• 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.

Nvidia announce Volta

Were only those with purple hair invited to attend?

There's just the minor issue of sacking delicate DRAM on top of 300w GPU dies :D

It won't be stacked on the actual die though, you wouldn't be able to cool the GPU surely. That slide makes it look more like it's the RAM chips that are stacked on each other.
 
There's just the minor issue of sacking delicate DRAM on top of 300w GPU dies :D

I know your not necessarily being serious, but modern GPU's are not consuming 300w. Even last the generation, the 580 was listed at 244w, the 680 195w, so in two generations time they will be nowhere near 300w.
 
My point being DRAM is *usually* quite delicate and GPUs are hot, even 195w ones. As wolvers pointed out, I had not noticed this is just die stacked RAM on the same substrate rather than a full on die stack

I recall reading about IBM working on stacking RAM on CPU/GPU, apparently AMD have sampled some APUs with 512MB/1GB stacked on board with silly bandwidth - wave goodbye to sub £150 GPUs in a year or so... Although only rumours ;)
 
I thought wattage figures these days were bogus? Around 2010, Nvidia was having problems with Fermi being hot and power hungry and they were in danger of straying outside the PCI-E spec (over 300W). People were up in arms at that time because NV's new cards had stated wattages which seemed to have been calculated using "fuzzy math". Took a break from computers after that, can someone fill me in? Is it just simply smaller process requiring less power?
 
That, better clock control and gating. Figures from every manufacturer should be taken with a does of sodium chloride though
 
My point being DRAM is *usually* quite delicate and GPUs are hot, even 195w ones. As wolvers pointed out, I had not noticed this is just die stacked RAM on the same substrate rather than a full on die stack

I recall reading about IBM working on stacking RAM on CPU/GPU, apparently AMD have sampled some APUs with 512MB/1GB stacked on board with silly bandwidth - wave goodbye to sub £150 GPUs in a year or so... Although only rumours ;)

According to that link Greg posted earlier someone is working on actually stacking RAM onto the core. I bet that will be low powered cores though that produce very little heat.
 
My point being DRAM is *usually* quite delicate and GPUs are hot, even 195w ones. As wolvers pointed out, I had not noticed this is just die stacked RAM on the same substrate rather than a full on die stack

I recall reading about IBM working on stacking RAM on CPU/GPU, apparently AMD have sampled some APUs with 512MB/1GB stacked on board with silly bandwidth - wave goodbye to sub £150 GPUs in a year or so... Although only rumours ;)

Yeah I believe this is just stacked RAM on the same substrate - it would make a huge huge difference to the trace latency not to mention the amount of space it would save on the PCB.
 
Back
Top Bottom