Any advice GTX580's vs GTX670's? My builder wants me to "upgrade" from GTX580's with liquid cooler to GTX670's with air cooled system.

Thanks.

---

Please check out my blog
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

The new ones are a bit faster and consume less power. But not that much.. I'd only consider the deal if you can get a good rpice for your old ones. Max 100$ per card upgrade.

Noise can also be a major concern. I guess you're using the water cooling due to noise reasons. I'm using high-end air cooling (Thermalright Shaman) and couldn't stand a stock blower type cooler, they've always been far too loud for my main rig.

MrS

---

Scanning for our furry friends since Jan 2002

Thanks so much for your reply. I was using sealed liquid cooler on the GTX 580's because that was the builder's (Maingear, Kenilworth, NJ, USA) recommendation. They are replacing my whole Shift Super Stock machine because it seems to be a "lemon" (no reflections on the fruit). The "upgrade" will be at their expense, not mine, the cooling choice theirs, their expertise.

I cannot say enough about the performance of the GTX580's. I went from 7,000,ooo credits in March 2012, gathered since 2007, to my current 25,000,000 credits crunching GPUGrid, Einstein, Milkyway and SETI on GPU.

And, I cannot say enough good things about Maingear. They have stood behind and are standing behind their guarantee.

My concern about the 670's is that the latest is not necessarily the greatest. Your response does not imply any problem in accepting the "upgrade".

I appreciate it very much.

---

Please check out my blog
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

For a free upgrade I would take it. Faster, pretty silent, and more power efficient. Do which brand they would be giving you?

EVGA , GB, MSI?

Sorry, I am going to show my ignorance, but I assume Nvidia, as were the GTX580's.

---

Please check out my blog
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

Yes, nVidia makes the chip. The GPU board is put together by someone else.. you'll find out if you get them :)

And don't run Milkyway on nVidia, neither the GTX580 nor the GTX670. AMDs are much more efficient at this project, but can't run e.g. GPU-Grid at all.

MrS

---

Scanning for our furry friends since Jan 2002

Yes that is a good point. I would highly recommend not running Milkyway. The 6xx series has abysmal OpenCl performance. This is what Milkyway uses.

---

Interesting about Milkyway, I did very well with it, you guys think mostly CPU?

I appreciate the help. The builder, Maingear, is a master builder, but know nothing of BOINC (although I am trying to teach them), so when they say they want to "upgrade" something on my machine, I pretty much cave in, unless I can find knowledgeable help to say no.

---

Please check out my blog
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

It's not the CPU. The 5xx series gpus do decent on Milkyway but ATI (AMD Radeon) do better. I would recommend switching to the 670, since you will see a performance increase on all of those projects except Milkyway. Also, if I'm not mistaken you also crunch for wcg. When they release their gpu app for HCC, the 670 will perform worse than a 580, since those also use OpenCl, just like Milkyway.

In short, personally I would take the 670s since you said they would be free. The difference is noticeable, especially here and at einstein. But should you decide to keep crunching Milkyway they will perform worse for that project only.

I was coming back to say, silly me, sure I saw Milkyway on GPU when I looked at Active tasks.

---

Please check out my blog
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

BTW: the problem with MW is not OpenCL, but double precision (DP). The big nVidia chips in the 400 and 500 series could do this at 1/2 the single precision (SP) performance (which is used at GPU-Grid and most other projects), but artificially capped DP performance at 1/8 of SP to protect their Tesla sales. Lower chips are at 1/12. GTX670 (and similar cards) run DP at 1/24 the SP performance. It's really only meant for development or an occasional instruction.

The high end AMDs since the 4000 series run DP at 1/5 or 1/4 the SP rate. Middle and low-end chips used to not support it all, but new ones (78x0, 77x0 and 75x0) support it at 1/16 the SP rate.

MrS

---

Scanning for our furry friends since Jan 2002

Lol. Wow. Too many projects and im beginning to get them confused.

If you are referring to my project list, I am crunching on seven machines. Six are CPU only, three hyper threaded quads and one hyper threaded dual core, one dual core, one ATOM, also hyper thread. The ATOM has two projects, the dual core has three, the hyper threaded dual core has four, the quads have eight. No project is on more than four machines, except SETI, to which we owe a huge debt, and WCG, which is, I believe crimped by BOINC considering it all as one project. WCG also has a resource share of 200.

---

Please check out my blog
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

My 680 and my 570 are roughly the same speed per WU at PrimeGrid on the Double Precision Cuda GenFer WUs. On anything Single Precision "as are most projects" my 680 is about 40%-50% faster. 600s series "Keplers" are much better for power usage also.
600s will run Milky Way fine but you will never excell against an ATI card there.
My EVGA 680 starts automatically downclocking at 70c so I keep it around 67c with 50% fan at 100% utilization. Heat is never a problem with it. I got the superclocked EVGA Signature 2 series with 2 fans and highly reccomend it.

Please increase your fan speed.

600s will run Milky Way fine but you will never excell against an ATI card there.

-> "600s will run Milky Way fine but doing so wastes valuable ressources" ;)

MrS

---

Scanning for our furry friends since Jan 2002