AMD Vega 10 - Vega 11 and Vega 20 GPUs Mentioned

Published by

Click here to post a comment for AMD Vega 10 - Vega 11 and Vega 20 GPUs Mentioned on our message forum
https://forums.guru3d.com/data/avatars/m/198/198862.jpg
Vega GPU's will heat things up big time. I just feel AMD should've done everything to release them this year.
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
The article is kind of confusing because there is basically two rumors and one piece of news. From what I can gather from reading everything and correct me if I'm wrong: The CTO only said that Vega is coming 2017, he didn't mention 10/11/20 or the specs of any of those. VideoCardz has a leak, from a source they claim has given them good info in the past that said the following: Vega 10: has 64 CUs, 24TF-16FP & 512GB/s. TDP is 225w and dual Vega 10 Q2 2017. VEGA 20: which is a 32GB 1TB/s 150w part with the same 64 CU's on 7nm. Vega 11: Replaces Polaris 10 As for the 6144 Vega 11 part of the article, that seems to come from Hilbert directly. Pretty sure during the AMA Raja said that Vega 11 was the big Vega. So it replacing Polaris 10 doesn't make much sense to me, unless it's an entire line up coming later. I'm inclined to think that Hilbert's prediction is more accurate - but that would also directly conflict with the VideoCardz source so idk.
https://forums.guru3d.com/data/avatars/m/16/16662.jpg
Administrator
Still writing/rewriting the article.
https://forums.guru3d.com/data/avatars/m/63/63215.jpg
Considering what was in the investors presentation, I fully believe Vega will not come this year and instead will indeed be 1st 1/2 of 2017.
https://forums.guru3d.com/data/avatars/m/115/115462.jpg
All of these sound great, but they should have come out much, much sooner. Next year is simply too late. By that time, even if the cards are really good, nvidia will likely have an answer, and I'm not even thinking about 1080Ti. In such a time frame they can as well do a refresh of their whole line by spring 2017 and we'll be back where we are now. AMD's best chance is to be even better in DX12 (and we can see nvidia struggling there), but at the same time for DX12 to spread out more. We'll see, I'm certainly looking forward to Vega 11, if that's indeed the big boy.
data/avatar/default/avatar20.webp
Ahhh... 200 million VR HMD in 2020... this clearly points to some brain issues with AMD and many others. This is a dreamland, which will NOT happen. Are they like Apple fanboys and hype-press, who expected 80 million iWatch sales? 24TF for Vega 10 is also a total BS.
https://forums.guru3d.com/data/avatars/m/242/242471.jpg
No its not, 32bit is ~12TFlops..
https://forums.guru3d.com/data/avatars/m/63/63215.jpg
No its not, 32bit is ~12TFlops..
Yeah. Which would mean they're aiming roughly for twice the performance of FuryX for that card.
https://forums.guru3d.com/data/avatars/m/237/237771.jpg
No its not, 32bit is ~12TFlops..
That could be nice at $450-500 US. If it is on GDDR5x and not HBM2 it may happen. I don't see a HBM2 part coming in under $700 MSRP.
Yeah. Which would mean they're aiming roughly for twice the performance of Fury for that card.
Fixed that for you FuryX was just under 9 Tflop.
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
That could be nice at $450-500 US. If it is on GDDR5x and not HBM2 it may happen. I don't see a HBM2 part coming in under $700 MSRP.
I don't see why they couldn't price it at $650 with HBM2. This isn't launching till next year and they were able to ship the Fury X at $650 six months after HBM1 went volume. A 12Tflop part at $650 with HBM2 would definitely beat a 1080 in the majority of titles and could potentially make up for the 6-8 month head start Nvidia had over AMD in terms of marketshare. Idk how much they'd be making on each card though.
https://forums.guru3d.com/data/avatars/m/63/63215.jpg
Fixed that for you FuryX was just under 9 Tflop.
Bit of trivia for you Loophole, what is the FuryX's Single-precision Tflop rating?
https://forums.guru3d.com/data/avatars/m/259/259654.jpg
To be honest, if these GPUs hold their ground as well as Tahiti/Hawaii and even Fiji seem to, I have no problem getting one around May and keeping it for 3-5 years. They are the center of the Scorpio and the next Playstation after all, it's not like they will be going anywhere.
https://forums.guru3d.com/data/avatars/m/237/237771.jpg
Bit of trivia for you Loophole, what is the FuryX's Single-precision Tflop rating?
8.6 Tflop 12 Tflop is only ~40% faster. The Fiji Pro (Fury) is rated at 7.2 Tflop.
https://forums.guru3d.com/data/avatars/m/63/63215.jpg
8.6 Tflop 12 Tflop is only ~40% faster. The Fiji Pro (Fury) is rated at 7.2 Tflop.
valid.
https://forums.guru3d.com/data/avatars/m/242/242471.jpg
But don't all gpu makers name SinglePrecision FP in 32bit anyway, idk 780TI 5.3Tflop, FuryX 8.6tflops, Titan XP 11tflops. Dunno why AMD wanted to mention this now,.. I guess to say Vega is also capable of half precision just like NV Pascal GP100
https://forums.guru3d.com/data/avatars/m/237/237771.jpg
FuryX is closest. We're talking 8.6Tflop vs 24Tflop. 12Tflop is the estimated 32bit precision by halving the 16bit number.
That is half precision performance. There are no concrete numbers on Fiji XT half precision. But is should be assumed to be 17.2 Tflop. That is still 40%. You said roughly twice as fast as Fiji XT and that is wrong. There really is no reason for you to continue to play this semantics game.
But don't all gpu makes name SP in FP32, dunno why AMD wanted to mention this,.. I guess to say Vega is also capable of half precision just like NV Pascal GP100
I don't think he is getting that.
https://forums.guru3d.com/data/avatars/m/63/63215.jpg
That is half precision performance. There are no concrete numbers on Fiji XT half precision. But is should be assumed to be 17.2 Tflop. That is still 40%. You said roughly twice as fast as Fiji XT and that is wrong. There really is no reason for you to continue to play this semantics game. I don't think he is getting that.
Sorry, numbers got mixed-up. 32bit single, 64bit double, 16bit half.
But don't all gpu makers name SinglePrecision FP in 32bit anyway, idk 780TI 5.3Tflop, FuryX 8.6tflops, Titan XP 11tflops. Dunno why AMD wanted to mention this now,.. I guess to say Vega is also capable of half precision just like NV Pascal GP100
Does it mean that Vega 10 has phenomenal half-precision performance? I can only find this source for Pascal half precision performance and it's not that impressive, we're talking Gflops rather than Tflops; https://en.wikipedia.org/wiki/GeForce_10_series
https://forums.guru3d.com/data/avatars/m/237/237771.jpg
Okay so if we are assuming 12Tflop performance and we also assume that vega will have the same per-core performance as Polaris the this card will be clocked at what 1320-1360MHz?
https://forums.guru3d.com/data/avatars/m/232/232079.jpg
Okay so if we are assuming 12Tflop performance and we also assume that vega will have the same per-core performance as Polaris the this card will be clocked at what 1320-1360MHz?
Pffft to me AMD is a Total-Flop!:rpg:
https://forums.guru3d.com/data/avatars/m/261/261885.jpg
I don't care much about teraflops and bandwidth. Just make a card which can do 4K60FPS for 500-600$.