Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Nvidia Volta.
- Thread starter AidenShaw
- Start date
- Sort by reaction score
- Status
very nice!... thx for sharing I would have thought that cards sandwiched together like that would be bad for heat though.
440 cores and 20 GPUs in 7U (InfiniBand version):
https://www.supermicro.nl/products/superblade/module/SBI-7128RG-F2.cfm
I thought maybe your Ti support was unofficial and they would also work here.@ about $5K per GPU
Never seen them over 50° under load. There's a lot of airflow through the cards from the chassis fans, and the inlet temperature is 15°.
[doublepost=1494334683][/doublepost]
That chassis sucks air through the front-mounted GPUs and exhausts to the rear - it's designed for fanless GPUs.
Last edited:
I see front slots in the manual but probably the GPU fans would not be effective enough and you would would not like to stand in the aisle.
![7128RG-F2_top[1].jpg](https://forums.macrumors.com/data/attachments/614/614135-808c6021cc4095a2f789adafe570a6ad.jpg "7128RG-F2_top[1].jpg")
I know the supported accelerators have no fans, normal GPUs do.
And I know there would be no intake from the front, but blades with front drives do not seem to provide much.
It would seem there's some intake from the sides, but it does not look like enough.
The manuals do not seem to mention cooling at all.
...full of InfiniBand switches, management controllers, network ports,...
Also note that the space is very tight around the GPUs - only "Founder's Edition" and blowerless form factor will fit. (And you need special low-profile right angle power plugs - there isn't enough space to plug a normal aux PCIe connector.)
I know, if you can do without the InfiniBand, the minimum should be 2 or 3 bays populated (a blade without IB costs less than 1K, while with it costs 2K+).
If you look at a blade drawing, it has back grills in these areas.
Two rumors about Volta GPUs:
1. NVIDIA is introducing at least two Volta parts this year, the first is in September and no info on the 2nd part. Either way, 1st part shows improved performance/mm and performance per clk which we haven't had for ages. (i.e it is a step forward over Maxwell per clock, which was not the case for Pascal)
2. NVIDIA's new parts will offer GDDR6, and HBM is reserved for Tesla and compute parts for now.
Source:
http://carbonite.co.za/showthread.php?t=158997&p=1094603#post1094603
1. NVIDIA is introducing at least two Volta parts this year, the first is in September and no info on the 2nd part. Either way, 1st part shows improved performance/mm and performance per clk which we haven't had for ages. (i.e it is a step forward over Maxwell per clock, which was not the case for Pascal)
2. NVIDIA's new parts will offer GDDR6, and HBM is reserved for Tesla and compute parts for now.
Source:
http://carbonite.co.za/showthread.php?t=158997&p=1094603#post1094603
I've seen statements that all of the "Founder's Edition" cards are identical - literally manufactured by Nvidia and put in Asus/EVGA/PNY/MSI... boxes. (At least for the 1080Ti.)
There are blower cards with rear power.
Arent currently Server GPUs made for specific Thermal designs? For example Under 150W TDP, Under 225W TDP, Under 300W TDP?
And all of what you are talking about is just standardizing this?
- Status