-
We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.
updatevideocardzAMD Polaris 10 engineering sample pictured
Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Mahigan
Senior member
It's 4GB HBM, like I have kept saying. 🙂
And 4GB is more than enough..
Oh and it's a constant 60FPS+, which means the minimum FPS was 60.
Like I said, faster than FuryX performance from a BaffintXT (Polaris 10) core. It's faster than a GTX 980 Ti and overclocks really well too.
And no... I can't tell you where I've gotten my info from.
And 4GB is more than enough..
Oh and it's a constant 60FPS+, which means the minimum FPS was 60.
Like I said, faster than FuryX performance from a BaffintXT (Polaris 10) core. It's faster than a GTX 980 Ti and overclocks really well too.
And no... I can't tell you where I've gotten my info from.
Last edited:
VirtualLarry
No Lifer
http://www.anandtech.com/show/10145/amd-unveils-gpu-architecture-roadmap-after-polaris-comes-vega
"With Polaris confirmed to use GDDR5, Vega is notable for being the first AMD architecture to use HBM2, and the first parts in general to use HBM tech since Fiji."
Polaris 10 and 11 is GDDR5 based according to anandtech. So you are wrong on the HBM part. Moreover using 4096 bit wide HBM controller to get 4 GB is cost prohibitive for a high volume mainstream part.
AMD can easily use a 256 bit GDDR5 memory controller with 8 Ghz GDDR5 chips and with improved memory compression techniques can feed the Polaris 10 GPU without bandwidth bottlenecks.
My guess is Polaris 10 - 40-48 CU, 2560-3072 sp, 4 shader engines, 64 ROPs, 256 bit GDDR5, 256 GB/s, 4/8GB. Price USD 249 - 399. I think there might be 3 SKUs as yields are still not good at 14nm FINFET for > 200 sq mm dies.
Thats the big or the small polaris? at what price range?
Mahigan
Senior member
It wasn't confirmed to use GDDR5. That's simply a different interpretation of Raja's words.
Raja said that HBM design and implementation was cost prohibitive and that AMD needed to recoup those costs. Thing is, AMD is sitting on HBM memory produced for their FuryX line. They have to use it somewhere in order to recoup costs. So AMD will use HBM for another generation. That's the Polaris generation. PCPer sad as much.
The way Anandtech interpretated this was that HBM was too expensive so AMD will use GDDR5, which is incorrect.
http://www.pcper.com/news/Graphics-...past-CrossFire-smaller-GPU-dies-HBM2-and-more
Oh and Polaris 10 will out perform both the FuryX and GTX 980 Ti.
Then you look at the shipping manifest:
And you notice that every GDDR5 circuit board mentions GDDR5 and the speed, because the traces are printed on the boards, but not BaffinXT (Polaris 10).
And you have your answer. HBM1.
Last edited:
DownTheSky
Senior member
With that kind of performance, for a new product, 4Gb seems little.
Do you really expect about 232mm^2 GPU (rumored size of Pascal 10) to outperform Fury X? 😀
Last edited:
Mahigan
Senior member
Not across the board but under certain conditions, yes.
Compute wise, it won't out perform Fiji under large parallel compute loads, ROP wise, it will due to the inclusion of color compression, larger L2 cache and a new memory controller.
Geometry performance wise? Yes.
Texture(texels/s) no.
So under certain conditions, it will outperform a FuryX and GTX 980 Ti. And then there's the overclocking potential.
Snarf Snarf
Senior member
Estimates have it around ~232mm^2 give or take. It really isn't out of the realm of possibility that Polaris 10 could match and beat a Fury X in some titles due to changes in u arch.
If it has less GCN4 cores yet clocks high enough thanks to the new process or uarch changes, then it'll probably outperform it too.
There are too many variables to consider and the specs are still unknown, yet it is entirely possible for a chip about 40% the size of Fiji to match it or even outperform it. I'd bet many of the improvements that were to go into 20nm GCN (what GCN 1.2 was actually supposed to be instead of what we'd got in Tonga and Fiji?) are now included in GCN4.
Most of the GPU is new. 232mm^2 is small compared to the outgoing stuff, yes, but I wouldn't rule out that possibility.
There are too many variables to consider and the specs are still unknown, yet it is entirely possible for a chip about 40% the size of Fiji to match it or even outperform it. I'd bet many of the improvements that were to go into 20nm GCN (what GCN 1.2 was actually supposed to be instead of what we'd got in Tonga and Fiji?) are now included in GCN4.
Most of the GPU is new. 232mm^2 is small compared to the outgoing stuff, yes, but I wouldn't rule out that possibility.
Last edited:
update：http://www.overclock3d.net/articles/gpu_displays/amd_polaris_10_engineering_sample_pictured/1
AMD's Polaris 10 engineering sample has been pictured and shown running Hitman at Ultra Settings at a constant 60+ FPS at 1440p, which is an impressive feat given that in our testing the the R9 Fury X was not even able to average 60FPS at these settings.
AMD's Polaris 10 engineering sample has been pictured and shown running Hitman at Ultra Settings at a constant 60+ FPS at 1440p, which is an impressive feat given that in our testing the the R9 Fury X was not even able to average 60FPS at these settings.
It's 4GB HBM, like I have kept saying. 🙂
And 4GB is more than enough..
Oh and it's a constant 60FPS+, which means the minimum FPS was 60.
Like I said, faster than FuryX performance from a BaffintXT (Polaris 10) core. It's faster than a GTX 980 Ti and overclocks really well too.
And no... I can't tell you where I've gotten my info from.
I fell for this last time with Fury X.... never again....
Mahigan
Senior member
To each his/her own.
Silverforce11
Lifer
It is hard to believe a 232mm2 chip is out performing Fury X. It would imply everything in the perfect scenario.
Perfect scaling for density at 14FF, perfect clock speed target, perfect architecture gains. But is it possible?
Let's get some educated guesses with info we know:
232mm2 * 2.4x 14FF density is 556mm2 equivalent on 28nm.
390X is 438mm2. Fiji is 600mm2. Fiji however, is a very unbalanced design, incapable of peak throughput in any game compared to the 390X, with the one exception Ashes in DX12 w/ Async Compute enabled. In other games, it's ~20% faster, so it's not the right chip to compare next-gen performance. We would have to assume AMD knew what they were doing with Fiji, a stop-gap testbed for HBM.
Taking Hawaii and extrapolating from that is a more accurate approach, IMO, as it's a more balanced chip front end/shader wise.
Immediately, there's potential for Polaris 10 to be 390X +20% from the die size comparison.
What's the major change in Polaris?
The Geometry Discard Accelerator. We're talking massively improved minimum frame rate performance (no longer geometry bound, for complex scenes).
Improved Command Processor => Increased shader "IPC".
Unless they mess up, Polaris 10 should be much faster than the 390X and that puts it potentially faster than Fury X.
------------------
As for whether 4GB HBM is enough, it clearly is, there's not a single example of 4GB not being enough at playable settings. None. And AMD's implementation of HBM as a "cache" in their dynamic vram system, if 4GB is not enough to cache, the overflow will go into dynamic system ram portion. It requires driver optimization that is game-specific to function, but it's been shown to work in Rise of the Tomb Raider at 4K. At 1440 and 1080p, it's a non-issue.
Perfect scaling for density at 14FF, perfect clock speed target, perfect architecture gains. But is it possible?
Let's get some educated guesses with info we know:
232mm2 * 2.4x 14FF density is 556mm2 equivalent on 28nm.
390X is 438mm2. Fiji is 600mm2. Fiji however, is a very unbalanced design, incapable of peak throughput in any game compared to the 390X, with the one exception Ashes in DX12 w/ Async Compute enabled. In other games, it's ~20% faster, so it's not the right chip to compare next-gen performance. We would have to assume AMD knew what they were doing with Fiji, a stop-gap testbed for HBM.
Taking Hawaii and extrapolating from that is a more accurate approach, IMO, as it's a more balanced chip front end/shader wise.
Immediately, there's potential for Polaris 10 to be 390X +20% from the die size comparison.
What's the major change in Polaris?
The Geometry Discard Accelerator. We're talking massively improved minimum frame rate performance (no longer geometry bound, for complex scenes).
Improved Command Processor => Increased shader "IPC".
Unless they mess up, Polaris 10 should be much faster than the 390X and that puts it potentially faster than Fury X.
------------------
As for whether 4GB HBM is enough, it clearly is, there's not a single example of 4GB not being enough at playable settings. None. And AMD's implementation of HBM as a "cache" in their dynamic vram system, if 4GB is not enough to cache, the overflow will go into dynamic system ram portion. It requires driver optimization that is game-specific to function, but it's been shown to work in Rise of the Tomb Raider at 4K. At 1440 and 1080p, it's a non-issue.
Last edited:
Maverick177
Senior member
It's 4GB HBM, like I have kept saying. 🙂
And 4GB is more than enough..
Oh and it's a constant 60FPS+, which means the minimum FPS was 60.
Like I said, faster than FuryX performance from a BaffintXT (Polaris 10) core. It's faster than a GTX 980 Ti and overclocks really well too.
And no... I can't tell you where I've gotten my info from.
You're not talking out of your ass, aren't you?😵
Mahigan
Senior member
No.
I'm not. Talk to game developers if you want. I've been discussing these issues with various developer's as well as other contacts ever since August of last year.
Polaris 10 achieves a constant 60+ FPS under Hitman at 1440p Ultra settings. That's faster than a GTX 980 Ti or FuryX.
It doesn't mean it will be faster under every condition but it is faster in that instance.
Last edited:
DrBombcrater
Member
Process improvements can throw up performance jumps like that. Remember the RV770, aka Radeon 4870? A 260mm2 chip that arrived and delivered performance equal to, and sometimes better than, NVidia's competing 570mm2 GT200.
Silverforce11
Lifer
It's not accurate to compare chips of vastly different architectures to each other, certainly not for AMD vs NV.
It's best to compare GCN to GCN, or Polaris 10 vs Hawaii/Fiji.
As said, the die size equivalence alone gives them >390X performance, easily.
It means, that if Polaris 10 is at least as powerful as Fury X then Radeon R7 370 sized GPU will be 2,5 times faster and it means that AMD would have to give up on reducing power consumption at all on Polaris 10. Sounds too optimistic, IMO.
http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/charts1/swb_1920h.png
http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/charts1/swb_1920h.png
Last edited:
Glo.
Diamond Member
The problem with information on OC3D is that they tested, and posted Hitman in DX11 condition, because for them DX12 was to buggy to test.
http://i.imgur.com/Ti5Ei1L.jpg Here we have ultra settings as well in 1440p. And it is on par with R9 390X.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 25K
-
-
-
