53 GHz in one box?

[limxdul]

Anyone else catch this? http://www.engadget.com/2009/02/28/cooler-masters-5-cpu-monstrosity-has-your-craptop-cowering-in-a/

 

[Zero82z]

Interesting, but you'd probably get a lot more production out of a box with 8 GPUs using a 9800GX2 or GTX295.

 

[sirmonkey1985]

uh.. why not just use a quad socket opteron system.. use 5 of those that will give you 20 processors.. well over the 53ghz mark

 

[Zero82z]

uh.. why not just use a quad socket opteron system.. use 5 of those that will give you 20 processors.. well over the 53ghz mark

Good luck fitting five quad-socket boards into an ATCS840.

 

[crazjayz]

Good luck fitting five quad-socket boards into an ATCS840.

I'm sure it can be done with a lot of modding. All 5 boards would have to be vertically parallel and all CPUs would have to be watercooled, much like they are in the 53Ghz. Making it fit would be quite the PITA though

 

[Jathanis]

Good luck fitting five quad-socket boards into an ATCS840.

Why? I just ordered pizza two nights ago, as soon as that's gone, my cases will be ready!

 

[Tigerbiten]

That would have been impressive a couple of years ago.
Now its a case of "Is it worth running ??".

The SMP client under Linux does between 250-450 PpDpGhz depending on the protien.
~450 x 53 = ~23,850 PpD if you get fast protiens.
If it pulls around 750 watts thats ~30 PpDpWatt.
But you could do only do half that with all slow protiens.

An 8 GPU cores box will do 4,500-6,500 PpDpCore.
8 x 4,500 = ~36,000 PpD with slow protien.
If it pulls ~1,000 watts that ~36 PpDpWatts and thats with slow protiens.
All fast protiens and your looking at over 50 PpDpWatt.

Luck ....................

 

[EvilAlchemist]

That would have been impressive a couple of years ago.
Now its a case of "Is it worth running ??".

Tiger, I think you are right to some degree ... but we don't know what the SMP2 Client will bring to the table ....

Another factor is easy of use / manageability factor

While the GPU2 client is stable, running the Linux SMP client is even easier ...

It really all depends on how much $$ and time you can throw at a problem.

 

[APOLLO]

I'm sure it can be done with a lot of modding. All 5 boards would have to be vertically parallel and all CPUs would have to be watercooled, much like they are in the 53Ghz. Making it fit would be quite the PITA though

It cannot be done. That many QS boards cannot fit in the case of the size chosen for this project even with all the internals removed and only the shell remaining. Heck, one quad board would be difficult without extensive modification, unless it's one of the Tyan models with hardly any expansion slots.

This is quite a dense powerhouse even if it's not my preferred way of doing things. And Tiger is right concerning F@H, with the advent of GPU2 this showcase system is all but obsolete, if one were on the quest for ultimate production in a given space. Other than that, I'm sure it would be a good setup for certain niche applications.

 

[EvilAlchemist]

with the advent of GPU2 this showcase system is all but obsolete

That is not 100% correct.
Remember that the GPU client can only run certain type of problems but the CPU can any type of problem, but some much slower then GPU.

There is a lot of work being done to speed up CPU processing speed and allow GPUs to run different type of work , but both will take some time ....

It is best to have a mixture of different clients running in your setups.

Protects you in terms of future PPD / Client Changes and helps the entire F@H project.

 

[APOLLO]

^ You're right about development and I was aware of the continued advancement of the CPU clients. I should have been clearer with my reply. I consider any system that is exclusively CPU-based for F@H old-fashioned. This CM setup doesn't seem to have support for high-end GPUs since the motherboards that constitute the specs for this system are ITX-based, and room is extremely limited.

I no longer desire to build F@H systems with only the CPU clients in mind. I still run CPU and GPU exclusive systems, but do not like them nor do I intend to continue building exclusive setups. It's not for the PPD that my viewpoint is directed thus, it's about the variety and especially the versatility, which I have been professing here for well over a year now ever since I ran the GPU1 client.

Just to reaffirm what I meant, it's not about the PPD that I find this setup lacks, but the versatility.

 

[limxdul]

That is not 100% correct.
Remember that the GPU client can only run certain type of problems but the CPU can any type of problem, but some much slower then GPU.

There is a lot of work being done to speed up CPU processing speed and allow GPUs to run different type of work , but both will take some time ....

It is best to have a mixture of different clients running in your setups.

Protects you in terms of future PPD / Client Changes and helps the entire F@H project.

Hrmmm this is interesting... is there anyway to id those type of jobs ahead of time and then add them to either the gpu or cpu queue? If not someone should defintely make one...