Full AMD Polaris 10 GPU has 2304 Stream Processors

[tviceman]

GloFo sinks another AMD ship. They have got to get Zen/Vega away from them ASAP.

On the contrary, pro-AMD forum posters here seem to think P10 is actually a better jump in most metrics vs. 28nm GCN parts than GP104 is vs. Maxwell parts which means the process is not to blame. http://forums.anandtech.com/showpost.php?p=38329784&postcount=58 Of course, he seems to think P10 is the spiritual successor to Tonga, while I think P10 is the successor to Pitcairn....

Anyways, I don't buy the excuse GloFo is to blame. 2 months ago all pro AMD posters said GloFo had better manufacturing capabilities than TSMC and 3 months ago a shrunk down Fiji would be only 20mm2 larger than P10. If P10 had come out with 390x performance and 120 watt power draw, would people be praising AMD for their amazing engineering or GLoFo for their ultra advanced fab process? Why is it that when AMD doesn't deliver it's not their fault? Aside from Pitcairn and Bonaire, GCN has been a power hungry architecture since it's first introduction. What were the excuses for Fiji and Hawaii having excessively high power draw and low headroom (aside from GCN 1.0 having good headroom)? Fiji had the benefit of lower-power, super-fast HBM and still couldn't match GM200 in perf/w.

Perhaps, as many have been saying over and over again, with Nvidia's R&D budget climbing higher than AMD's 3+ years ago, and with Nvidia having fewer projects in development, perhaps it's just a matter of AMD having less resources to spec, build, test, rebuild, retest, etc. than Nvidia and that is coming to full fruition before our very eyes. Chips are getting harder to build and AMD is spending less on R&D.... hmmm.....

 

[Headfoot]

In the land of team-based discussion and fanboyism everything is black or white.

I think it's perfectly fair, and most likely, that both the process and the architecture have failings and successes. Clearly the GF process is dense, and yielding fairly well. Clearly, TSMC have much, much more experience manufacturing modern GPUs. I very much doubt GF's process for GPUs is to TSMC's level. I doubt anyone in the entire industry is as good at GPUs as TSMC. I also think its perfectly likely that nVidia sent more engineers for more time to sit and work with the TSMC engineers due to their proportionately larger R&D budget. It isn't either or.

 

[tviceman]

In the land of team-based discussion and fanboyism everything is black or white.

I think it's perfectly fair, and most likely, that both the process and the architecture have failings and successes. Clearly the GF process is dense, and yielding fairly well. Clearly, TSMC have much, much more experience manufacturing modern GPUs. I very much doubt GF's process for GPUs is to TSMC's level. I doubt anyone in the entire industry is as good at GPUs as TSMC. I also think its perfectly likely that nVidia sent more engineers for more time to sit and work with the TSMC engineers due to their proportionately larger R&D budget. It isn't either or.

It is a fair assessment to look at the entire picture. It seems like many people on this forum think they can run AMD, make decisions about chip design, and execute product development better than AMD can. There is always 101 excuses when products don't meet the level 11 hype generated on this (and other) forums.

To me, Polaris 10 seems like the normal evolution of Pitcairn and GCN given AMD's decreased R&D, spread out product development, and recent M.O. of pushing out chips at maximum speeds and very little headroom.

 

[Silverforce11]

Looks like a 490 is coming soon, with GDDR5! That cannot be Vega (HBM2).

 

[24601]

Looks like a 490 is coming soon, with GDDR5! That cannot be Vega (HBM2).

Sounds like a dual-gpu part tbh.

 

[JDG1980]

Sounds like a dual-gpu part tbh.

But if that were the case then you'd expect it to have 16GB (2x 8GB).

Also, AMD has released a chart describing their naming scheme, and the diagram indicates that a '9'-series card is supposed to have a memory bus wider than 256 bits.

 

[NostaSeronx]

But if that were the case then you'd expect it to have 16GB (2x 8GB).

The 16 GB model is flash-locked. :sneaky: (8 Gigabyte = 16 x 4 Gigabit or 8 x 8 Gigabit, thus 16 Gigabyte = 32 x 4 Gigabit or 16 x 8 Gigabit.) ((Now was the choice to use 4 Gigabit price or a TDP measure to keep to dies under 275-300W TDP.))

 

[sirmo]

Looks like a 490 is coming soon, with GDDR5! That cannot be Vega (HBM2).

There are two Vegas.. Vega 10 and Vega 11.. http://www.anandtech.com/show/10145/amd-unveils-gpu-architecture-roadmap-after-polaris-comes-vega

Vega 10 is probably 490 and 495.. with GDDR5 (512-bit bus?) and Vega 11 HBM2 Fury and Fury X type cards.

 

[icy1007]

For those who want to believe PR, let me test your logic..

NVIDIA says the 1070 has 64 ROPs. Is this the complete truth?

Look at the diagram and look at the actual test result of Rasterizer performance.

^ If you know anything about NV's architecture layout, you would have quickly realized Rasterizers are within a GPC cluster, if it's cut, bye bye ROPs.

Look at it's fillrate performance:

Ohh look at that! Nowhere near the 1080 with full 64 ROPs. It looks to be missing quite a few, like it's only got 48 ROPS usable.

What a coincidence, each GPC has 16 ROPs, four for the full GP104 equates to 64, 3 for the 1070 equates to 48 ROPs.

Do you trust AMD or NV PR?

Need I remind you, 970 fiasco?

Or recently with AMD, 150W RX 480! lol

The ROPs are not apart of the GPC. You can count all 64 ROPs in that image you posted. They are the dark blue squares between the GPCs and the L2 cache. The GTX 1070 has 64 ROPs.

 

[zentan]

Amazing if true...a 192 GB/s bandwidth GTX 1070 destroying the 256 gb/s bandwidth RX 480 by 50%. The GTX 1070 having worse specs on paper make it look all the more impressive in comparison.

This argument just cuts through all the "48ROP" and "6GB GDDR5 effective" conspiracies. :thumbsup:

 

[Silverforce11]

It's actually not a conspiracy if the 1070's fillrate is so far below the 1080. But believe whatever you want.

 

[itsmydamnation]

There are two Vegas.. Vega 10 and Vega 11.. http://www.anandtech.com/show/10145/amd-unveils-gpu-architecture-roadmap-after-polaris-comes-vega

Vega 10 is probably 490 and 495.. with GDDR5 (512-bit bus?) and Vega 11 HBM2 Fury and Fury X type cards.

Most Guesses are expecting the smaller of the two (vega 10) to be 4096 shaders with HBM and be the part with all the extra PCI-E/GMI interface to go in the HPC SOC. With Vega 11 being even bigger and thus using HBM as well.

Given that a 4096/256/64 part with HBM should be under ~400mm excluding the extra GMI interface size, it would be pretty disappointing to not see a bigger chip even if it doesn't launch until mid 17.

 

[Samwell]

It's actually not a conspiracy if the 1070's fillrate is so far below the 1080. But believe whatever you want.

Maxwells Pixelfillrate is depending on SMMs as is Pascals, but ROPs are not locked to GPCs. 980TIs Fillrate is as high as if it had 88 ROPs instead of 96. You see the same with 1070 and with every Maxwell/Pascalchip. More ROPs can still help with MSAA.

 

[Headfoot]

This argument just cuts through all the "48ROP" and "6GB GDDR5 effective" conspiracies. :thumbsup:

uhhh. No. It doesn't. Memory bandwidth is not the same thing as pixel fillrate. That argument discredits the 6GB effective argument, but it does not discredit the 48 ROP argument.

 

[Rvenger]

This argument just cuts through all the "48ROP" and "6GB GDDR5 effective" conspiracies. :thumbsup:

A partially enabled GPC would have to come into play in order to have this effect.

 

[ElFenix]

Of course, he seems to think P10 is the spiritual successor to Tonga, while I think P10 is the successor to Pitcairn....

 

[Sweepr]

WCCFTech thinks RX 490 is dual GPU Polaris 10:

http://wccftech.com/amd-rx-490-dual-gpu

Makes sense as a stop gap solution till Vega.

 

[jj109]

If the 25% ROPs are disabled, you should be able to easily run the memory allocation test and see that only the lower 6GB are full speed. Since Pascal is evidently just a die shrink of Maxwell (lol), the ROP and memory bus structure should be the same right?

Also dual GPU P10, single or dual card, is a stupid idea. Non-AFR multi-GPU mode will be slower than a GTX 1070 due to inherently poor scaling, and AFR mode will have the frame rate of a GTX 1070 but double the input latency due to the extra frame buffer required for AFR scaling.

 

[tviceman]

The die size is nearly the same as pitcairn, it has the same bus width as pitcairn (assuming Tonga had a magical 384-bit bus) and it is supposed to be the same TDP as Pitcairn.

 

[RussianSensation]

The die size is nearly the same as pitcairn, it has the same bus width as pitcairn (assuming Tonga had a magical 384-bit bus) and it is supposed to be the same TDP as Pitcairn.

You didn't understand what he is saying. Even though it is a spiritual successor to Pitcairn, it is also an R9 380 successor in the market segment/tier at $199-239. It will also be a Pitcairn R7 370 market tier successor in the $149-179 product segments with RX 470.

R9 380 -> RX 480 (replacement tier)
R7 370 -> RX 470 (replacement tier)

This is similar to GTX 970-> 1070, GTX 980 -> 1080, GTX 960 -> 1060. The naming scheme by NV and AMD could not make it more clearer what replaces what. It is also why AMD compared R9 380 to RX 480 in their marketing slides.

Because of HardOCP and PC gamers not understanding the AMD launch strategy, initially too much wrong information was spread about P10 being a Hawaii replacement. It is not and never was.

That means you are both correct. Engineering wise, there is no Tonga replacement this round but Tonga was likely a test vehicle for future architectural features (memory compression, UVD, pixel fillrate efficiency, etc.) It's likely Tonga will never have a replacement from this point of view.

Even though GTX 960 was basically half of a 980, 1060 isn't half of a 1080. We would still consider 1060 a 960 replacement even though based on the engineering comparison it now has 48 ROPs instead of 32.

While RX 480 isn't as good as Pitcairn, NV raised 960's price from $199 to $249-299 when moving to the 1060.
http://www.anandtech.com/show/8923/nvidia-launches-geforce-gtx-960

7870 was much closer to 680 than 480 is to the 1080 but 7870 came out at $349. It seems 480 was more in-line with a 7850 level product line and the pricing also reflects that. The biggest issue for Polaris 10 is its terrible power usage given the 14nm mode and supposedly 2-2.8X perf/watt improvements that we never saw. It sets a terrible precedent for Vega. It means their Vega chip should use more power and be a larger chip to be able to match a 1080...unless Vega has major changes to its architecture and/or 14nm node matured dramatically by Q1 2017.

 

[ElFenix]

The die size is nearly the same as pitcairn, it has the same bus width as pitcairn (assuming Tonga had a magical 384-bit bus) and it is supposed to be the same TDP as Pitcairn.

That's not an argument as to why it can't be a successor to both.

It has roughly the same number of xtors as Tonga (which to me makes it interesting to see what level of generational change AMD is getting) and was launched into the same price range. Seems like a match to me.

And afaik no Tonga had this "magical"* 384 bit bus exposed.





*Why the need to be pejorative?

 

[gamervivek]

WCCFTech thinks RX 490 is dual GPU Polaris 10:

http://wccftech.com/amd-rx-490-dual-gpu

Makes sense as a stop gap solution till Vega.

The wccftech article is a joke. This howler,

Here is what was explicitly stated on one of the shipping manifests: the Baffin XT GPU (aka Polaris 10) was the C98 variant.

​

And he didn't mean Polaris 11 there, since in the very next line he says,

In fact both C94 and C98 boards represent iterations of the RX 480 (namely 4GB and 8GB).

Finally he links C99x board to dual Polaris 10 despite the fact that the second digit identifies the chip used.

Zauba shipping manifests for Hawaii were C676, C675 and C673 for xt, pro and x2 respectively and more recently,

C882 - Fiji nano
C888 - Fiji x2 (Gemini)
C880 - Fiji XT

So C98x and C99x are definitely not the same chip, C99x could be a polaris 10 board if C99x was for Polaris 10 chips but that isn't clear either. He could've linked the new prices on zauba for the cards,

 

[JDG1980]

The biggest issue for Polaris 10 is its terrible power usage given the 14nm mode and supposedly 2-2.8X perf/watt improvements that we never saw. It sets a terrible precedent for Vega. It means their Vega chip should use more power and be a larger chip to be able to match a 1080...unless Vega has major changes to its architecture and/or 14nm node matured dramatically by Q1 2017.

What we got from RX 480 in terms of power usage was certainly disappointing. However, there's good reason to think it's an issue that can be fixed even in Polaris without a full redesign.

First of all, the GloFo 14LPP process is still very immature at this point; as it improves, AMD should be able to tweak things so that the new chips run on lower voltages. Even the existing cards do fine on 1050mV for everyone who tried it, and that alone lowers power usage by 20-30W, and improves performance at the same time by reducing the amount of throttling from hitting the power limit.

Secondly, a non-negligible amount of the power draw for Polaris was due to the high speed of the GDDR5 clock (2000 MHz). This isn't just speculation; AMD specifically mentioned this as a culprit in their statement about PCIe power compliance. When GDDR5X becomes widely available and cheap, it could dramatically reduce power usage for P10 while simultaneously increasing performance (because P10 is memory bottlenecked at 256GB/sec). That makes the perf/watt numbers look a lot better.

Vega will be using HBM2, so that alone provides some measure of added perf/watt efficiency (as we saw with the Fiji cards, especially the Nano). And by the time it's ready, GloFo should have the worst kinks worked out of their process, so hopefully AMD won't have to cram enough volts through their chips to kill an elephant. All this should make Vega's perf/watt much better than what we see with RX 480 even if architectural improvements are discounted completely.

 

[RussianSensation]

All this should make Vega's perf/watt much better than what we see with RX 480 even if architectural improvements are discounted completely.

It's not enough because of how far behind the RX480 is in relation to the GTX1070/1080.

AIB 1080 is 2X faster than an RX 480 @ 1440p/4K, while maintaining 210W typical power usage with a peak of 224W. AMD would need a full generational leap from P10 to Vega to come close to GTX1080 if there were to target ~ 320-330mm2 die within the same 210-225W power envelope.

My calculation using typical gaming power usage:

1440p
Palit 1080 = 100% / 210W = 0.476 perf/watt rating
RX 480 = 51% / 163W = 0.313 perf/watt rating

Palit 1080 has a 0.476 / 0.313 = 52% perf/watt advantage

But it's even worse for the reference 1080 against reference RX 480 where the perf/watt advantage goes up to almost 80% in favour of NV.

You actually believe AMD will improve perf/watt by 50-80% from Polaris 10 with Vega? After their performance and perf/watt misrepresentation with Fury X and Polaris 10, I highly doubt it.

I maintain that AMD will need a larger die size and more power usage to match a GTX1080 with Vega 10. Their only way to compete will be on price or price/performance. This means GTX1080Ti/Titan P should have 0 competition this generation again. That's why I also predict a 2nd consecutive generation of NV selling GTX570 cut-down flagship under the x80Ti brand (980Ti and soon GTX1080Ti). I will be pleasantly surprised if NV actually releases a 3840 CC / 96 ROP / 240 TMU 384-bit GTX1080Ti. If they do, and Pascal scales almost linearly in GPU demanding games, then that's another 1.5X increase over the 1080. AMD will have no chance. To me right now this is AMD's HD2000 series generation unless I see some major changes. What masks AMD's engineering failure is RX 480's $199-239 price. The power usage should have been 110-120W with R9 390X performance, not 163-167W with R9 390 level of performance.

Go back to HD4000 series. When was the last time AMD actually had an significantly inferior $200-250 level chip? It hasn't happened until now. GTX1060 is not only going to be more power efficient but also faster. In fact, one could argue that in the last 5 years, that AMD actually had a superior product line in the $200-250 space. While AMD's cards were less efficient, they were at least as fast or faster. This time AMD's 1060 competitor is both slower and less efficient. Since the $200 RX 480 4GB is MIA and was released in limited quantities, how do they expect to sell RX 480 for $240-270 when GTX1060 is a $250-300 card? AMD is blowing it this time.

 

[flopper]

It's not enough because of how far behind the RX480 is in relation to the GTX1070/1080.



You actually believe AMD will improve perf/watt by 50-80% from Polaris 10 with Vega? After their performance and perf/watt misrepresentation with Fury X and Polaris 10, I highly doubt it.

It will happen with Navi, Vega be an improved version anyhow.
People dont care much about power/wattage anyhow.
they care about running their games and leave the debates to forum users.