I look at the elite computers and I see that the CPUs are taking about 50 seconds to compute one credit, but that it takes only about six seconds for their GPUs to crunch one unit. Is this correct? In other words are the best GPUs crunching about eight times as fast as the best CPUs? Thanks.

---

A factor of 8 sounds about right. My GTX460's are turning out WU's about every 10 min, while the CPU takes around 1:45. So for me, it's actually more like 10x.

---

lets not forget that GPU's can't get their max usage from SETI@HOME. Check out DNETC, MILKYWAY, Collatz, or the PSP Sieve at Prrimegrid. These projects aren't working on complex data such as seti and are able to burn through WU's around 20-30X if not faster than a CPU

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

Thanks. Got another question: I understand that the top CPUs usually compute eight or maybe 16 instances. How many instances do the top GPUs crunch?

---

Thanks. Got another question: I understand that the top CPUs usually compute eight or maybe 16 instances. How many instances do the top GPUs crunch?

Depends on the GPU, GTX 280 and lower only one efficiently, GTS 450 and GTX 460 2 Efficiently. Not sure about the GTX 470, but I think it does 3 efficiently. and of course 480, 570, and 580 all can do 3 WU's without incurring efficiency loss. and it takes them mere minutes per WU instead of the hour + it can take a CPU.

Qick edit: I don't know anything about AMD/ATI GPU's running on this project, so someone else may be able to fill in the blanks!

---

I think the most anyone is running is 3 WU's at a time on some of the higher end GTX cards (470/480, a few people with 460's).

---

Thanks. Got another question: I understand that the top CPUs usually compute eight or maybe 16 instances. How many instances do the top GPUs crunch?

Depends on the GPU, GTX 280 and lower only one efficiently, GTS 450 and GTX 460 2 Efficiently. Not sure about the GTX 470, but I think it does 3 efficiently. and of course 480, 570, and 580 all can do 3 WU's without incurring efficiency loss. and it takes them mere minutes per WU instead of the hour + it can take a CPU.

Qick edit: I don't know anything about AMD/ATI GPU's running on this project, so someone else may be able to fill in the blanks!

I currently run 2 at a time on my ATI 5850 I know a few folks even run 3 at a time on the 5850. The 5870 and 5970 can do 3 with ease

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

I think the most anyone is running is 3 WU's at a time on some of the higher end GTX cards (470/480, a few people with 460's).

Guess it depends on how you feel about it. But 2 WU's on my GTX 480 puts the gpu at 85-90% usage and completes both of them in about 13-15 minutes. If I add a third to the mix it stretches the times out to 18-20 minutes.

So I get 4 units done in ~30 minutes. Or 6 done in ~40. I also tend to watch TV or youtube, browse etc. while letting Seti run and find that I get a touch of video lag if I enable three.

---

Traveling through space at ~67,000mph!

...I understand that the top CPUs usually compute eight or maybe 16 instances.
...

No, the top CPUs compute either one or two instances per physical CPU core. Two is only practical on Intel processors which have hyperthreading. Though there's a cc_config.xml setting which allows more instances, it is intended for simulating a multi-core CPU on a lesser actual processor and is inefficient; using that would ensure the CPU is no longer "top".

                                                                 Joe

...I understand that the top CPUs usually compute eight or maybe 16 instances.
...

No, the top CPUs compute either one or two instances per physical CPU core. Two is only practical on Intel processors which have hyperthreading. Though there's a cc_config.xml setting which allows more instances, it is intended for simulating a multi-core CPU on a lesser actual processor and is inefficient; using that would ensure the CPU is no longer "top".

                                                                 Joe

An i7 quad will compute up to 8 at a time max, the 980x will do 12 max. However there have been people report, like Steve, that HT causes issues with his rig so he is probably only crunching 4 at a time. Personally I agree with the inefficient rating on running more than one per core even with HT on, because shouldn't Seti be using all the clocks when crunch and HT is effectively stealing spare clock cycles which shouldn't be there, or does Seti still have execution space on the processor it isn't using? I guess what I'm trying to figure out is if an HT processor is making room for the extra instance or if it has room for it to begin with. Not entirely learned on the fine details of HT. In execution it seems to work well, but logically my brain is saying it shouldn't be done, but then again no software is without flaw or substance.

---

Traveling through space at ~67,000mph!

Thanks. Got another question: I understand that the top CPUs usually compute eight or maybe 16 instances. How many instances do the top GPUs crunch?

The CPU machines usually compute the same number as they have CPU cores (real or Hyper-Threading) So those top machines usually have multiple CPUs, each with multiple cores, and possibly hyper-threading too. (16 tasks is probably 2 x quad core Xeons with hyperthreading)

Although the GPUs have multiple arithmetic units only the top ones actually have multiple GPU chips, which enable them to do several units concurrently. A standard GPU cruncher uses the CPU to break down the work unit into multiple sub-tasks that can be sent to the GPU to do the actual maths, with lots of calculations being done in parallel by the GPU.

The faster throughput of the GPU is because it may have 100 or more actual arithmetic units that can all be used in parallel, where the CPU has one per core. How much speed up depends on the actual functions being processed. Some are relatively simple to break down into sub-task and calculate in parallel. Other jobs need to be worked though in sequence, where the next task depends on the outcome of the last one, so the GPU has little advantage there. This is why some projects get a HUGE boost from a GPU, others (like Seti) get a useful improvement, and for others it would be a waste of time.

Ian

.....(16 tasks is probably 2 x quad core Xeons with hyperthreading)....

When my money hits right and I get my gaming machine where I want it, I want to build myself a dual xeon machine for crunching. Would be soooo nice. Make one hell of a media server too.

---

Traveling through space at ~67,000mph!

I want to build the twin xeon for gaming, video editing, and crunching.

But I do need to save some money first.

---

Janice

Xeon rigs would not be my first choice in gaming rigs. Too expensive and the motherboards are horrid for gaming, bad chip sets especially if you want to run an SLi rig. However it would make one heck of a cruncher. Really wish you could find an x58 board that supported two processors and the i7 had the interconnects there. Man dual i7's with quad gpus would make one hell of a RAC, not to mention heater.

---

Traveling through space at ~67,000mph!

My rig does gamimg, video editing, and crunching all at the same time. I only shut down crunching if I'm doing a very intensive game, like Dirt2, Crysis, or H.A.W.X. 2. I do create and burn DVD's under full crunch without any problems. I also do Starry Night Pro Plus, or Adobe Photoshop under full crunch.

Steve

---

Warning, addicted to SETI crunching!
Crunching as a member of GPU Users Group.
GPUUG Website

My rig does gamimg, video editing, and crunching all at the same time. I only shut down crunching if I'm doing a very intensive game, like Dirt2, Crysis, or H.A.W.X. 2. I do create and burn DVD's under full crunch without any problems. I also do Starry Night Pro Plus, or Adobe Photoshop under full crunch.

Steve

*Whispers to Steve* My core 2 quad does the same.

Nothing to see here move along! (black helicopters in the background fly by)

---

Traveling through space at ~67,000mph!

steve's not gota rig its an monster with all sorts of things hanging out and more in than a rush hour subway

Xeon rigs would not be my first choice in gaming rigs. Too expensive and the motherboards are horrid for gaming, bad chip sets especially if you want to run an SLi rig. However it would make one heck of a cruncher. Really wish you could find an x58 board that supported two processors and the i7 had the interconnects there. Man dual i7's with quad gpus would make one hell of a RAC, not to mention heater.

Well I don't see gaming being a problem with this board: http://www.hardwaresecrets.com/article/EVGA-Classified-SR-2-Motherboard/1029

Dual Xeon Processors, 7 PCI Express slots, 2 NF200 Chips.

(Don't forget the house-sized case and insane power requirements) :D

---

Well I don't see gaming being a problem with this board: http://www.hardwaresecrets.com/article/EVGA-Classified-SR-2-Motherboard/1029

Dual Xeon Processors, 7 PCI Express slots, 2 NF200 Chips.

(Don't forget the house-sized case and insane power requirements) :D

Indeed, also don't forget the house sized price tag of $600-700! I totally forgot about that board though, makes me want to see if there are any other boards like this out there. Someone had built and benchmarked the "ultimate" game machine with a tri-SLi 480 setup and Xeon's but now I can't find it to save my life. Want to say it was PC mag, but I'm not sure.

---

Traveling through space at ~67,000mph!

Thanks, everyone, for your useful information. Clyde

---