that is failed logic.

Do you have something real to go with, here? I'm not saying you are wrong, just that you keep posting the same blurb and claim it is saying something that it really doesn't.

For my sake, I am not sure how the confirmed Vega chip in the MI25 card, which AMD claims to be some 20-30% faster than Titan XP is smaller than this Vega 10 that you are saying is obviously Big Vega. I just don't get it.

Besides the fact that one of your favorite wccftech posts some month or so ago claimed that Vega 10 = small Vega and Vega 11 = big Vega, in a reversal of the Polaris scheme (remember that the number has nothing to do with the size, merely the release schedule)

 

[linkgoron]

I was using common sense. I was letting everyone know that the same card they used to test AOTS 3 weeks ago, was the same card they used for the Doom benchmarks in todays article. I was not arguing, its a fact.

We all know big Vega to be Vega 10, correct?

Quote:
"AMD Vega 10 With 8 GB HBM2 – Device ID “687F:C1” Confirmed"

quote:
Vega’s Confirmed Specs
Members of the press inside the demo room were able to spot some key specifications pertaining to Vega by taking a look at the expanded statistics in Doom. 8GB of HBM2 memory for the consumer version of Vega was confirmed.

So in conclusion what we are seeing IS big Vega 10, correct? Yes it is.
Next conclusion we come to after seeing the 2 leaked benchmarks is that this card is Vega 10, otherwise known as big Vega and is a little faster than a gtx1080 in AMD favored games.

It seems rather simple.

Nobody (except AMD) knows if Vega 10 is Big Vega or not. You could read this post, where multiple sources (quoted from around September) speculated that Vega 11 is Big Vega.

edit: copied from that thread

http://www.guru3d.com/news-story/amd-vega-10-vega-20-and-vega-11-gpus-mentioned-by-cto.html

Vega 11
Vega11 - lets call it "Big Vega' which is to replace the FIJI / Fury (X) parts. So as suggested, where Vega10 would get 4096 stream processors and could compete with the GeForce GTX 1080 and GTX 1070 Vega11 on its end would feature 6144 stream processors (=rumor) and would be lined against the GP100/GP102 aka Big Pascal GPU (Titan X). This all is obviously known and discussed many times already. Then there now is some new and intersting info.

http://www.fudzilla.com/news/graphics/40662-amd-allegedly-pulls-vega-launch-forward

The Vega 10 GPU is rumored to be a smaller chip with up to 4096 Stream Processors and this is the chip that AMD needs in order to compete with Nvidia's new GP104 GPU and Geforce GTX 1080/1070 graphics cards. The Vega 11, is a bigger chip, rumored to come with up to 6144 Stream Processors and compete with Nvidia's future GP100 flagship graphics card.

 

[Arachnotronic]

If this is big Vega there is little reason for AMD to produce little Vega. Unless little Vega is GDDR5/X then there would still be a reason to sell it between P10 and big Vega. Also this Doom performance preview looks to be about where a Polaris GPU would hit with 9-10 TFlops. With the Doom 4K demo we either have a small Vega with better gaming/TFlop ratio than Polaris or a perf/W optimized big Vega demo where it's gaming/TFlop is barely changed from Polaris.

So you're telling me there's no room in between a 5.8TFLOP RX 480 and a 12.5TFLOP Fury Next for a "small Vega"? Interesting.

 

[HurleyBird]

So you're telling me there's no room in between a 5.8TFLOP RX 480 and a 12.5TFLOP Fury Next for a "small Vega"? Interesting.

There's room in terms of performance, but HBM2 memory is going to be pretty borderline economically for that performance bracket. We also know that a Polaris 12 chip exists which might already be targeted between Polaris 10 and Vega 10, assuming Vega 10 is small Vega.

 

[happy medium]

SO your suggesting that Polaris 10 is faster than Polaris 11,
but Vega 10 will be slower than Vega 11

So Vega 10 has 4096 sp's, 8gb HBM (Confirmed) at about 200 watts, and Vega 11 would then have 6144 stream processors, ~ 50% faster than Vega 10 ,and I'm guessing 16gb of HBM at about 250 watts.

I guess its possible, but I wouldn't get my hopes up.

 

[HurleyBird]

SO your suggesting that Polaris 10 is faster than Polaris 11,
but Vega 10 will be slower than Vega 11

As has been pointed out about a hundred times by this point, the code names are based on production date, not performance or die size. 10 is made first, then 11, and then 12, and so on. That's the only pattern. There's nothing else to read into it.

 

[Arachnotronic]

As has been pointed out for about a thousand times, the code names are based on production date, not performance or die size. 10 is made first, then 11, and then 12, and so on. That's the only pattern. There's nothing else to read into it.

yeah, that was the PR party line, but it turned out to be complete nonsense. In an interview with PCPer, Koduri said they focused on getting Polaris 11 right in order to service the OEMs and then shifted focus on Polaris 10.

 

[IllogicalGlory]

As has been pointed out about a hundred times by this point, the code names are based on production date, not performance or die size. 10 is made first, then 11, and then 12, and so on. That's the only pattern. There's nothing else to read into it.

Seriously, this has been pointed out every time this comes up. Don't know how regulars haven't picked up on this yet.

That being said, WCCF holds that Vega 11 is the "580" or an upgraded 480 in a similar vein to the R9 285, I guess. It could really be either.

 

[HurleyBird]

yeah, that was the PR party line, but it turned out to be complete nonsense. In an interview with PCPer, Koduri said they focused on getting Polaris 11 right in order to service the OEMs and then shifted focus on Polaris 10.

The names don't refer to when the design was finished. They refer to when the design was started. In what universe would it be the other way around? And of course they'd want to get Polaris 11 in shape as fast as possible to demo to OEMs since there's a lead time for those systems and you need to make the deadline or you're out of the running.

And even then, did you forget that the RX 480 launched first, followed by the RX 470, followed by the RX 460?

But even putting aside all of that, what possible motivation would AMD have to lie about their naming scheme? Even assuming the tinfoil hat stuff is correct and AMD brought up Polaris 11 first, and then decided to switch around the code names because larger chip showed up in the market as the RX 480 before the smaller chip, how does that prove in the slightest that AMD is secretly naming Polaris chips based on die size or performance? It's the dumbest conspiracy theory ever. You're really grasping at straws.

 

[Vesku]

So you're telling me there's no room in between a 5.8TFLOP RX 480 and a 12.5TFLOP Fury Next for a "small Vega"? Interesting.

Whatever is in the Doom demo box is getting performance similar to a Polaris card with 9-10 TFlops. Now it could be the same chip in the 12.5 TFlop MI25 card just at lower clocks or it could be a smaller chip with ~9 TFlops and some gaming improvements from Polaris. I suppose it could also be a Vega clocked at the 12.5 TFlop level that has a regression in gaming/TFlop ratio but I'm giving AMD engineers some credit and think that's very unlikely.

If it is big Vega than small Vega would have to be roughly 1070 performance and using HBM2 on such a die would be an odd decision both in matching performance to bandwidth and meeting the expected retail price.

 

[HurleyBird]

Whatever is in the Doom demo box is getting performance similar to a Polaris card with 9-10 TFlops. Now it could be the same chip in the 12.5 TFlop MI25 card just at lower clocks or it could be a smaller chip with ~9 TFlops and some gaming improvements from Polaris.

I've heard some mumbling that it may be a cut down version of the 4096SP card. But I don't buy it. Why would AMD demo a cut down GPU, without specifying what it is, as a way to generate good publicity? That makes zero sense. They'd want to put the best foot forward. I suppose it's possible that yields are just terrible and they had to settle for a cut down chip, but that's reaching. If it is a cut down version, I'd be more likely to attribute that to be yet another dumb execution mistake by AMD management.

And if it's clocked lower than they expect it to be in retail, then that sounds like another dumb execution mistake. Slap a water cooler (or some other beefy cooling solution) on it and run it at the clocks you expect to be at when you launch. Again, unless you're just getting really disappointing chips back from the foundry.

 

[Arachnotronic]

I've heard some mumbling that it may be a cut down version of the 4096SP card. But I don't buy it. Why would AMD demo a cut down GPU, without specifying what it is, as a way to generate good publicity? That makes zero sense. They'd want to put the best foot forward. I suppose it's possible that yields are just terrible and they had to settle for a cut down chip, but that's reaching. If it is a cut down version, I'd be more likely to attribute that to be yet another dumb execution mistake by AMD management.

Someone found on LinkedIn that the full GPU has 4096 cores.

Anyway, I don't know why people are so desperate for there to be a higher performance Vega...this thing seems to go toe-to-toe with the 1080 (even edging it out), which itself is 20-30% faster than the Fury X. It will probably consume less power than the Fury X, so it'll be substantially more efficient.

If one is an AMD buyer, this is a big step up from what you can buy today. I don't get why people seem "disappointed" in it.

 

[swilli89]

Someone found on LinkedIn that the full GPU has 4096 cores.

Anyway, I don't know why people are so desperate for there to be a higher performance Vega...this thing seems to go toe-to-toe with the 1080 (even edging it out), which itself is 20-30% faster than the Fury X. It will probably consume less power than the Fury X, so it'll be substantially more efficient.

If one is an AMD buyer, this is a big step up from what you can buy today. I don't get why people seem "disappointed" in it.

You do realize this is an enthusiast board right? And that nVidia has faster chips already faster than 1080? And what is an "AMD Buyer?"

 

[nurturedhate]

Someone found on LinkedIn that the full GPU has 4096 cores.

Anyway, I don't know why people are so desperate for there to be a higher performance Vega...this thing seems to go toe-to-toe with the 1080 (even edging it out), which itself is 20-30% faster than the Fury X. It will probably consume less power than the Fury X, so it'll be substantially more efficient.

If one is an AMD buyer, this is a big step up from what you can buy today. I don't get why people seem "disappointed" in it.

The tone of your post would leave one to think you think rather lowly of people who "buy" "AMD". That seems rather "disappointing".

It's simple really. There's two vega chips. They will be sold in concert with polaris. This vega chip seems to match GP104. You salvage some and you have a 1070 competitor. There's the 480, 470, and 460 as well.

Do you want to go on record stating you believe the other vega chip is smaller than polaris 11?

 

[Arachnotronic]

You do realize this is an enthusiast board right? And that nVidia has faster chips already faster than 1080?

One faster chip. And I would say that the performance level of the GTX 1080 is solidly enthusiast class, most gamers (even enthusiasts) are running much slower GPUs.

And what is an "AMD Buyer?"

People who are locked into the AMD ecosystem, either because they love AMD/dislike NVIDIA (brand loyalty/hatred is a thing, and it's not necessarily bad or good, it just "is") or own a FreeSync monitor. It also includes people who have historically used AMD, had good experiences with AMD, and want to continue to use AMD because of said good experiences because, -- why use something different if you've had good luck with the vendor you're currently using?

 

[Arachnotronic]

The tone of your post would leave one to think you think rather lowly of people who "buy" "AMD". That seems rather "disappointing".

You're choosing to read into it that way, that's not my problem.

Do you want to go on record stating you believe the other vega chip is smaller than polaris 11?

I've already stated what I believe. I think Vega 10 is as large as it gets for the enthusiast Vega lineup, but that's just my opinion. Why do you care so much about my opinion?

 

[nurturedhate]

You're choosing to read into it that way, that's not my problem.



I've already stated what I believe. I think Vega 10 is as large as it gets for the enthusiast Vega lineup, but that's just my opinion. Why do you care so much about my opinion?

So you are of the opinion that vega 11 is smaller than polaris 11 then.

 

[happy medium]

So you are of the opinion that vega 11 is smaller than polaris 11 then.

why would Vega 11 be smaller than Polaris 11?

I think the Vega in the article is big Vega. Call it Vega 10 or 11, don't matter.

 

[nurturedhate]

why would Vega 11 be smaller than Polaris 11?

That's my thought as well happy. but arach believes vega 10 is the larger vega and it matches gp104 then a cut down version would be around 1070, then we have 480, 470 and 460 there's no where to fit another card in between all that. With all of that arach would seem to believe that vega 11 is smaller than polaris 11 as there's no where else in the lineup for it to go since he says there isn't a larger vega.

Remember Happy, this isn't my thoughts on vega. This is the conclusion of Arachnotronic, maybe you should ask him why he says what he says.

 

[happy medium]

That's my thought as well happy. but arach believes vega 10 is the larger vega and it matches gp104 then a cut down version would be around 1070, then we have 480, 470 and 460 there's no where to fit another card in between all that. With all of that arach would seem to believe that vega 11 is smaller than polaris 11 as there's no where else in the lineup for it to go since he says there isn't a larger vega.

Remember Happy, this isn't my thoughts on vega. This is the conclusion of Arachnotronic, maybe you should ask him why he says what he says.

OK,,
I think it would be easier to just say what I think.
Performance wise.............
I think big Vega(Fury) is in between a gtx1080 and a Titan XP.
Small Vega (490) fits in between the gtx1070/1080
480 vs gtx1060

Sucks that both will have a large gap between the 480/1060 and the gtx1070/490

 

[raghu78]

I think by now we can pretty much combine all the bits of information we have been getting with the recent MI25 press release.

http://videocardz.com/63700/exclusive-first-details-about-amd-vega10-and-vega20

http://videocardz.com/64677/amd-ann...erator-radeon-instinct-mi25-for-deep-learning

http://wccftech.com/amd-vega-10-4096-gcn/

videocardz got it spot on with their leaks a few montha back. Vega 10 is the big flagship GPU. I think Vega Vega 11 is Polaris 10 with the Vega architecture and most probably G5X . I do not think HBM2 is going to be cost competitive and available in very high volume to be used in the sub USD 300 market.

Vega 10 - 12.5 TFLOPS , 1525 Mhz , 4096 cores, 64 ROPs, 512 GB/s, 250W . 8GB HBM2 (Radeon) / 16 GB HBM2 (Firepro) using 8 Hi stacks .
Vega 11 - 7 TFLOPS , 1525 Mhz, 2304 cores, 32 ROPs, 320 GB/s, 8 GB G5X

I am pretty sure the Polaris 11 chip will also be refreshed with Vega GFX 9 architecture. We will see Vega 10 in Q1 2017, Vega 11 in Q2 (probably mid 2017) and the smallest Vega chip in Q3 2017. I think the key factor which will determine AMD's product competitiveness is how much of an improvement does Vega bring over Polaris for perf/sp at same clock. If that value is <10% then AMD will have a very tough time against Pascal and Pascal refresh. If perf/sp increase is around 20% then they have a fighting chance provided pricing is aggressive.

btw I think Vega 10 will probably double up on the entire front end too with 8 shader, raster and tesselation engines. So its going to be quite a big chip with a die size close to 450 sq mm.

 

[zinfamous]

SO your suggesting that Polaris 10 is faster than Polaris 11,
but Vega 10 will be slower than Vega 11

So Vega 10 has 4096 sp's, 8gb HBM (Confirmed) at about 200 watts, and Vega 11 would then have 6144 stream processors, ~ 50% faster than Vega 10 ,and I'm guessing 16gb of HBM at about 250 watts.

I guess its possible, but I wouldn't get my hopes up.

Your guess is anyone's guess, but it has long been "substantiated" that there is a 16GB HBM Vega, anyway. This 8GB card doesn't sound like the flagship to me.

I'm not sure why big Vega needs to be "1080" and small Vega needs to be "1070". Small Vega could be "1080", with cutdown small Vega ~1070. Yeah, there's still a gap...but it looks like 480 is already filling that now as it is consistently pulling ahead of 1060 with the last 3 updates. 1060 has remained completely stagnant in performance since release. (I know, it's only slight performance gains, so no real gap-filling. just saying! )

You could just read Ryan's article today about Instinct, that shows a Vega card that supposedly outperforms Titan XP in, whatever this is supposed to be:

So, yeah, there is a Vega out there that is certainly bigger than this thing in the Doom box. Will it show up in dGPU consumer cards? Dunno...

 

[raghu78]

Your guess is anyone's guess, but it has long been "substantiated" that there is a 16GB HBM Vega, anyway. This 8GB card doesn't sound like the flagship to me.

Its very simple. 16 GB HBM2 can be realized by using 8 Hi stacks on a 2048 bit memory bus. btw its very likely that Vega 10 Radeon uses 4 Hi stacks on 2048 bit memory bus thus providing 8GB HBM2 and Vega 10 Firepro will use 8 Hi stacks on 2048 bit memory bus to provide 16 GB HBM2 . I think Vega Radeon will launch in Q1 2017 while Vega Firepro will launch by mid-2017 along with Zen Naples. AMD's marketing for MI25 is all about the synergy of Zen Naples and Vega Firepro.

 

[HurleyBird]

You could just read Ryan's article today about Instinct, that shows a Vega card that supposedly outperforms Titan XP in, whatever this is supposed to be:

So, yeah, there is a Vega out there that is certainly bigger than this thing in the Doom box. Will it show up in dGPU consumer cards? Dunno...

The same graph shows Fiji outperforming the TitanXP, so I wouldn't read that much into it. Chances are that it's Vega 10 in both the Doom test and this new professional card.

 

[swilli89]

The same graph shows Fiji outperforming the TitanXP, so I wouldn't read that much into it. Chances are that it's Vega 10 in both the Doom test and this new professional card.

Looking at this graph I'd say a more meaningful takeaway is that this Vega card is showing to be about 41% faster than Fiji. Will be interesting to see the price if true..