Nemesis
That would be a great name for the top Polaris card.
I really think this time round it is going to be AMDs GPUs coming out on top.
Looking at the specs for the top Pascal NVidia card, things are not looking good. Cores disabled and 300W TDP.
-
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
Nemesis
That would be a great name for the top Polaris card.
I really think this time round it is going to be AMDs GPUs coming out on top.
Looking at the specs for the top Pascal NVidia card, things are not looking good. Cores disabled and 300W TDP.
Nemesis
That would be a great name for the top Polaris card.
I really think this time round it is going to be AMDs GPUs coming out on top.
Looking at the specs for the top Pascal NVidia card, things are not looking good. Cores disabled and 300W TDP.
GP100 will likely never come to the consumer space, it has a lot of extra circuitry on Die for dealing with all of the NVlink connections as well as the Double precision shaders. Considering how dire yield appear to be for them so far and with current rumours, including The 1080Ti will be using full GP104.
I think overall AMD (well ATI cough...) do have better engineers and can make better products. The problem with them is money, so poor Marketing, drivers are hit and miss and even if they do release a much better product they will still fall behind Intel or Nvidia.
With low level API better chips will finally be used fully, so thus is not gonna be a problem any more. Drivers for nvidia are much worse in past year than it was for AMD. I do agree about money and marketing, but with market attitude towards AMD being so hostile I don't think anything would please anyone
/OT
Well you are right, P100 have lots of thing not needed in a gamer card, but misses a few things as well...if i'm right it doesn't have any ROPs for example.
If it has no ROPs or TMUs then it is effectively headless and pure compute.
If that is true then it would have been better for them to use a different part moniker such as TP or CP instead of GP, the G generally meaning graphics. Etc.
Well as far as i know they haven't...it is just called the Tesla P100
You can see i called it by that name in my post before
The mezzanine card is called the P100 and Ofcourse it is headless. But the gpu die is called GP100, it is a matter of if it has ROPs or not, I it does not then it is headless at heart as it has no render outputs.
GP100 will most likely surface in the consumer space as the Titan.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 31,182
Makes sense, and they can even move NVLink into the consumer space to make it truly stand out at the top of the pyramid
So what you guys are saying is that nvidia hid Graphic parts from GTC spec sheets? Or are you guys saying that nvidia just forgot to tell about them? Or that nvidia will add them later on, but then it is not GP100 then but something else. And what's this dream about mobo manufacturers spinning off seperate mobo line just for nvidias nvlink? 
So what you guys are saying is that nvidia hid Graphic parts from GTC spec sheets? Or are you guys saying that nvidia just forgot to tell about them? Or that nvidia will add them later on, but then it is not GP100 then but something else. And what's this dream about mobo manufacturers spinning off seperate mobo line just for nvidias nvlink?
Mobo manufacturers are irrelevant here
AMD or Intel have to support it. And there is no reason they ever will.Makes sense, and they can even move NVLink into the consumer space to make it truly stand out at the top of the pyramid
Would be a stupid move, no one in their right mind would buy a new motherboard just to install a new GPU.
I disagree, sure they would not sell in great quantities but people would buy.
Don't think it will happen though
I disagree, sure they would not sell in great quantities but people would buy.
Don't think it will happen though
Unless its an IBM mobo with an IBM CPU, it won't happen in your lifetime.
NVLink is a GPU-GPU-CPU interconnect. CPU has to support it.
Not to mention it's completely useless in consumer space since PCIe bandwdith is not a problem.
Last edited:
I can put the latest Xeon processors in my RVE motherboard if it suits me even though there are things on the CPU the board won't support.
I can use the original Titan as a gaming card even though it is actually a compute card.
The point is GP100 will make a fantastic gaming card even if not everything on the card is usable.
You're wasting your breath mate it goes in one ear out the other. People just can't get it through their heads. It's like how they still want pooled VRAM in mGPU.
re: Pascal pretty obvious there will be some sort of GP110 in 2017 or whenever they can get more than 0.7 good dies per wafer.
It would be doubtful that Nvlink came to the consumer space unless they made their own sockets.
The whole point of Nvlink is having low latency, high bandwidth connections between gpu's. You don't get that with PCIe ports since they all go back to the PCIe controller. There are no direct connections between ports, The only way this would work is if Nvidia worked with a mobo manufacturer that made a custom port for their cards, or had a pci x32 connector where the extra pins were Nvlink pins.
That way the first x16 worth of pins are normal PCIe but then you have the extra pins for Nvlink.
AMD's 'Coherent Fabric' is more likely to appear in consumer parts, but only between CPU-GPU. But that is only if it can work over PCIe links and since they can have Dual mode Controllers on GPU and CPU to support 'CF' and PCIe.
Now the one on the far right is an PCIe X32 socket.
The whole point of Nvlink is having low latency, high bandwidth connections between gpu's. You don't get that with PCIe ports since they all go back to the PCIe controller. There are no direct connections between ports, The only way this would work is if Nvidia worked with a mobo manufacturer that made a custom port for their cards, or had a pci x32 connector where the extra pins were Nvlink pins.
That way the first x16 worth of pins are normal PCIe but then you have the extra pins for Nvlink.
AMD's 'Coherent Fabric' is more likely to appear in consumer parts, but only between CPU-GPU. But that is only if it can work over PCIe links and since they can have Dual mode Controllers on GPU and CPU to support 'CF' and PCIe.
Now the one on the far right is an PCIe X32 socket.
Last edited:
Caporegime
- Joined
- 9 Nov 2009
- Posts
- 25,725
- Location
- Planet Earth
Another leak:
http://vrworld.com/2016/04/13/amd-polaris-10-gpu-beat-competing-pascal/
Polaris 10 GPU “67DF:C4” Specifications
14nm FinFET, GlobalFoundries
Diffused in USA (New York state)
Assembled in Taiwan
2304 Cores (silicon: 2560)
36 Enabled Core Clusters (silicon: 40)
New GPU Architecture (not GCN “1.5”)
256-bit Memory Controller
8GB GDDR5/GDDR5X (when available)
That means the GPU is being made in the Samsung fab in New York and apparently the article says the base clockspeed is currently 1.05GHZ and it might hit as high as 1.15GHZ. It will be interesting to see what the maximum boost clockspeed is.
http://vrworld.com/2016/04/13/amd-polaris-10-gpu-beat-competing-pascal/
Polaris 10 GPU “67DF:C4” Specifications
14nm FinFET, GlobalFoundries
Diffused in USA (New York state)
Assembled in Taiwan
2304 Cores (silicon: 2560)
36 Enabled Core Clusters (silicon: 40)
New GPU Architecture (not GCN “1.5”)
256-bit Memory Controller
8GB GDDR5/GDDR5X (when available)
That means the GPU is being made in the Samsung fab in New York and apparently the article says the base clockspeed is currently 1.05GHZ and it might hit as high as 1.15GHZ. It will be interesting to see what the maximum boost clockspeed is.