Nvidia Voltra

[TDMVPDPOY]

lmao amd

here are the voltra specs http://wccftech.com/sk-hynix-gddr6-nvidia-volta-gpu-gtc-2017/

moving to ddr6
while amd vega10 currently is still on ddr5, and vega11 will be hbm2 memory below 1080 performance....

[baseline bum]

Holy fuck the Volta V100 chip has a die size of 815mm^2.

[Wild Cobra]

Holy fuck the Volta V100 chip has a die size of 815mm^2.

Over 1 sq inch.

I've seen larger chips, but that was when they were still using 0.25 micron technology.

[TDMVPDPOY]

if the chips continue to get smaller, why not make it a dual card since they are going away from sli?

[Wild Cobra]

if the chips continue to get smaller, why not make it a dual card since they are going away from sli?

That one chip consumes as much as 300 watts. Smaller is not easy to do. Do you realize what is involved in 12 micron production?

[illusioNtEk]

Can't wait for next gen laptops... finally going to have the same power as desktops have

[baseline bum]

if the chips continue to get smaller, why not make it a dual card since they are going away from sli?

Dual gpu cards still use SLI man.

[Wild Cobra]

Dual gpu cards still use SLI man.

Yes, but on-board would be far less susceptible to signal to noise ratios.

[ElNono]

The biggest issue with SLI is that the biggest performance gain is with AFR (Alternate Frame Rendering, where one card draws even frames and the other odd frames) and it suffers from micro-stuttering, which is inherent to multi-GPU (especially dual) configurations.

The reason is that one of the cards always has to transmit the full frame buffer over to the master for display post-rendering. Latest SLI bridges can do up to 650 MHz, which is good enough for up to 5K resolutions, but the transfers are unavoidable, and because frame rendering time is not constant or exact, a small delay can really cause partial frame renders, thus the stutter.

G-Sync or FreeSync would correct for that, but then you have dynamic refresh rates...

Now if you put them on the same card, you could optimize by using dedicated memory per framebuffer, but then you run into cache coherency issues, etc. Multi-CPU configurations are always difficult when you have to eventually do serial work (one frame at a time).

[TDMVPDPOY]

was watching the NVidia computex 17 conference...

regarding volta and whatever....seems like NVidia wants to challenge intel or whoever is making cpus, since the volta gpu is just as powerful as a cpu, why the need for a cpu from a competitor when NVidia could make their own cpu = gpu volta...I wonder if they jump to cpu, would they give intel/amd a run for their money with cpu competition? the way the ceo was talkin about volta, looked like he wanted volta to be mainstream....

hence was watching the NVidia laptop segment how they have a gaming laptop 18mm with a 1080gtx comparing it to ps4pro makes no fkn sense....how old is a ps4pro vs this new laptop, hence that laptop can always be upgraded, while the ps4pro u cant upgrade the core component like cpu/gpu...they should've compared it to other gaming laptops manufactures