• 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.

A question about Nvidia Pascal's unified memory

cbn

cbn

Lifer
http://blogs.nvidia.com/blog/2014/03/25/gpu-roadmap-pascal/

According to the above blog and other info on the web Nvidia Pascal will have unified memory. This will allow the CPU to share memory with the GPU and the GPU to share memory with the CPU.

Does anyone see a problem with this working on Core 2 level hardware? Often times these machines ship with limited amount of RAM and being able to use the Pascal Video card's VRAM as System RAM would be a real benefit IMO.
 
Nvidia Pascal's unified memory is useless for your desired effect ...

Unified virtual memory mostly only poses a benefit to heterogeneous systems such as APUs ...

Using video memory as a basis for the system is useless because it will end up been slower since paging from video memory will have a higher latency than system memory due to the fact that the described scenario would have to deal with PCIE bus latencies ...
 
cbn said:
Does anyone see a problem with this working on Core 2 level hardware? Often times these machines ship with limited amount of RAM and being able to use the Pascal Video card's VRAM as System RAM would be a real benefit IMO.
Click to expand...

Why would anyone pair a late 2016 architecture with a July 14, 2006 architecture? I don't even....

We already know from user testing that Q9550 @ 4.0ghz bottlenecks a GTX570/HD7950. Considering GTX750Ti = GTX480 = GTX570, that means even the crappiest $150 Pascal chip will be bottlenecked by the fastest Core 2 Quad. I can't even comprehend who would be considering pairing a Pascal GPU with a Core 2 Duo/Quad CPU for gaming next year.
 
Last edited:
RussianSensation said:
Why would anyone pair a late 2016 architecture with a July 14, 2006 architecture? I don't even....

We already know from user testing that Q9550 @ 4.0ghz bottlenecks a GTX570/HD7950. Considering GTX750Ti = GTX480 = GTX570, that means even the crappiest $150 Pascal chip will be bottlenecked by the fastest Core 2 Quad. I can't even comprehend who would be considering pairing a Pascal GPU with a Core 2 Duo/Quad CPU for gaming next year.
Click to expand...

I'm thinking the entry level Pascal cards, not the midrange ones.

Example: Cards that would be listed in this thread--> http://forums.anandtech.com/showthread.php?t=2403091
 
ThatBuzzkiller said:
Using video memory as a basis for the system is useless because it will end up been slower since paging from video memory will have a higher latency than system memory due to the fact that the described scenario would have to deal with PCIE bus latencies ...
Click to expand...

I think I see your point.

I believe NVlink is meant to be the answer to that problem, but needs to be supported on the CPU.
 
cbn said:
I think I see your point.

I believe NVlink is meant to be the answer to that problem, but needs to be supported on the CPU.
Click to expand...

NVlink only solves the interconnect bandwidth throughput issue, not the latency issue ...

NVlink/Omni Scale Fabric are meant to increase network traffic capabilities ...
 
cbn said:
I think I see your point.

I believe NVlink is meant to be the answer to that problem, but needs to be supported on the CPU.
Click to expand...

ThatBuzzkiller said:
NVlink only solves the interconnect bandwidth throughput issue, not the latency issue ...

NVlink/Omni Scale Fabric are meant to increase network traffic capabilities ...
Click to expand...

Ultimately, it's all really meant for the clustered compute world (think: GPGPU supercomputers). There will be a whole new GPU interface that replaces PCIe if Nvidia succeeds on this mission, but I don't think that will ever translate to the consumer market. I do think they'll adapt NVlink to replace SLI bridging, but it's not really shaping up to be a consumer tech push.
 
NVLink won't turn up in any Intel processor. If it's used for anything in the consumer market, it will be as an SLI bridge... but honestly, it isn't going to make that much difference. Latency is still going to suck.
 
Why on earth would you buy a Pascal chip for an Core 2 in order to get more memory, when its way cheaper to just BUY memory.

Not to mention it wont even work like that.
 
cbn said:
http://blogs.nvidia.com/blog/2014/03/25/gpu-roadmap-pascal/

According to the above blog and other info on the web Nvidia Pascal will have unified memory. This will allow the CPU to share memory with the GPU and the GPU to share memory with the CPU.

Does anyone see a problem with this working on Core 2 level hardware? Often times these machines ship with limited amount of RAM and being able to use the Pascal Video card's VRAM as System RAM would be a real benefit IMO.
Click to expand...

Pascal designed to support fine grained SVM with IBM Power8 architecture.

Coarse-grained SVM is already possible with OpenCL 2.0.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top