[Info-vax] The Future of Server Hardware?

[JohnF]

Stephen Hoffman <seaohveh at hoffmanlabs.invalid> wrote:
> JohnF said:
> 
>> Just curious -- nowadays this particular kind of stuff would be
>> done using some cuda kind of architecture. Is that right?
> 
> As with most stuff in computing, "it depends".
> 
> But yes, for some tasks, gazillions of GPUs in gazillions of servers is 
> the go-to design.
> 
> For stuff like Facebook, not so much.
> 
> If you're looking to render images or brute-force passwords, definitely 
> yes.  (This is why I was pointing to bcrypt and scrypt a while back, 
> too.  But I digress.)

Thanks for the info. So vectorizable, "definitely yes";
parallelizable, "not so much."

> If you're doing your own rack-n-stack, you look at whether your 
> algorithms can be adapted to and run in a GPU.  If it can, then you can 
> load the boxes with graphics cards or with add-on accelerators such as 
> the Intel Xeon Phi PCIe boards.
> 
> If you're a vendor building a server, it varies.  For the very high-end 
> servers, IBM Roadrunner is built on a mix of x86-64 and PowerXCell 
> chips.  Cray Jaguar is built on x86-64 chips, but Cray is now busy 
> adding (retrofitting?  refitting?) GPUs to the design.
> 
> This whole area has been a moving target: ATI Streams, Microsoft 
> DirectCompute, Nvidia CUDA and OpenCL provide (among other details) 
> programming languages used to avoid dealing directly with the GPUs.  

Is there a pretty standard whitebox configuration people typically
put together as a gpu platform -- MB make and model, power supply
and cooling options, memory, etc? And is nvidia/cuda the current
favorite among the gpu/"language" options to put in that box?

> For OpenVMS, not so much; the vector options (see the VAX vector 
> architecture and VVIEF for some related details) and the add-on 
> co-processors are long gone from the VMS market.

Good vms segue.

-- 
John Forkosh  ( mailto:  j at f.com  where j=john and f=forkosh )