Bay Trail and Folding

[rwh202]

Just thought that I'd post my recent experience of folding on an Intel J2900 powered desktop. For those that don't know, the J2900 is a low power system on a chip, incorporating most of the controllers and IO as well as the quad core CPU and GPU on a single package, normally intended to be used in tablets and netbooks.

I got a Lenovo E50 tower that came with windows 7/8 professional. It's available cheaper with windows 8.1. The power supply is a 65 W laptop one and the only fan is a small one over the processor. Inside a large case and with the benefits of the fan, the processor is able to run constantly at its maximum frequency of 2.67 GHz whilst staying around 40 degC. The whole system draws less than 20 W from the wall when folding and that's with a spinning 3.5 inch HD. I intend to install Mint on a USB stick and see if that can bring the power down.

The four atom cores aren't as powerful as those in the bigger i7 chips but they run at a healthy clock speed. Benchmarks put the performance below that of a dual core i3 and SMP folding using the V7 client seems to support that - I've seen between 2500 and 3500 ppd depending on WU. However, running the Chrome NaCl client the performance is a lot better - over 10k PPD with each WU taking around 17 minutes. I'm guessing that the Chrome version isn't able to draw on the added features and powers of the i7 type cores and so it is handicapped less.

I realise that 500 PPD/W is not in Maxwell territories for efficiency, but I was encouraged to find a way to fold CPU WUs for low cost and not be wasting electricity that could better serve GPUs. I'm now tempted to see if I can find some inexpensive mini-ITX boards with these processors and fit 4 or so into a standard case with a high efficiency PSU and possibly exceed 1000 PPD/W.

[Napoleon]

I'm guessing that the Chrome version isn't able to draw on the added features and powers of the i7 type cores and so it is handicapped less.

From NaCl log:

Code: Select all

main.js:77 DEBUG:                 :-)  G  R  O  M  A  C  S  (-:
main.js:77 DEBUG:                 Groningen Machine for Chemical Simulation
main.js:77 DEBUG:                 :-)  VERSION 4.6.5  (-:

IIRC, that's what current A? cores are based on, so not much of a difference there. I've been under the impression that NaCl is somewhat separate from the unified benchmarking scheme. No QRB but the base points are high. I guess PG is trying to steer mid/low range CPUs towards NaCl. My old 2C/4T Atom330 is still able to get over 2500 PPD from NaCl (overclocked 1.6GHz 2.0GHz) but is just barely capable of completing SMP WUs before Timeout. Usually a NaCl WU takes about 1 hour.

That's Moore's law in action for you: minor increase in (TDP) wattage, but you get twice as many real CPU cores, twice as much cache, higher clocks, 4x overall performance and an iGPU in the bargain.

IMHO, it isn't much of a surprise that the NaCl servers are overloaded, given the hefty head start NaCl PPD has over traditional QRB WUs.

NaCl seems to require a bit of babysitting, though. Looks like the Chrome process which spawns NaCl starts taking up more and more of CPU time with each returned WU. So you'll want to finish a WU, close the NaCl tab and restart NaCl every now and then. Furthermore, NaCl acts like the v6 client when it comes to uploads and downloads. They aren't concurrent, so PPD of a very fast CPU would become more or less capped by internet connection and/or NaCl servers not being able to keep up with it.

I'm now tempted to see if I can find some inexpensive mini-ITX boards with these processors and fit 4 or so into a standard case with a high efficiency PSU and possibly exceed 1000 PPD/W.

Could get a bit complicated. The efficiency curve isn't a flat horizontal line. More importantly, (switching) PSU rails tend to require a minimum load for stable operation!

[readbueno]

Hi. I had wondered why the NaCl slowed dowṇ. It appears that this program like chrome, does not tidy up after itself and fills the computer with various amounts of rubbish every time you get a new work uniṭ. My acer travelmate laptop starts doing a WU in about 45 minutes, but after it has done several, it slows down to many hours or even days for a work uniṭ. My desktop running ubuntu14.04 seems to be just whizzing along in comparison. Question? is it worth shutting down, cleaning up, and restarting after several WUs before this process slows the computer to a standstill and what would be the optimum timing for this?

rb.

[ChristianVirtual]

[hijack]Anyone any experiences with Xeon D like http://ark.intel.com/products/87039/Int ... e-2_00-GHz ?[/hijack]

[rwh202]

NaCl seems to require a bit of babysitting, though. Looks like the Chrome process which spawns NaCl starts taking up more and more of CPU time with each returned WU. So you'll want to finish a WU, close the NaCl tab and restart NaCl every now and then. Furthermore, NaCl acts like the v6 client when it comes to uploads and downloads. They aren't concurrent, so PPD of a very fast CPU would become more or less capped by internet connection and/or NaCl servers not being able to keep up with it.

I don't know how OS / chrome version dependent this is. Recently I've not been seeing so many issues - I've got a few windows 7 and OS X systems that have months of running with no apparent slowdown. They even happily rode out the assignment problems and picked up again afterwards. However, in the early days there were definite problems where the client would hang, fill up disks etc. etc. Then, I had a script to just kill and relaunch chrome once a day.
To get around the lack of concurrent upload and download, I've toyed with the idea of running two instances of the client (VMs if necessary) but I think the overhead will be too high to give any benefit.

[hijack]Anyone any experiences with Xeon D like http://ark.intel.com/products/87039/Int ... e-2_00-GHz ?[/hijack]

They look interesting, but at around $800 vs $150 for a system I don't think I'll be trying one on a whim!