[PCgamer] Pascal to be 10x faster than Maxwell!

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
Nice bit of CEO math from Jen Hsun. :) Half of the gain was from being able to connect twice as many GPUs in the system, and going from FP32 to FP16 was another 2X improvement.

Stacked memory will give a great performance boost, but don't expect 10X miracles.
 

monstercameron

Diamond Member
Feb 12, 2013
3,818
1
0
Nice bit of CEO math from Jen Hsun. :) Half of the gain was from being able to connect twice as many GPUs in the system, and going from FP32 to FP16 was another 2X improvement.

Stacked memory will give a great performance boost, but don't expect 10X miracles.


Maybe if you take raw shader count, die size and shader ipc...
 

sxr7171

Diamond Member
Jun 21, 2002
5,079
40
91
You never know what may be going on in the background. You may want to read pcie 4 specification to see, if Nvidia had a "difference of opinion" into where it must progress as a standard. Just like Nvidia did with the Adaptive Sync.

Nvidia has significant marketshare in discrete, so i wouldn't be surprised, if they tried to push some proprietary chippery, as they used to before for enabling SLI. More money is more money. I don't even blame Nvidia for state of affairs, but rather their customers, who're simply not sending a strong enough message to cut this crap. If you keep buying from an abusive seller, well, abuse should be expected.

Right now we're not using PCI-E 3.0 x16 to it's fullest. So I'm not understanding how faster link would make so much difference.
 

Pottuvoi

Senior member
Apr 16, 2012
416
2
81
Nice bit of CEO math from Jen Hsun. :) Half of the gain was from being able to connect twice as many GPUs in the system, and going from FP32 to FP16 was another 2X improvement.

Stacked memory will give a great performance boost, but don't expect 10X miracles.
Indeed, although the FP16 should give nice performance boost in some cases. (HDR post processing, rendering.)
 

Omar F1

Senior member
Sep 29, 2009
491
8
76
Right now we're not using PCI-E 3.0 x16 to it's fullest. So I'm not understanding how faster link would make so much difference.
Exactly, gaming-wise we almost never been interface-bandwidth limited.

So, in three years if we're going to choose between either Nvidia or AMD compatible motherboards that would be a problem.
 

ultimatebob

Lifer
Jul 1, 2001
25,135
2,445
126
Exactly, gaming-wise we almost never been interface-bandwidth limited.

So, in three years if we're going to choose between either Nvidia or AMD compatible motherboards that would be a problem.

Something tells me that the big OEM's like Dell and HP will refuse to play this vendor lockin game and stick with the open PCI standards.

The gaming motherboard manufacturers will probably love it, though.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
PCIe 4.0 is coming with Skylake. So no worries.

It may be Skylake-E first, but it will hit the consumer fast. So any special graphics interface is still far away.

Skylake.png

PCIe-4-PCIe-3.1-specifications.jpg
 

DeathReborn

Platinum Member
Oct 11, 2005
2,743
734
136
Something tells me that the big OEM's like Dell and HP will refuse to play this vendor lockin game and stick with the open PCI standards.

The gaming motherboard manufacturers will probably love it, though.

While both Dell & HP will have systems using NVLink for HPC/Compute purposes I do agree that general consumer machines will still use PCIe until Intel (if they ever do) adopts NVLink. If Intel does then AMD would have to follow.

I think for Tesla the orientation of the mezzanine connector would be very suitable. It should also be possible to have a PCIe riser card to attach them to which would also open up a new kind of SLi bridge over NVLink cables instead of PCIe.