AMD Polaris architecture to succeed Graphics Core Next
- Thread starter wachie
- Start date
IMACOMPUTA
Member
Only relatively:
Thank you.
I guess I'm kind of lost then. Whats with all the doom and gloom about Polaris 10? Nvidia comes out with a non hbm2 card between Polaris 10 and Vega, right?
Its a joke, don't get so defensive!
You know, these past few years it seems to have become AMD's defense : Nvidia are evil and bla bla bla. It's their fault they are in this mess. Their competitors are investing in their platform and it's good to see AMD launching initiatives of their own.
Everyone benefits.
Lastly you really are not the one qualified to pull the fanboy card. You are every bit as much of a fanboy.
On both sides with equally dubious arguments.
IMACOMPUTA
Member
Definitely
polarizeme
Member
Jathanately.
Gemüsepizza
Member
If the rumored numbers and the scale in this picture are accurate (and that's a pretty big "if"), this could mean that Vega 10 is about 88% bigger than Polaris 10, so 232mm² for P10 and 437mm² for V10. The problem for AMD is that the nvidia GP104 has a size of 333mm², and nvidia already has a 610mm² chip which they could release next year. So a small comparison:
2016
AMD: 232mm²
nvidia: 333mm² (+44%)
2017
AMD: 437mm²
nvidia: 610mm² (+40%)
Again, this is of course a lot of speculation. And AMD is also using a slightly smaller process. But if this is how things will be, it would be pretty bad for AMD.
IMACOMPUTA
Member
Gemüsepizza;202221511 said:
Thanks for that.
Bigger size is not automatically a positive. One reason why 14nm was mostly being used for mobile SoC so far is due to their limited sizes. The bigger the size the worse the yield, the higher the cost for the chip will be. So in that regard AMD is playing it safe (after having wanted to switch to a smaller node for ages already) while Nvidia gambles.Gemüsepizza;202221511 said:
Bolivar687
Banned
Nobody is defending AMD. Nobody is calling anyone evil. Nobody is pulling fanboys cards. It's just becoming very apparent how frequently and predictably certain users show up to argue away any Nvidia criticisms and downplay positive AMD news. It's gotten to the point where people are starting to draw up Bingo cards. Really.
Nvidia is in a great place - they don't need you guys to be so vigilant or so sensitive. Let's try to have some fun and see how these new architectures stack up to eachother.
AMD dies are usually more dense than Nvidia's (Maxwell was an exception though, so this might not still be true), especially this time due to the difference in nodes. It's more that you expect a dramatic boost over GM100, though. That's what I'm trying to warn you about. I will admit that I didn't know GP104's die size before though.
As an example, Hawaii is 438 mm² while GK110 is 561 mm².
Same could be said about AMD really. Strangely you seem less inclined to point that out. :b
SneakyStephan
Banned
Again? Jesus christ I wish they would stop fucking with their gpu names every gen, it's so annoying
They went from the x800 cards being the high end, to the x900 being the high end, to inventing yet anothe tier with the fury brand (and worse: a 390 being the same card as a 290), then there's the whole rx xxx
It's such a pain in the ass to compare gpus to older ones these days with all the obfuscation tactics.
Nvidia used to pull this shit too back in the day and it was equally obnoxious.
Gemüsepizza;202221511 said:
This is what it comes down to right now, it's all rumors but it's all we have and it's worrying because a lack of competition means more bad pricing in the midrange and high end. I don't want the gpu market to become like the cpu market where amd only compete in the lower end and the low end is the only bracket where there is any value to be had.
I remember how amazing the hd4870 was and how awesome the value of that card was, it was the best bang for buck gpu I ever bought (alongside the radeon 9800 pro) and I want more of that: Price/performance that blows away the previous gen.
It's logically much more likely that the bigger die will be faster than the smaller one. I too said I hope amd manage to exceed expectations based on die sizes , if they can have a crazy efficient gpu architecture with polaris (and later with vega) it's good for everyone.
@ the bolded
While intel's 14nm process is significantly smaller, the difference between tsmc 16nm ff+ (pascal) and samsung's 14nm ff+ (polaris and vega) is basically non existent, it doesn't make sense to expect that to be a factor.
dgrdsv
Member
As I've said I don't get why people think that NV is launching only GP104 this year. For all we know GP106 may be no more behind P10 than P10 itself will be behind GP104. The reality of this "market split" won't hold for more than one quarter and what will happen by the end of the year is NV having 16nm Pascal products from GP107 ($100-200) to GP104 ($400+) while AMD having 14nm Polaris products from P11 ($100-200) to P10 ($300-400) which will basically mean that AMD will loose the higher end of the market while NV will transfer the additional margins from there to battle AMD's marketshare push in lower segments.
This is all in all a bad situation for AMD unless we don't know something about their high end Polaris strategy.
Bolivar687
Banned
Just because Nvidia may compete at the mainstream much sooner than AMD will compete at the high end. However, all of these terms are incredibly ambiguous and we're reading entirely too much into so little information IMO. Still, I've been saying that if the GTX 1070 competes closely with Polaris 10, then I agree AMD will be in trouble, as they would have botched yet another architecture launch opportunity.
dgrdsv
Member
This is something which isn't really viable either if we're going off the die sizes leak we have currently.
P10 = 232mm^2 = ~Hawaii transistor complexity = 390-390X+ performance
GP104 = 333mm^2 = ~GM200 transistor complexity = 980Ti+ performance
390/X and 980Ti do not compete in this generation and for them to compete in the next one either AMD must do some heavy architectural miracles in Polaris or NV must screw up with Pascal updates.
There were rumors of a third GP104 SKU and I've been assuming that it's something akin to "1060Ti" which may definitely go against P10. But for now, with what we know, I'm not expecting P10 to compete with 1070.
Granted all these leaks can easily be false including the die sizes.
Bolivar687
Banned
The cut down Polaris 10 and GTX 1070 are said to share a similar memory interface, VRAM capacity, memory type, and core count. And before anyone brings up the difference between Maxwell cores and the GCN cores which they launched a full year later than, I want to note some of the estimations in here are strongly based on comparisons to AMD'S three year old arcitecture. Which seems presumptuous on a node shrink this big and when we know almost everything about Polaris is new. I'll also reiterate this one last time before launch:
The way these leaked die sizes are being attributed to specific cards is almost completely made up. Not necesarily saying they won't end up being accurate, though.
The way these leaked die sizes are being attributed to specific cards is almost completely made up. Not necesarily saying they won't end up being accurate, though.
As you said, it's not too crazy of an idea to assume the following if Vega does consist of two chips like Polaris does:
P11: 470 series
P10: 480 series
V11: 490 series
V10: whatever fancy name they'll come up with for their halo part
I can see Vega 11 end up in the 300-350mm² range, but what about Vega 10? I don't think AMD are willing to go with a monster die like Fiji right off the bat. They wouldn't even need to because GP100 isn't going to make it into consumer products anytime soon, if ever (something something speculated GP102). Perhaps they're gonna do another Hawaii and manage to come close enough to GP100 in gaming performance. All that DP functionality takes up a huge chunk of die space.
I totally agree with what's been repeated so many times. P10 and GP104 aren't going to compete with eachother. It's basically HD7870 vs GTX 680, which is totally fine if they price these right and AMD can get some mobile/OEM wins and are competitive in the low-mid range. At best I can see P10's fastest SKU punch up against something like a 1060Ti. And as dr_rus also mentioned, GP106 isn't too far off either. Something like this sounds reasonable enough to assume: P11 < GP106 < P10 < GP104 < ...
Sure it sucks for whoever is sitting on current high end cards and want a >50% performance increase. I know in past generations a second in line chip was able to bring a decent performance jump over the previous high end chip, but things are a bit different now. 28nm was around for so long all they could do was go up in size and/or cut DP functionality. GM200 is a 601mm² gaming beast and it's hard to say how much faster the upcoming cards will be. I'm not denying GP104 will be faster though. 14FF/16FF+ is likely to stay around for a quite a while like 28nm did, so neither company is in a rush to show their biggest cards yet.
I could be totally wrong about all this though.
Inspectah_Deck
Member
Amen
If I compare this thread to the Pascal thread, no thats not the case.
I don't understand why NV's 610mm² chip (GP100) is even talked about in this thread. I'd be exceedingly surprised if they ever sold that as a consumer graphics product. It doesn't make sense as such. It seems more likely that they'll have a gaming-focused chip between GP100 and GP104 in terms of size at some point later on.
I've used many video cards from every manufacturer (including 3DFX and Matrox), but it's simply the case that over the past decade NV has, overall, catered significantly better to the needs of the enthusiast consumer than AMD has.
This is utterly ridiculous. Ever since a few years back when Nvidia had the gall to call out consoles for what they are (mid-range products at best even at their respective launches) the prevailing attitude on this forum towards the company has been one of disdain and (often completely unwarranted) negativity. On the other hand, every time AMD messes up, it's "give them another chance" or "don't you want competition??".
I've used many video cards from every manufacturer (including 3DFX and Matrox), but it's simply the case that over the past decade NV has, overall, catered significantly better to the needs of the enthusiast consumer than AMD has.
Bolivar687
Banned
I'm going to give you the benefit of the doubt and assume we're having a miscommunication here, that you think I'm speaking of the forum at large (rather than just PC GAF), where console gamers got butthurt at Nvidia's comments you referenced and they could feasibly be more partial towards the company making their chips.
Because it would be really hard to take you seriously otherwise.
If you truly believe the "prevailing attitude" on GAF is disdain and negativity for Nvidia and forgiveness and patience for AMD, I can go back and link you the Pro Duo thread, the Fury X reveal, the Nano thread, or the half dozen enthusiastic threads with Maxwell and Pascal in the title over the last two years. But I'd rather not because it's the weekend, I'm down the shore at my dad's, and that's the most absurd thing I've heard this week. I'm not denying those things are said but calling them the prevailing attitude is disingenuous. You can even look at the last few pages here, with doubt and concern for the Polaris launch, and the only derision towards Nvidia coming from Maxwell and Kepler owners looking for a reason to switch.
dgrdsv
Member
Your examples are about real product launches though and they actually highlight what Durante's said. The general atmosphere has for some years been like this: NV is stupid they won't be able to compete and AMD will "screw them" (there's a thread named just with this word around here) because they are smart (because of consoles/HSA/GCN/HBM/anything else really); then the product comes out and turns out that it's NV screwing AMD again and again and this is where the opposite happen.
But the thing is that while the second one is about real products and benchmarks which are an objective way to compare the companies and base your opinion upon, the first one is just baseless speculation from people who don't know what they're talking about most of the time - hence why their argumentation is down to drawing bingo cards, not because of what you've said.
Irobot82
Member
Is there more proof of the P10 being that size other than a LinkedIn profile?
Let's look at it another way. Both rumor mills I've seen point to the 1080 and the P10 as having around 2560 cores. How does that work with such a small die size?
It seems like you are concentrating so hard on the die sizes, which on AMD's side is pure conjecture, but we've had data leaks that show the P10 has at least 2304 (probably cut version) most likely 2560.
Edit: Oh I see. The 390 has about that many cored. The fury has 4096. So if it's the same cores at 390 level, shrunk to 14nm with higher core clocks, seems it should still be able to compete at a 980ti level.
Console users = All of AMD folks? Its one thing to call out a small segment, but you even used GAF itself as overtly being on defense for AMD, which calls to mind the "Gaf are Sony fanboys" arguments.
Not to make a mountain out of a moehill, but that's a pretty reaching argument. Those people saying Nvidia were doing PR damage and dismissing consoles with their overt statements at that time are not all the same people who buy AMD, nor is it right to conflate the two.
All this guy is saying is, the cheerleading and defense force can stop for a corp like Nvidia who is huge and doesn't need people defending their honor and besmirched reputation online. Especially to quite the fervor it seems every single time there seems to be some butting of heads between the two big faceless entities.
It's been brought up before that the LinkedIn die size could be pure bogus, but if we were to believe the 36/40CU rumor with a 256-bit memory controller, we should more or less end up around that size.
I hadn't seen this yet, but apparently some leaked scores were posted on GFXBench:
http://videocardz.com/59468/amd-polaris-10-and-11-opengl-benchmarks-spotted
As usual, take it with a grain of salt.
Locuza
Member
It looks like Project F was indeed Polaris 10:
http://videocardz.com/59487/amd-polaris-11-and-10-gpus-pictured
~230 mm²
http://videocardz.com/59487/amd-polaris-11-and-10-gpus-pictured
~230 mm²
230 mm2 is very small for the supposedly better of the two cards. Outside of some breakthrough in computing density apart from the new node, I don't think it'll come close to the performance needed for good VR.
Will be cheap and power efficient, though.
If you are selective for some reason.
Gemüsepizza;202221511 said:
Well, define bad for amd. If it's to win an arms race to find out who can build the fastest possible card, then nvidia will win this year. Amd has stated more than once that these are going to be mainstream cards. Think 970 and 980 level for a better price point and way better power performance. This is exciting because it can dramatically increase the market base of computers powerful enough to run graphics intensive games, and even better in the laptop space it can make fairly thin and light gaming laptops a reality, both in terms of heat and battery life. Since the home pc market is dying in favour of laptops, this has a bigger impact overall.
Next year is when the real super-cards come out. Vega and Volta will be much more powerful, with a more mature process, better memory and larger die sizes.
Volta won't release next year. GP100 consumer variant will.
Isn't the current minimum requirement for VR around 390X/970 performance? I'm paraphrasing here, but AMD said they wanted to bring that performance level to a lower price point. It's starting to become very likely P10 will end up in that performance bracket.
Re VC article: took long enough for a die picture to show up. LinkedIn guy redeemed.
shark sandwich
tenuously links anime, pedophile and incels
If AMD can come somewhat close to Nvidia performance-wise, but do it at a cheaper price, it could work out very well for them. Some of their best products (4000 + 5000 series) did this.Gemüsepizza;202221511 said:
dgrdsv
Member
Are 390X or Fury not good for VR?
I wouldn't be so sure about this. Volta is coming in 2017, that's 100%. The question is will they release it as a GeForce card in 2017 or will it go into supercomputers only.
It'll likely be faster than the 970. Wait for benchmarks instead of comparing die sizes and densities.
michaelius
Banned
More like 50$ cheaper since I don't see AMD releasing those cards below 250$.
And I got no idea what's the point of VR ready cards for everyone marketing when cheapest helmet costs 600$+ accessories.
GhostTrick
Banned
What I see is that it'll be on par with R9 390/390x. Lower CU number but same kind of performances for a far lower TDP.
Well the 390x is probably the worst perf/watt product AMD sell at stock settings (an undervolt can really reduce power usage and not impact performance) so using that as the baseline does give the worst case P10 scenario. It also gives AMD the biggest figure to use in their marketing materials so it seems the most likely to be honest.
If we assume that 2.5x is best case and it is more realistically 2x on average, it only applies to the chip not total board power, the P10 will use GDDR5 modules that use the same power as on the 390x and we use a figure of 90w for the memory tdp that means the 390x die has a TDP of 185 Watts. That would mean a similar performing P10 would be in the region 90 Watts giving a total board TDP of 180 watts. Based on that I think a lower bound for a 150 watt P10 based board would be around the 290 performance level.
I think that is a very conservative lower bound assuming a 150 watt TDP. I am using 150 Watts as that means you can make parts with 1 6pin connector which the OEMs will like, perhaps retail parts will have two so you can have more power headroom but that is pure speculation. It seems safe to say that P10 will as a minimum offer current Hawaii performance in a much lower power envelope but if that is all it offers it does not mesh with statements made by people at AMD as you can already get that performance for around $300.
If the price point for the full P10 is $299 then between 390X and Fury X performance would be pretty near to stock 980Ti performance. I can totally see Fury X performance from the top P10 being realistic and based on the size of the chip I can see something like Pitcairn (7870 and 7850) where the top one is $350 and the cut down one that offers 90% of the performance is $250.
Irobot82
Member
This is a little confusing. The image the show as "Polaris 10" has the exact same marking as Tonga. So, where is the P10 die?
Edit: I see, they are using that rendering marketing material as the gpu and guessing the size off that.
SneakyStephan
Banned
If all they can do is offer r9 390x performance at 180w then all they will have accomplished in power efficiency terms is to match last gen maxwell (gtx980 has 170w TDP at the same performance level)
That would be really really terrible
I'm still thinking most of the rumors we have atm are bullshit, much of it comes from WCCFtech ffs they don't have any sources.
I might have talked too soon. If they can get Fury or 390X performance on such a small die, it might be the greatest product AMD in some time.
If the WCCF article is to be believed and the 1080 will only be slightly more powerful than the 980ti(And based on the way the worded it, it seems like it may even be slower than the 980ti) wouldnt that make Polaris pretty much in the same ballpark as Pascal in terms of target audience? Both cards would be around the same performance-wise as the 980ti with Polaris possibly costing much less.
Well Fury Nano is also in the performance window at 175 Watts too so its not like AMD cannot do it at 28nm. The idea that there will be no perf/watt improvement over Fury Nano, even if using GDDR5 rather than HBM is daft.
Ignoring the rumours though lets just look at the last node shrink. The 7870 was 212mm^2, 2.8 Billion transistors, had a TDP of 175 watts and had 1280 SPs vs the 6970 that was 389mm^2, 2.64 Billion transistors, had a TDP of 250 Watts and had 1538 Sps yet at launch the 7870 was faster than the 6970 and after drivers matured for GCN that gap increased to >30% faster. The 7870 was also very close (and did eventually exceed) the GTX580 performance which was a 520mm^2 GPU.
For the P10 if AMD is targeting a similar 10-20% performance gain over Hawaii that puts it bang into Fury X performance territory. At 4k the Fury X is 25% faster than the 390X and neck and neck with the 980Ti, at 1440p the Fury X is 21% faster than the 390X and neck and neck with the 980Ti, at 1080p the Fury X is 16% faster than the 390X and the 980Ti is 11% faster than the Fury X. Latest high end TPU review.
So based on very rough perf/watt improvements Fury X performance in a 232mm^2 GPU with a TDP in the 150 watt rage seems very possible, based on the last node shrink where a node was skipped (32nm) this sort of performance uplift also seems in line. None of the rumours are outlandish, all of it adds up and all of it has been done in the past so it would be nothing new and it is far from exceptional.
It would not surprise me for AMD to launch the 480 and 480X at $249 and $349 offering 390X performance and Fury X performance respectively. Perhaps if they really want market share they might offer something like $229 and $299 respectively instead but the pricing depends on so many factors that those are just guesses based on previous launches.
Saintruski
Unconfirmed Member
I'd rather buy a Arctic cat or skidoo...
(I wonder how many people will understand this)
(I wonder how many people will understand this)
Got it!
Supposed benchmarks for the Polaris 10:
https://gfxbench.com/compare.jsp?be...wname1=AMD+67DF:C4&D2=AMD+Radeon+(TM)+R9+390X
Take with a HUUUUGE grain of salt.
If they are true it's worse than I though.
https://gfxbench.com/compare.jsp?be...wname1=AMD+67DF:C4&D2=AMD+Radeon+(TM)+R9+390X
Take with a HUUUUGE grain of salt.
If they are true it's worse than I though.
What am I looking at?
Similar threads
- Poll