Vega/Navi Rumors (Updated)

Page 65 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.

Valantar

Golden Member
Aug 26, 2014
1,792
508
136
the vega with 64 CUs is big(!!) vega -> the vega 10. small vega will be vega 11, the 'sweet spot' GPU according to AMD. It's starting to be frustrating, why does somebody new come every single day and try to spread this nonsense, that the 64 CU 4096 SP vega 10 is '''''small'''' vega with its close to 500mm2 die size?
they will be much bigger, that's almost 100% certain

edit: formatting
That might be, but that would signal such a radical architecture change I have trouble imagining them calling it GCN anymore ... Since Polaris 10's area extrapolated to 64 CUs is 412mm2, that would mean that they've added ~25% area per CU. I can't see that coming from anything but the CUs (probably a small area increase for the HBCC over a traditional memory controller, but not even close to 100mm2 at 14nm). That's a huge difference - and would require either each CU having at least 25% more SPs, or IPC increasing by at least 25%/CU (with a net power usage drop from P10 per CU (as 64CUs is a 77% increase from P10, while a 250W TDP is a 66% increase from the RX 480) for it to make sense.
 

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,425
8,388
126
After the coming Polaris respin, how cheaply do you think AMD can sell the "tuned" up Radeon RX480 for..? Are the yields there, where AMD can go lower on the price for the 480/580..?

$149 bucks..? (-$50).
Taking on the 1080p & 1440p consumers. Leaving Radeon RX Vega brand for VR & 4k.

Is Polaris 12 a mobile GPU or Polaris on 14LPP (as opposed to LPE)?

Polaris 12 is a smaller Polaris, nothing special. 640 SP's.

All of Polaris is already made on LPP, the rumor Videocardz was talking about was of minimal credibility.

Someday, I will learn

not gonna happen, unfortunately

Polaris 10 GPUs will cost up to 299$.

that's not gonna happen either, fortunately.

Not that I think they can squeeze such performance out of Polaris 10 with this 1 year refresh, but if the 570 or 580 (whatever the new top Polaris becomes) manages 85-95% of 1070, then $299 is more than fair, imo.

Besides, that "up to $299" price target reflects AMD's own words about a year ago, just ahead of Polaris launch. If it happens, it just means AMD were right all along, and this was their goal all along, and further discounts the previous speculation that 480X only launched at that up to $250 price "due to nVidia pricing and performance." Not that any of that speculation isn't true, but none of it isn't not yet true, either.


Can y'all keep the polaris rumors to a polaris thread? kthxbai

AT Moderator ElFenix
 
Last edited:

w3rd

Senior member
Mar 1, 2017
255
62
101
"Fiji was 596mm2 with 4096 SPs/64 CUs. Hawaii was 438mm2 with 2560 (although that includes more FP64 components due to its 1/2 DP compute rate vs. Fiji's 1/16)."

-Valantar



And what process node was Fiji & Hawaii on..?
 

Valantar

Golden Member
Aug 26, 2014
1,792
508
136
"Fiji was 596mm2 with 4096 SPs/64 CUs. Hawaii was 438mm2 with 2560 (although that includes more FP64 components due to its 1/2 DP compute rate vs. Fiji's 1/16)."

-Valantar



And what process node was Fiji & Hawaii on..?
Uh, yeah, they're on 28nm, isn't that reasonably clear from the fact that I subsequently compare them to Polaris and use that shrink as a basis to guesstimate Vega die sizes? Or is that not clear? I know I didn't spell it out, but I really didn't think that would be necessary in a discussion like this. Or did you not read past that part? Did you think I was implying the 40%+ die area shrink between Hawaii and Polaris was due to architectural changes? Don't you think I'd be slightly more amazed at that?

If you're implying that the comparison isn't relevant: of course it is, as Hawaii is the (semi-)direct predecessor to Polaris 10 (it does have 4 CUs less), and Fiji is the same to Vega. Unless there's a high-end AMD chip in between that only you know of?


Edit: damn autocorrect
 
Last edited:

Snarf Snarf

Senior member
Feb 19, 2015
399
327
136
"Fiji was 596mm2 with 4096 SPs/64 CUs. Hawaii was 438mm2 with 2560 (although that includes more FP64 components due to its 1/2 DP compute rate vs. Fiji's 1/16)."

-Valantar



And what process node was Fiji & Hawaii on..?

Come on man, he's using the scaling on 28nm as a point of reference when inferring how die size has changed on 14nm. Don't be dense, all we have to speculate off of are known quantities and this kind of extrapolation is common on this forum. It's purely academic and for fun, we get kicks off of trying to guess what might be correct in future releases using what knowledge is available to make estimates.

Also there's a quote button friend, use it so that the whole post gets referenced and context of the original statement isn't lost.
 

Valantar

Golden Member
Aug 26, 2014
1,792
508
136
I dont see anything in those slides saying 64CU is the big VEGA.
Me neither. It doesn't say anything about CUs at all, only what products it's slated to replace in various server/HPC workloads. Heck, Fiji (AMD's only 64-CU part until now) is only represented by the Pro S9 Nano in that chart, so by that logic Vega 11 (small Vega, single-slot card) should have 64 CUs as it slots in in the same market segment.


Also, I love how VideoCardz has taken a picture watermarked by another site, blurred out the watermark, and added their own. Not that I expect much more from them, but still... Stay classy, VideoCardz.
 
Mar 10, 2006
11,715
2,012
126
Me neither. It doesn't say anything about CUs at all, only what products it's slated to replace in various server/HPC workloads. Heck, Fiji (AMD's only 64-CU part until now) is only represented by the Pro S9 Nano in that chart, so by that logic Vega 11 (small Vega, single-slot card) should have 64 CUs as it slots in in the same market segment.


Also, I love how VideoCardz has taken a picture watermarked by another site, blurred out the watermark, and added their own. Not that I expect much more from them, but still... Stay classy, VideoCardz.



64CUs.
 
Reactions: CatMerc

IllogicalGlory

Senior member
Mar 8, 2013
934
346
136
I dont see anything in those slides saying 64CU is the big VEGA.
Vega 11 is positioned to replace Polaris 10/11, where Vega 10 replaces Fijix2. Plus it's not actually mentioned anywhere in the slides. Vega 10 and Vega 20 are the important ones, and both have 64 CUs.

You can go on believing that there is a bigger Vega than 64CUs, but there isn't. Vega has a different kind of design than Polaris - high clocks, fewer CUs.
 
Last edited:

Glo.

Diamond Member
Apr 25, 2015
5,765
4,668
136
Vega 11 is positioned to replace Polaris 10/11, where Vega 10 replaces Polaris 10x2. Plus it's not actually mentioned anywhere in the slides. Vega 10 and Vega 20 are the important ones, and both have 64 CUs.
In professional market.

P.S. Don't you think that otherwise, rumors about RX 500 lineup, would be a little illogical...?
 

IllogicalGlory

Senior member
Mar 8, 2013
934
346
136
In professional market.

P.S. Don't you think that otherwise, rumors about RX 500 lineup, would be a little illogical...?
Yes, and Vega 10 replaces Polaris 10x2 in the professional market, clearly being a more expensive, higher performance solution than the one that replaces Polaris 11. Vega 20 is Vega 10 on 7nm with double precision. It would make no sense to put the smaller chip in that position.

I don't know what you mean about the RX 500 series, can you elaborate? All we know about that so far are the Polaris refreshes.
 

Glo.

Diamond Member
Apr 25, 2015
5,765
4,668
136
Yes, and Vega 10 replaces Polaris 10x2 in the professional market, clearly being a more expensive, higher performance solution than the one that replaces Polaris 11. Vega 20 is Vega 10 on 7nm with double precision. It would make no sense to put the smaller chip in that position.

I don't know what you mean about the RX 500 series, can you elaborate? All we know about that so far are the Polaris refreshes.
Well if Vega 11 would be positioned to replace Polaris 10 series, overall, then RX 500 series, on which we know are Polaris refresh would be pointless and illogical...
 

IllogicalGlory

Senior member
Mar 8, 2013
934
346
136
Well if Vega 11 would be positioned to replace Polaris 10 series, overall, then RX 500 series, on which we know are Polaris refresh would be pointless and illogical...
I never said it was, only that where it was positioned in the professional market was indication of its performance segment.

Edit: Furthermore, the existence of Vega 10x2 is pretty much a guarantee that it's the largest Vega chip. What sense would it make to make a dual GPU design out of the small chip?
 

zinfamous

No Lifer
Jul 12, 2006
110,810
29,564
146
Well if Vega 11 would be positioned to replace Polaris 10 series, overall, then RX 500 series, on which we know are Polaris refresh would be pointless and illogical...

at the risk of getting the hammer for talking Polaris again...all we seem to know is RX 580 and below (I think 580, maybe 570 and below) are the Polaris refreshes. It is possible that 590/590X are the Vega 11 cards (which may not come until the end of the year...or even 2018? if that Videocardz-rebranded slide is to be believed), where Vega 10 is the "Radeon RX Vega" lineup. However, AMD did mention that "Vega cards will be called Radeon Vega," so I assume this means all Vega cards, but in that slide, I see no Vega 11 in the 2017 quarter/half roadmap.

I think it is likely that for dGPU this year, Vega 10 is the flagship, Radeon refresh fill out the midrange (but would probably still leave a huge gap, unless there is a cut-down Vega 10 that slots somewhere between 1070/1080 performance).

"Vega 11" may appear in cut-down form as parts for Raven Ridge and/or Apple-branded laptops/iMacs/macPros or whatever--so no Vega 11 dGPU until next year, when it actually does replace Polaris? Maybe we see 590/590X...and quite possible 580X? but then it wouldn't be called Vega, so maybe a different naming scheme for the next card that fills out the 290 > 390 line.

Perhaps I have forgotten some information somewhere or I am just reading more into that slide up there--it only talks about Vega 10, so maybe it is just about Vega 10 parts in that one slide, and not the full Vega story for 2017.
 

Mopetar

Diamond Member
Jan 31, 2011
8,016
6,466
136
at the risk of getting the hammer for talking Polaris again...all we seem to know is RX 580 and below (I think 580, maybe 570 and below) are the Polaris refreshes. It is possible that 590/590X are the Vega 11 cards (which may not come until the end of the year...or even 2018? if that Videocardz-rebranded slide is to be believed), where Vega 10 is the "Radeon RX Vega" lineup. However, AMD did mention that "Vega cards will be called Radeon Vega," so I assume this means all Vega cards, but in that slide, I see no Vega 11 in the 2017 quarter/half roadmap.

From what we know of big Vega, it seems as though there will be 3 consumer parts and I think the fill the Fury, Fury X, and Fury Nano categories. I don't know what Vega 11 ends up getting called, but it seems like a mixed lineup almost doesn't make sense. Sure you can call them Vega Fury or something like that to denote the difference between the old and new Fury, or just call it the Vega, Vega X, Vega Nano and drop Fury completely since it isn't something that I think will get a lot of fond memories down the road. At best it was ahead of its time.

If small Vega is going to take the x90 slot, it seems like the cut version should probably take the x80 slot and relegate Polaris to x70 and below. A few powers speculated that there won't be an X cards since we've moved from R9 to RX and at some point, it adding more X's just gets ridiculous. No idea how to work Vega into the name without it getting cumbersome though.

Personally, I think it would be cool if they dropped numbers with Vega and played more into the astronomy side of things. Big Vega gets designated as Vega Black Star or something like that. It's a little bit like a Black Edition throwback in a way. So Vega Black Star, Vega Black Star X, and Vega Black Star Nano for the big Vega Cards. Below that the smaller Vega 11 can be labeled Red Star and Red Star X to denote cut and full chips. If you need another rung for an even smaller die, just pick another color like blue or white.
 

xpea

Senior member
Feb 14, 2014
449
150
116
Guys,

in your die size estimation, you forgot that Maxwell has 4 times more cache than Kepler (cache used by the tile rasterization), taking out space from pure ALUs count. Vega will be the same thus wont scale linearly its ALUs compared to Polaris...

my 3cts
 

Ajay

Lifer
Jan 8, 2001
16,094
8,106
136
Guys,

in your die size estimation, you forgot that Maxwell has 4 times more cache than Kepler (cache used by the tile rasterization), taking out space from pure ALUs count. Vega will be the same thus wont scale linearly its ALUs compared to Polaris...

my 3cts

I was thinking the same thing.
 

w3rd

Senior member
Mar 1, 2017
255
62
101
What if.....
V10 is not found anywhere, because what you see in these slides & leaks is V11. (Masked as V10.)
With V10 being held behind some tight lips.

And AMD's respin of Polaris is a market grab maneuver at ultra low costs.
(AMD's version of the 8800GT).




Is that a viable rumor?
 

AtenRa

Lifer
Feb 2, 2009
14,003
3,361
136
OK i forgot about the second slide, it seems Vega 10 with 64CUs will be the big one. And from the AMD Vega architecture release back in January it seems that each Vega CU will be a lot faster than GCN, making direct CU comparisons invalid.
 
Reactions: french toast

Valantar

Golden Member
Aug 26, 2014
1,792
508
136
What if.....
V10 is not found anywhere, because what you see in these slides & leaks is V11. (Masked as V10.)
With V10 being held behind some tight lips.

And AMD's respin of Polaris is a market grab maneuver at ultra low costs.
(AMD's version of the 8800GT).




Is that a viable rumor?
It's not a rumor at all until/unless people start spreading it around and talking about it. Up until then, it's an idea.

As for the merit of the idea, it seems unlikely. Why present misleading information in confidential closed-door meetings to high level industry partners? That would seriously erode their trust in the company, even if performance matches up to what's presented. I imagine an AMD datacenter sales rep calling up ... say, Google, and telling them "Oh, remember that awesome Vega 10 GPU that we showed you and that you ordered ten thousand of for your datacenter? Yeah, that was Vega 11, only we didn't want to tell you. Sorry. Vega 10 does exist, and it's even better, we just didn't want to tell you about it. We cool?" That wouldn't go over well.

I mean, there's nobody outside of AMD dictating what their codenames mean, so why on earth make a system and subsequently lie about it? If the purpose was to "hide" Big Vega, that's both a risky and impractical way of doing it. And that could easily be done by naming Big Vega 12. Or 13. Or $$$. Or >9000. Or whatever. They're free to codename their products whatever they want, after all.
Guys,

in your die size estimation, you forgot that Maxwell has 4 times more cache than Kepler (cache used by the tile rasterization), taking out space from pure ALUs count. Vega will be the same thus wont scale linearly its ALUs compared to Polaris...

my 3cts
That makes some sense, and would make up some of the ~100mm2 "missing" between the known photos of Vega 10 and an "upscaled" Polaris 10@64 CUs. 100mm2 is a lot of cache, though. I guess a combination of bigger caches and bigger CUs?
In light of the above, that does seem to make sense.

Still, the Fury X clocks in at 8.6tflops at 1050MHz. As such, all it would take for it to reach 12tflops would be a clock speed increase to ~1470MHz - lower than what's rumored for Vega. From the (rather awful) clock scaling of Fiji/GCN (Polaris included) that would reuqire some architectural changes and not just a better process, but it still means that (given that rumors on CU count and clock speed are accurate) there is no IPC gain between Fiji and Vega. Which doesn't exactly jive with AMDs statements that one of the biggest focal points for Vega's development is/has been IPC increases.

Unless Vega's NCU contains fewer SPs than previously? So that 64 CUs add up to less than 4096 SPs? That would of course mean better IPC if tflops performance matches "1500MHz Fiji". That sounds odd to me, but on the other hand Fiji is (as mentioned by someone above) a very imbalanced design, with a "throw more CUs at it" approach to performance. Perhaps they found it would make more sense to rebalance the design, give each CU fewer SPs, increasing cache, adding more TUs and ROPs and such to balance out the design more? I don't even remotely know enough about GPU architectures to do more than ask questions about this, but in my head at least it sounds like something isn't adding up.

On the other hand, could some of the arch improvements be specifically focused on increasing gaming/real-time 3D performance per tflops? After all, that's where Nvidia is beating them the most soundly. So the IPC in terms of tflops might be the same, but real-world gaming (and similar) performance might be better?
From what we know of big Vega, it seems as though there will be 3 consumer parts and I think the fill the Fury, Fury X, and Fury Nano categories. I don't know what Vega 11 ends up getting called, but it seems like a mixed lineup almost doesn't make sense. Sure you can call them Vega Fury or something like that to denote the difference between the old and new Fury, or just call it the Vega, Vega X, Vega Nano and drop Fury completely since it isn't something that I think will get a lot of fond memories down the road. At best it was ahead of its time.
That's what I'm thinking too. RX Vega to replace R9 Fury/Nano.

If small Vega is going to take the x90 slot, it seems like the cut version should probably take the x80 slot and relegate Polaris to x70 and below. A few powers speculated that there won't be an X cards since we've moved from R9 to RX and at some point, it adding more X's just gets ridiculous. No idea how to work Vega into the name without it getting cumbersome though.
Here I have to disagree with you. While adding infinite Xes does get silly at some point, AMD traditionally keeps the same chip on the same numbered tier. As such, it would make sense for Vega 11 to be the RX 590 and 590X (perhaps they'll add some other "step-up" signifier? Revive the "Pro" moniker? That would create confusion with the Radeon Pro lineup, though. Hm), rather than 580 and 590. (Even if I on principle agree that renaming a clock-bump chip refresh on the same naming level a year later is a bad tactic. An RX 480 refresh should be the 570 for it to make any sense.)

Personally, I think it would be cool if they dropped numbers with Vega and played more into the astronomy side of things. Big Vega gets designated as Vega Black Star or something like that. It's a little bit like a Black Edition throwback in a way. So Vega Black Star, Vega Black Star X, and Vega Black Star Nano for the big Vega Cards. Below that the smaller Vega 11 can be labeled Red Star and Red Star X to denote cut and full chips. If you need another rung for an even smaller die, just pick another color like blue or white.
Oh god. Please don't, AMD.
"AMD Radeon RX Vega Black Star X (16GB edition)"?
"Sapphire Radeon RX Vega Black Star X 16GB Nitro+"?
"MSI Radeon RX Vega Black Star X 16GB Gaming X OC"?
"Asus ROG Strix Radeon RX Vega Black Star X 16GB OC"?
No. Thank. You.
They've already established a high-end naming scheme with the Fury series: Radeon R(x) (Product name) (X). No reason to change that after just one generation.

Not to meniton: If they're to stick to the astronomical naming, at least suggest things that exist (such as Polaris, Vega and Navi ...). Red/white/brown dwarfs, neutron stars, black holes, red/yellow/blue giant/supergiant. Plenty to pick from. Or are you suggesting they start a naming scheme based on David Bowie song and album titles? I for one would rather buy a Radeon RX Rebel Rebel or Radeon RX Dancing In The Street than an RX Vega Black Star. Maybe an SFF version called Radeon RX Space Oddity? A really great blower cooler on the Radeon RX Wild is the Wind? A business lineup called Radeon Black Tie White Noise? I think we might be on to something here.
 
Last edited:
Reactions: guachi
Status
Not open for further replies.
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |