Soldato
- Joined
- 30 Nov 2011
- Posts
- 11,356
If Vega has HBM2 and Polaris has large amounts of GDDR5 then that's a power saving right there, if they are also waiting for the process to mature then it's not inconceivable
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD Polaris architecture – GCN 4.0
- Thread starter Boomstick777
- Start date
If Vega has HBM2 and Polaris has large amounts of GDDR5 then that's a power saving right there, if they are also waiting for the process to mature then it's not inconceivable
Ok so forgetting all the mixups over the naming scheme, (have they really gone with the bigger number being the smaller chip). I get the feeling that AMD might have pulled a blinder here.
Here is the dilemma, you are going to launch a new set of GPU's but you don't want to show the opposition just how fast they are, (now most of us have expected the bigger chips from this first round of new GPU's to beat the currant 980ti/FuryX, just like I'm sure both AMD and NVidia have.) so you show a benchmark that is locked at 60fps, one that the currant top tier cards can just about manage to max out and voila you have shown off your new card, shown that it is at least as fast as the currant crop, without actually showing just how fast it actually is.
It could be a genius move by AMD, but then returning to the naming stuff, when you have two of your top people saying different things about which card is which, maybe AMD just blindly stumbled into this benchmarking piece of genius by mistake.
The problem with this is that it's also confusing to your customers. Your early adopters now have no idea of what they want or how fast it will be. It seems like the info is a damp squib because we still don't know how fast these things are.
AMD still need to learn how to market their products in an exciting fashion. Yes, I know it's GDC, but even so, it's streamed to the world, non-developer people are interested and watching. There's nothing to stop AMD making this exciting instead of dull. A little showmanship would go a long way.
Soldato
The polaris 10 part was matching a 1.38ghz 980TI and the fury (stock), the fury x gets better performance in 2560x1440.
But the other side of things is that all we can tell from that demo is that Polaris 10 can match those two and maintain 60FPS at that resolution.
We know nothing about its upper end from that demonstration since it was frame capped. Just that it can match the aforementioned parts. AMD are keeping its top end under wraps atm along with all other details. But we know it has Fury levels of performance when run in dx12.
I guess the other thing to consider (though I don't want to get my hopes up too much) is that this could have been cut down Polaris 10.
Obviously Polaris 10 is going to have an X and non-X version, so what if Hitman was running on the cut down?
Soldato
We don't know, that is the thing. AMD showed us this tid bit so we can guage its performance. But in terms of its max performance and what specs it has, it is all conjecture.
For all we know it could be cut down polaris 10, it was said that the AMD engineers had rather smug grins when asked about its performance at CES. So it could run at 1.5x this fps when uncapped on the full part.
The Polaris 10 and 11 naming scheme is confusing. I guess they are following Nvidia's naming convention where GM200 was the higher card and GM204 was the lower card.
Let's hope Polaris 10 comes in two variants (X and Pro?) and is at least as fast as a TitanX with 8GB+ memory.
I haven't seen a 60 fps version of the Hitman demo yet but if it holds 60 fps at max detail then it's potentially a monster card for a sub £400 price. AMD could have shown the max fps for the card but I reckon this time they are holding the performance numbers closer to their chest.
Let's hope Polaris 10 comes in two variants (X and Pro?) and is at least as fast as a TitanX with 8GB+ memory.
I haven't seen a 60 fps version of the Hitman demo yet but if it holds 60 fps at max detail then it's potentially a monster card for a sub £400 price. AMD could have shown the max fps for the card but I reckon this time they are holding the performance numbers closer to their chest.
Last edited:
Soldato
A short demonstration was shown of it running hitman at 2560x1440p60 frame capped with constant fps on the capsaicin stream last night. it was only a short demonstraction but means it beats a stock titan x considering that makes it match a 1.38ghz 980Ti.
But yeah, i think they are keeping the specs for polaris 10 and 11 under wraps till nvidia decide to show something or until release date.
Soldato
How on earth would a new GPU be twice as fast as the previous gen stuff? they would have to have at least 8192 shaders (more like over 10000) and god know what else to do that. Not going to happen this year, or next year TBH.
Also this constant "it needs at least 8gb memory" is annoying too. Name a game that is playable on current hardware that uses more than 4gb? yes some of the top games use 4-6gb of memory at 4-5k but they also run at sub 20fps. Whats the point!?! At realistic (60fps min, not average and 1080p or 1440p) settings games still use under 4gb.
The shrink to 14nm and the extra power available within the envelope by going to HBM will get them a lot of the way there. I think there might be a mad top-end Vega in sight by end of Q4 2016
Associate
He didn't said twice as fast, just as fast as 980Ti, which is possible as they can fit that performance in less than half sized GPU now.
Also for the top cards comparison:
Fury X transistor count 8.9B
980Ti transistor count 8B
Big Polaris (Vega) transistor count ~16-18B
Big Pascal transistor count ~17B
WE should mention newer achitecture with improvements as well.
Agreed. I think that both AMD and Nvidia have been saving up big architectural changes because they've been waiting on the process shrink/transposer/HBM to take full advantage.
They've had longer to get ready for this process shrink than any other shrink in history, and I think they want to make the most of it after being stuck in 28nm limbo for years.
Soldato
The main issue with GCN is that it has always had higher theoretical performance than the competitors parts. But AMD has not been able to make full use of that performance till Async.
Now GCN4 has certain features that help fix this and should improve Shader utilisation overall, allowing the parts to work far closer to their theoretical performance in single threaded and async situations.
So even if Polaris 10 was to have fewer shaders than fiji, it could still beat it due to more efficient and better utilised shaders. The only times it would be worse than fiji is in very async heavy games, which Deus Ex should be.
Soldato
ninja edit i think, im sure he said twice as fast originally lol
AMD are looking rly strong in dx12 stuff, i think thats just going to get better too
and thats great, forward thinking etc
but for now atleast it needs to perform with games built on old dx too,
people play a ton of old coded games, and cheaply ported games, we dont all live in dx12 land yet
i would like to hear them talk more about that and what they can do about it
The 10/11 switch makes it pretty obvious they have nothing interesting until Vega.
Let everyone think 11 will be something tasty for a while to keep interest up, then BAM let it slip that they reversed the names and we already saw 11. 60fps in BF4, yawn, had that for a long time already.
Same goes for Li'l Pasquale. At least Nvidia didn't put out a demo of GP100 and call it GP104.
Let everyone think 11 will be something tasty for a while to keep interest up, then BAM let it slip that they reversed the names and we already saw 11. 60fps in BF4, yawn, had that for a long time already.
Same goes for Li'l Pasquale. At least Nvidia didn't put out a demo of GP100 and call it GP104.
Associate
If this is the case... they better be cheap. Like £150 max for a P11.
The 290X/390X has been with us since Oct 2013. If they price the P11 at £200+, they can shove it
+1
Though a more realistic, but still reasonable pricing would be for P11 to replace the current 280/280X. So in the range of £150 to £200. I wonder if they'll do 2 versions with different amounts of memory. Perhaps a 3GB and a 4GB? Or even a 4GB and an 8GB? If so, then the 950/960 are doomed and Nvidia will lose out on the semi-mid-range customers, unless they have a card to compete in that price bracket.
Also who keeps on saying that the P10 is on par with 980ti or weaker than a FuryX? Current benchmarks of Hitman 1440p DX12 show that Fury X averages mid 60s at best while still occasionally dropping below 60. That is if current benchmarks are to be believed. Digital Foundry have yet to do their benchmarks, but their one video so far has a 970 at 1440p max DX12, fluctuating between ~35-~55fps.
As I said, AMD played it smart and showed Polaris 10 maintaining solid 60, thus implying it is capable of more. Whoever had the idea that AMD isn't quite showing their full hand, I feel like they might be on to something. Now I'm actually a bit more excited.
I still probably won't buy a thing till end of 2016/early 2017 though. I want to see what Nvidia has too and what both companies 'big' hitters are.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 30,185
Vega isn't Polaris exactly, from what they've said. It's an argchitectural upgrade. They're claiming significantly higher performance per watt than Polaris, and HBM2 won't bring much in the way of performance per watt, if anything.
Something others seem to have missed is that AMD have stated that Polaris 10 & 11 are not the only chips - others will follow.
They also stated that Polaris has been made compatible with GDDR5 and HBM.
If the images of Polaris 10 & 11 chips in the slide were scaled somewhat correctly, then Polaris 10 must be around 350mm2 or slightly under. That's fairly big, and no surprise that it apparently matches or beats Fury X / 980Ti. Koduri hinted that the big Polaris would be shown later (with HBM).
Given the significant reduction in chip sizes with 14nmFF, it shouldn't be difficult for them to have 6 stacks (or maybe even 8 stacks) of HBM1, and therefore have a 6GB or 8GB Polaris HBM1 card. Tbh I don't even think it's needed in terms of capacity ... but I suspect the extra bandwidth is absolutely necessary.
One other thing ... the 2018 architecture. It'll be very interesting to see what that memory turns out to be. I assume it'll be either Wide IO or their own proprietary standard. If they need more bandwidth that early, then HBM isn't gonna work ... too much power consumption. Also very interesting that NVIDIA seem to have abandoned HMC entirely (no surprise), and will be keeping HBM2 for Volta, probably until 2020. That would be a major divergence.
Something others seem to have missed is that AMD have stated that Polaris 10 & 11 are not the only chips - others will follow.
They also stated that Polaris has been made compatible with GDDR5 and HBM.
If the images of Polaris 10 & 11 chips in the slide were scaled somewhat correctly, then Polaris 10 must be around 350mm2 or slightly under. That's fairly big, and no surprise that it apparently matches or beats Fury X / 980Ti. Koduri hinted that the big Polaris would be shown later (with HBM).
Given the significant reduction in chip sizes with 14nmFF, it shouldn't be difficult for them to have 6 stacks (or maybe even 8 stacks) of HBM1, and therefore have a 6GB or 8GB Polaris HBM1 card. Tbh I don't even think it's needed in terms of capacity ... but I suspect the extra bandwidth is absolutely necessary.
One other thing ... the 2018 architecture. It'll be very interesting to see what that memory turns out to be. I assume it'll be either Wide IO or their own proprietary standard. If they need more bandwidth that early, then HBM isn't gonna work ... too much power consumption. Also very interesting that NVIDIA seem to have abandoned HMC entirely (no surprise), and will be keeping HBM2 for Volta, probably until 2020. That would be a major divergence.
Last edited: