Weekly Stats - April 13, 2014

[GLeeM]

I will start a new thread for performance tuning SETI once I get some initial benchmarking done.

I look forward to this, it has the potential to help the TeAm be more productive and save money on our hobby; errr ... hmmm... OK, save money on electricity so we can buy more hardware

 

[lane42]

This could mean war........A FUN war
Nice fleet melon.

 

[Assimilator1]

Yep, both CPU and GPU.

I also just checked and at idle the system pulls right about 100w. Thises things hate using power. Heck, the Dual X5650's I have here only pull about 325 from the wall with all 12 cores loaded.

Yes, AP = AstrpPulse and MB = MultiBeam. For the last 3-4 days the SETI guys have been going crazy with AP. The key I have found for AP on a GPU is you have to dedicate a CPU core to EACH unit. Otherwise the GPU is idle 90% of the time. So sometimes there are no CPU only units being processed but the CPU's are still always loaded.

I don't do ATi because the *NIX drivers for nVidia have always been better. I don't give a crap that they are not open source. At least nVidia keeps them in sync, with a few exceptions like Optimus, with Windows. I say *NIX because it includes Solaris and Linux. I built the development lab at Sun Microsystems where they did the validation for nVidia on Solaris x86.

Another reason is that nVidia provided development resources, not just a few pieces of hardware, to the SETI developers to create the CUDA version of the MB app.

One thing AMD has going for them in the GPU space now that is extremely attractive is the heterogeneous APU's. I'm this close -><- to picking up an A10-7850K just to play with that. But for discrete GPU's they don't have anything I am interested in. I think they are going in the right direction for sure but it is not enough to pull me from nVidia just yet.

Part of it is probably nostalgia in thinking back to just how mind blowing it was the first time I managed to get Quake 2 running in OpenGL on a Riva TNT card back in, what, 1997 or so? It was like someone had turned on the lights and I was seeing the world for the first time!

I will start a new thread for performance tuning SETI once I get some initial benchmarking done. Perhaps I can even package it up and we can build an ad-hoc database of performance metrics using the same set of work units on various platforms.

I am a hardware nut I must confess. If I were to turn on everything I had here I could probably quadruple my output using CPU alone but I would be drawing 30-40KW of power not counting cooling

Lol, that's a lot of power!
I think TNT was '98, I had a TNT2 in '99
What's *NIX? :$

I tried getting a SETI benchmark thread going about 6yrs ago (god I can't believe it was that long ago!), didn't get much response in terms of times but it would be interesting to see what you get.
I don't get what the pull is about APUs, aren't they just low power chips for mobile devices?

Funny how make different cards favour different projects, MW@H likes OpenCL which IIRC is much better developed on ATI than NVidia, so ATI/AMD cards destroy NVidia cards there.

 

[The Melon]

This could mean war........A FUN war
Nice fleet melon.

I should be pulling ahead of you today in what seems to be a mostly arbitrary number, RAC. Wars like this are fun because it can inspire people to find ways to make more efficient use of what they have. That and I don't have as much idle crap laying around and I am using less 4-6 generation old hardware so I am not really increasing my power draw much, if at all!


Assimilator1, *NIX is a generic term used to refer to UNIX and UNIX like OS's all at once. Linux is not by the strict definition a proper decedent of SYS V or BSD UNIX.

The right tool for the job sure is relevant here.

APU is simply what AMD is calling their Fusion Architecture chips. The most recent generation of which are capable of essentially behaving like the "GPU Cores" are just another CPU core in terms of memory and system access. It's not like in the past where integrated graphics is logically separated in memory space and still behaves like a discrete GPU requiring the CPU to manage it's system access. I'm not sure how the SoC's of mobile devices manage memory as I really don't pay to much attention but I think it is more like traditional CPU's where the graphics portion is logically separated.