Nvidia Volta.

[AidenShaw]

24 April 2017.

With about two weeks before Nvidia's expected announcement of Volta GPUs at the GPUtech Conference starting 8 May in San Jose, here's a thread to collect rumours and comments on Nvidia's next generation GPU architecture.

Have at it, all the usual suspects.

​

Meet the Green Team.

 

[flat five]

ha, the ad i get fed when reading your post:

i buy a lot of tools so i get a lot of ads like this..
i'm mainly a blue or yellow type but i have a few of these green ones too


(i , or a mod, will delete this O.T post in a little bit.. found it funny_ish though)

 

[AidenShaw]

ha, the ad i get fed when reading your post:

I had a case where after looking for underwater (scuba) trips I got underwear ads for days.

Some of the ads were a lot of fun, though.

 

[AidenShaw]

I just ordered twenty GTX 1080 Ti cards.

I really hope that turns out to be a $14K mistake .

 

[Gazember]

Will Volta GPUs come with their own diesel electricity generators?

 

[DPUser]

Will Volta GPUs come with their own diesel electricity generators?

No, "clean" coal.

 

[flat five]

I just ordered twenty GTX 1080 Ti cards.

for what?
and do you also do the installation/IT work? or just the buying?

 

[AidenShaw]

for what?
and do you also do the installation/IT work? or just the buying?

CUDA

All of the above. We're a small, rogue research group in a big company. We don't let IT know what we're doing.
[doublepost=1493163118][/doublepost]

Will Volta GPUs come with their own diesel electricity generators?

Hydrogen fuel cells - only by-products are dihydrogen monoxide and Science.

 

[Prince134]

Big wish the green team provide Mac web driver for Volta sooner than they gave us for Pascal.

 

[AidenShaw]

Big wish the green team provide Mac web driver for Volta sooner than they gave us for Pascal.

If y'all send clear (official) feedback to Apple that full CUDA support for standard PCIe graphics cards on the mMP is essential - then this will happen.

Otherwise, it's hard for Nvidia to make the case that a few sales to fanatic cMP users are worth the effort of developing Nvidia drivers for Apple OSX.

It would also help to make a case for VR on Nvidia - although I'm ignorant on that topic. (Have systems with quad Titan X cards, and not a single monitor connected.)

 

[Prince134]

Yep, that's why we wish, but they've just done that for Pascal. Just so much later.

 

[Stacc]

I thought the rumors for this were maybe just some sort of pascal based tablet or set top box. It seems a bit early for Volta.

 

[PortableLover]

Wonder how much faster these voltas will be?

 

[tuxon86]

Wonder how much faster these voltas will be?

They go up to 11...

 

[Silencio]

I just ordered twenty GTX 1080 Ti cards.

I really hope that turns out to be a $14K mistake .

I recently ordered 8 GTX 1080's right before the 1080 Ti's were announced, so I feel your pain.

 

[AidenShaw]

TechPowerup: NVIDIA's Volta Reportedly Poised for Anticipated, Early Q3 2017 Launch

 

[Stacc]

TechPowerup: NVIDIA's Volta Reportedly Poised for Anticipated, Early Q3 2017 Launch

I wonder if we will get consumer cards this year or just something that goes into servers like GP100 was last year. Regardless its great to see Nvidia being so aggressive.

 

[cube]

- Automatic driver updates without login
- FreeSync support
- OpenCL optimization

 

[AidenShaw]

I wonder if we will get consumer cards this year or just something that goes into servers like GP100 was last year. Regardless its great to see Nvidia being so aggressive.

Considering the timing and the availability of GDDR6 and HBM2, several pundits have predicted that GV100 will roll out first as a cost-no-object HPC offering, and GV10x chips for the workstation and consumer markets will follow by the end of the year.

So, a repeat of the Pascal rollout is quite possible.

 

[Flint Ironstag]

24 April 2017.

With about two weeks before Nvidia's expected announcement of Volta GPUs at the GPUtech Conference starting 8 May in San Jose, here's a thread to collect rumours and comments on Nvidia's next generation GPU architecture.

Have at it, all the usual suspects.

View attachment 697380​

View attachment 697381
​

Meet the Green Team.

Just couldn't let us enjoy Pascal eh?

 

[AidenShaw]

Just couldn't let us enjoy Pascal eh?

Pascal's old...

Seriously, though, Nvidia's chip transitions have been slow affairs. (Pascal was announced last May, but I just ordered twenty Pascal cards to upgrade Maxwell systems - because it's just within the last few weeks that any reseller dropped the "one per customer" or "two per customer" rule on them.)

Enjoy Pascal for at least the rest of this year, unless you want to drop $5K to $10K on single GV100 chips in Q3.

 

[cube]

Aiden made me curious about whether his new cards were for workstations or servers.

 

[AidenShaw]

Aiden made me curious whether his new cards were for workstations or servers.

All servers. Not a single monitor will be connected to any card.

And twenty cards means that I'm upgrading five servers.

These are Titan X (Pascal) cards in a server that I'm not upgrading. Note the 24 DIMM slots.

 

[cube]

Is that Supermicro?

 

[AidenShaw]

Is that Supermicro?

No - HPE ProLiant ML350 Gen9 https://www.hpe.com/us/en/product-c...p.hpe-proliant-ml350-gen9-server.7271259.html

Rack-mount model. Dual socket E5-26xx v4. 2400 watt power (quad 800 watt hot plug supplies).

For dev/test I also have a bunch of DL380 Gen9 (similar dual socket E5-26xx v4, but in 2U chassis with support for two GPUs).

The DL380s have a pair of hot plug 1400 watt power supplies.