Intel Avoton

fivemack · 2013-12-11, 09:30

I've just picked up an Atom C2750 system.

It's about the size of a cereal box; uses 20W idle and 32W running eight threads of ecm; ecm performance with eight threads is 60% that of my i7/4770 with eight threads. Unfortunately it heats up at 4C per minute unless I put a fan on top of the heat-sink, but 10cfm of air flow keeps it beautifully cool.

Single threads are quite slow - slower than the 2,4GHz Q6600 was, Silvermont cores aren't as performant as Merom cores - but I've got eight of them using really very little energy.

Unfortunately the machine cost about the same as the i7/4770 did - the board is exotic enough that I had to buy it through a friend with a small business with an account with a Supermicro reseller, and it takes ECC SODIMMs. It has quad gigabit ethernet; as far as I can tell all this does for me is add the extra risk that some day Linux boot will rearrange the port numbering so eth0 is no longer the one with the cable plugged into it.

It requires ubuntu-13.10 or later; ubuntu-13.04 installer kept failing mysteriously, and at the rescue command prompt 'cat /proc/cpuinfo' told me the machine only had one core.

Obviously what I really want is six Atom C2750 sharing a PSU and case; this could be done for £1000 but I don't think anyone's planning to do it.

tului · 2013-12-12, 17:26

Saving my ducats in hopes of 16c/32t Haswell EP dual or quad socket. Oh yes. A man can dream. Still, a dense, decent IPC system like that sounds pretty good too.

VictordeHolland · 2013-12-12, 18:27

That is some nice energy efficiency you got there!

I was looking at Intel ARKs and it states that processor only has the 64bit instruction set. Since it usually states SSE, AVX or something like that, I assume it has none of those? http://ark.intel.com/products/77987
If that is the case, it won't be able to do P-1 and LL/DC test nearly as fast as NFS and ECM I suppose.

If only the price would be a bit lower....
The chip+SuperMicro board is €390 in the Netherlands, add +/- €130 for 2x8GB DDR3L-1600 SODIMMs. Something like €520 in total for CPU+mobo+mem combo.

You could get a 4770(non-k) for €260 or €190 for a 4670(non-k). €120 for 2x8GB DDR3-1600, which leaves €140 for a decent motherboard and a CPU cooler, that is certainly possible.
The 4770 combo will use about 110W or so under load, but can be used for more diverse purposes.

sdbardwick · 2013-12-12, 20:23

CPU World is generally reliable WRT chip capabilities. Looks like C2000 series supports up to SSE4.2, so no AVX or FMA3.

Mark Rose · 2013-12-12, 20:28

Quote:

Originally Posted by VictordeHolland

You could get a 4770(non-k) for €260 or €190 for a 4670(non-k). €120 for 2x8GB DDR3-1600, which leaves €140 for a decent motherboard and a CPU cooler, that is certainly possible.
The 4770 combo will use about 110W or so under load, but can be used for more diverse purposes.

I'm running mprime 28.1 on a 4770 at stock speeds, Hyper 212 Evo, 32GB @ 1600, 1 ssd, 1 spinning rust, ASUS Z87 Pro, Corsair CX500M power supply, 2 case fans, that consumes 120 watts at the wall with a kill-a-watt meter. Removing the spinning rust will drop 5 watts, removing a couple sticks of RAM would also help, as would using a more efficient/smaller power supply. You could get it down to 100W with an 80 PLUS Platinum supply.

VictordeHolland · 2013-12-13, 02:37

Quote:

Originally Posted by sdbardwick

CPU World is generally reliable WRT chip capabilities. Looks like C2000 series supports up to SSE4.2, so no AVX or FMA3.

That makes this board a lot more interesting!

Fivemack, could you post more detailed benchmarks? I'm just curieus how it performs on other tasks :).

henryzz · 2013-12-13, 11:16

Windows 8 actually requires SSE2 so all cpus should have it from now on.

fivemack · 2013-12-13, 12:12

Which benchmarks would interest you? Bear in mind I don't own a Windows license; I can offer gnfs-lasieve4I14e, msieve and mprime results.

I've turned the machine off for the moment, because the small 8000rpm fan that I was using for air-flow made it very difficult to think while in the same room as the computer; a better fan should be appearing sometime next week.

VictordeHolland · 2013-12-13, 14:34

Could you do the Prime95 benchmark, or is that not included in Mprime?

Electricity is quite expensive in the Netherlands (it is heavily taxed), €0.22/kwh if I'm remembering correctly. So running one of these might just be more economically efficient than a 4770.
35W x 24h x 365days = ~307kwh or ~€68 a year
110W x 24h x 365days = ~964kwh or ~€212 a year

ldesnogu · 2013-12-13, 22:27

Quote:

Originally Posted by fivemack

Which benchmarks would interest you?

SPEC 2000 and SPEC 2006. Bring it to work, I'm sure you'll find people able to help you on that.

fivemack · 2013-12-13, 22:42

prime95 benchmark:

Code:

Intel(R) Atom(TM) CPU  C2750  @ 2.41GHz
CPU speed: 2557.83 MHz, 8 cores
CPU features: Prefetch, SSE, SSE2, SSE4
L1 cache size: 24 KB
L2 cache size: 1 MB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
TLBS: 128
Prime95 64-bit version 27.9, RdtscTiming=1
Best time for 768K FFT length: 29.983 ms., avg: 30.016 ms.
Best time for 896K FFT length: 40.490 ms., avg: 40.545 ms.
Best time for 1024K FFT length: 38.525 ms., avg: 38.575 ms.
Best time for 1280K FFT length: 50.985 ms., avg: 51.025 ms.
Best time for 1536K FFT length: 64.529 ms., avg: 64.555 ms.
Best time for 1792K FFT length: 80.745 ms., avg: 80.867 ms.
Best time for 2048K FFT length: 84.762 ms., avg: 84.783 ms.
Best time for 2560K FFT length: 103.742 ms., avg: 103.927 ms.
Best time for 3072K FFT length: 134.613 ms., avg: 134.720 ms.
Best time for 3584K FFT length: 164.774 ms., avg: 164.825 ms.
Best time for 4096K FFT length: 173.640 ms., avg: 173.707 ms.
Best time for 5120K FFT length: 231.854 ms., avg: 231.913 ms.
Best time for 6144K FFT length: 285.665 ms., avg: 285.746 ms.
Best time for 7168K FFT length: 433.498 ms., avg: 434.008 ms.
Best time for 8192K FFT length: 370.349 ms., avg: 370.561 ms.
Timing FFTs using 2 threads.
Best time for 768K FFT length: 15.362 ms., avg: 15.419 ms.
Best time for 896K FFT length: 22.276 ms., avg: 22.343 ms.
Best time for 1024K FFT length: 19.808 ms., avg: 19.848 ms.
Best time for 1280K FFT length: 26.182 ms., avg: 26.236 ms.
Best time for 1536K FFT length: 33.567 ms., avg: 33.616 ms.
Best time for 1792K FFT length: 44.518 ms., avg: 44.581 ms.
Best time for 2048K FFT length: 49.014 ms., avg: 49.071 ms.
Best time for 2560K FFT length: 55.743 ms., avg: 55.806 ms.
Best time for 3072K FFT length: 76.805 ms., avg: 77.687 ms.
Best time for 3584K FFT length: 92.031 ms., avg: 92.130 ms.
Best time for 4096K FFT length: 102.465 ms., avg: 102.528 ms.
Best time for 5120K FFT length: 135.553 ms., avg: 135.651 ms.
Best time for 6144K FFT length: 168.475 ms., avg: 168.572 ms.
Best time for 7168K FFT length: 260.328 ms., avg: 260.988 ms.
Best time for 8192K FFT length: 230.479 ms., avg: 230.556 ms.
Timing FFTs using 3 threads.
Best time for 768K FFT length: 10.611 ms., avg: 10.681 ms.
Best time for 896K FFT length: 15.145 ms., avg: 15.215 ms.
Best time for 1024K FFT length: 13.518 ms., avg: 13.572 ms.
Best time for 1280K FFT length: 17.828 ms., avg: 17.993 ms.
Best time for 1536K FFT length: 22.505 ms., avg: 22.669 ms.
Best time for 1792K FFT length: 29.790 ms., avg: 29.888 ms.
Best time for 2048K FFT length: 32.257 ms., avg: 32.430 ms.
Best time for 2560K FFT length: 37.192 ms., avg: 37.278 ms.
Best time for 3072K FFT length: 50.394 ms., avg: 50.440 ms.
Best time for 3584K FFT length: 60.280 ms., avg: 60.442 ms.
Best time for 4096K FFT length: 65.845 ms., avg: 65.913 ms.
Best time for 5120K FFT length: 86.996 ms., avg: 87.048 ms.
Best time for 6144K FFT length: 107.230 ms., avg: 107.373 ms.
Best time for 7168K FFT length: 173.414 ms., avg: 173.645 ms.
Best time for 8192K FFT length: 141.248 ms., avg: 141.335 ms.
Timing FFTs using 4 threads.
Best time for 768K FFT length: 7.997 ms., avg: 8.044 ms.
Best time for 896K FFT length: 11.920 ms., avg: 12.001 ms.
Best time for 1024K FFT length: 10.228 ms., avg: 10.319 ms.
Best time for 1280K FFT length: 13.530 ms., avg: 13.589 ms.
Best time for 1536K FFT length: 17.148 ms., avg: 17.227 ms.
Best time for 1792K FFT length: 23.060 ms., avg: 23.160 ms.
Best time for 2048K FFT length: 25.743 ms., avg: 25.836 ms.
Best time for 2560K FFT length: 28.496 ms., avg: 28.560 ms.
Best time for 3072K FFT length: 41.473 ms., avg: 41.640 ms.
Best time for 3584K FFT length: 47.136 ms., avg: 47.250 ms.
Best time for 4096K FFT length: 52.966 ms., avg: 53.194 ms.
Best time for 5120K FFT length: 70.786 ms., avg: 70.887 ms.
Best time for 6144K FFT length: 87.231 ms., avg: 87.347 ms.
Best time for 7168K FFT length: 145.186 ms., avg: 145.423 ms.
Best time for 8192K FFT length: 114.705 ms., avg: 114.796 ms.
Timing FFTs using 5 threads.
Best time for 768K FFT length: 6.580 ms., avg: 6.693 ms.
Best time for 896K FFT length: 9.693 ms., avg: 9.849 ms.
Best time for 1024K FFT length: 8.446 ms., avg: 8.505 ms.
Best time for 1280K FFT length: 11.136 ms., avg: 11.204 ms.
Best time for 1536K FFT length: 13.897 ms., avg: 14.046 ms.
Best time for 1792K FFT length: 18.405 ms., avg: 18.674 ms.
Best time for 2048K FFT length: 20.326 ms., avg: 20.759 ms.
Best time for 2560K FFT length: 22.975 ms., avg: 23.153 ms.
Best time for 3072K FFT length: 33.160 ms., avg: 33.266 ms.
Best time for 3584K FFT length: 37.202 ms., avg: 37.797 ms.
Best time for 4096K FFT length: 41.982 ms., avg: 42.156 ms.
Best time for 5120K FFT length: 54.826 ms., avg: 55.125 ms.
Best time for 6144K FFT length: 68.186 ms., avg: 68.248 ms.
Best time for 7168K FFT length: 120.475 ms., avg: 120.560 ms.
Best time for 8192K FFT length: 91.575 ms., avg: 91.856 ms.
Timing FFTs using 6 threads.
Best time for 768K FFT length: 5.670 ms., avg: 5.766 ms.
Best time for 896K FFT length: 8.291 ms., avg: 8.595 ms.
Best time for 1024K FFT length: 7.172 ms., avg: 7.273 ms.
Best time for 1280K FFT length: 9.451 ms., avg: 9.502 ms.
Best time for 1536K FFT length: 12.076 ms., avg: 12.175 ms.
Best time for 1792K FFT length: 15.885 ms., avg: 16.008 ms.
Best time for 2048K FFT length: 18.070 ms., avg: 18.429 ms.
Best time for 2560K FFT length: 19.526 ms., avg: 19.639 ms.
Best time for 3072K FFT length: 28.947 ms., avg: 29.230 ms.
Best time for 3584K FFT length: 32.725 ms., avg: 33.037 ms.
Best time for 4096K FFT length: 36.957 ms., avg: 37.287 ms.
Best time for 5120K FFT length: 49.353 ms., avg: 49.805 ms.
Best time for 6144K FFT length: 61.004 ms., avg: 61.391 ms.
Best time for 7168K FFT length: 117.642 ms., avg: 117.809 ms.
Best time for 8192K FFT length: 81.761 ms., avg: 82.102 ms.
Timing FFTs using 7 threads.
Best time for 768K FFT length: 5.132 ms., avg: 5.221 ms.
Best time for 896K FFT length: 7.212 ms., avg: 7.471 ms.
Best time for 1024K FFT length: 6.403 ms., avg: 6.532 ms.
Best time for 1280K FFT length: 8.229 ms., avg: 8.334 ms.
Best time for 1536K FFT length: 10.362 ms., avg: 10.516 ms.
Best time for 1792K FFT length: 13.993 ms., avg: 14.147 ms.
Best time for 2048K FFT length: 16.054 ms., avg: 16.294 ms.
Best time for 2560K FFT length: 17.005 ms., avg: 17.100 ms.
Best time for 3072K FFT length: 24.632 ms., avg: 24.879 ms.
Best time for 3584K FFT length: 27.618 ms., avg: 27.951 ms.
Best time for 4096K FFT length: 32.041 ms., avg: 32.310 ms.
Best time for 5120K FFT length: 41.953 ms., avg: 42.364 ms.
Best time for 6144K FFT length: 51.665 ms., avg: 52.022 ms.
Best time for 7168K FFT length: 107.093 ms., avg: 107.704 ms.
Best time for 8192K FFT length: 68.442 ms., avg: 69.024 ms.
Timing FFTs using 8 threads.
Best time for 768K FFT length: 4.512 ms., avg: 4.636 ms.
Best time for 896K FFT length: 6.647 ms., avg: 6.816 ms.
Best time for 1024K FFT length: 5.611 ms., avg: 5.744 ms.
Best time for 1280K FFT length: 7.381 ms., avg: 7.460 ms.
Best time for 1536K FFT length: 9.323 ms., avg: 9.468 ms.
Best time for 1792K FFT length: 12.812 ms., avg: 13.047 ms.
Best time for 2048K FFT length: 15.092 ms., avg: 15.263 ms.
Best time for 2560K FFT length: 15.270 ms., avg: 15.433 ms.
Best time for 3072K FFT length: 23.370 ms., avg: 23.526 ms.
Best time for 3584K FFT length: 25.182 ms., avg: 25.791 ms.
Best time for 4096K FFT length: 29.965 ms., avg: 30.495 ms.
Best time for 5120K FFT length: 39.626 ms., avg: 40.108 ms.
Best time for 6144K FFT length: 48.869 ms., avg: 49.139 ms.
Best time for 7168K FFT length: 103.284 ms., avg: 104.345 ms.
Best time for 8192K FFT length: 66.686 ms., avg: 68.078 ms.
Best time for 61 bit trial factors: 7.432 ms.
Best time for 62 bit trial factors: 7.592 ms.
Best time for 63 bit trial factors: 8.820 ms.
Best time for 64 bit trial factors: 11.852 ms.
Best time for 65 bit trial factors: 14.856 ms.
Best time for 66 bit trial factors: 15.797 ms.
Best time for 67 bit trial factors: 15.741 ms.
Best time for 75 bit trial factors: 15.421 ms.
Best time for 76 bit trial factors: 15.382 ms.
Best time for 77 bit trial factors: 15.384 ms.

These are of course pretty awful; the Haswell time at 768k is 3.511ms on one thread, 0.954ms on eight. The i7/4930K time is 4.386ms on one thread, 0.904ms on twelve.

2013-12-11, 09:30	#1
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 3×2,141 Posts	Intel Avoton I've just picked up an Atom C2750 system. It's about the size of a cereal box; uses 20W idle and 32W running eight threads of ecm; ecm performance with eight threads is 60% that of my i7/4770 with eight threads. Unfortunately it heats up at 4C per minute unless I put a fan on top of the heat-sink, but 10cfm of air flow keeps it beautifully cool. Single threads are quite slow - slower than the 2,4GHz Q6600 was, Silvermont cores aren't as performant as Merom cores - but I've got eight of them using really very little energy. Unfortunately the machine cost about the same as the i7/4770 did - the board is exotic enough that I had to buy it through a friend with a small business with an account with a Supermicro reseller, and it takes ECC SODIMMs. It has quad gigabit ethernet; as far as I can tell all this does for me is add the extra risk that some day Linux boot will rearrange the port numbering so eth0 is no longer the one with the cable plugged into it. It requires ubuntu-13.10 or later; ubuntu-13.04 installer kept failing mysteriously, and at the rescue command prompt 'cat /proc/cpuinfo' told me the machine only had one core. Obviously what I really want is six Atom C2750 sharing a PSU and case; this could be done for £1000 but I don't think anyone's planning to do it. Last fiddled with by fivemack on 2013-12-11 at 11:19

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
AMD vs Intel	dtripp	Software	3	2013-02-19 20:20
Intel NUC	nucleon	Hardware	2	2012-05-10 23:53
Intel RNG API?	R.D. Silverman	Programming	19	2011-09-17 01:43
AMD or Intel	mack	Information & Answers	7	2009-09-13 01:48
Intel Mac?	penguain	NFSNET Discussion	0	2006-06-12 01:31

2013-12-12, 17:26	#2
tului Jan 2013 2²×17 Posts	Saving my ducats in hopes of 16c/32t Haswell EP dual or quad socket. Oh yes. A man can dream. Still, a dense, decent IPC system like that sounds pretty good too.

2013-12-12, 18:27	#3
VictordeHolland "Victor de Hollander" Aug 2011 the Netherlands 2230₈ Posts	That is some nice energy efficiency you got there! I was looking at Intel ARKs and it states that processor only has the 64bit instruction set. Since it usually states SSE, AVX or something like that, I assume it has none of those? http://ark.intel.com/products/77987 If that is the case, it won't be able to do P-1 and LL/DC test nearly as fast as NFS and ECM I suppose. If only the price would be a bit lower.... The chip+SuperMicro board is €390 in the Netherlands, add +/- €130 for 2x8GB DDR3L-1600 SODIMMs. Something like €520 in total for CPU+mobo+mem combo. You could get a 4770(non-k) for €260 or €190 for a 4670(non-k). €120 for 2x8GB DDR3-1600, which leaves €140 for a decent motherboard and a CPU cooler, that is certainly possible. The 4770 combo will use about 110W or so under load, but can be used for more diverse purposes.

2013-12-12, 20:23	#4
sdbardwick Aug 2002 North San Diego County 1010101101₂ Posts	CPU World is generally reliable WRT chip capabilities. Looks like C2000 series supports up to SSE4.2, so no AVX or FMA3.

2013-12-13, 11:16	#7
henryzz Just call me Henry "David" Sep 2007 Cambridge (GMT/BST) 5886₁₀ Posts	Windows 8 actually requires SSE2 so all cpus should have it from now on.

2013-12-13, 12:12	#8
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 3·2,141 Posts	Which benchmarks would interest you? Bear in mind I don't own a Windows license; I can offer gnfs-lasieve4I14e, msieve and mprime results. I've turned the machine off for the moment, because the small 8000rpm fan that I was using for air-flow made it very difficult to think while in the same room as the computer; a better fan should be appearing sometime next week.

2013-12-13, 14:34	#9
VictordeHolland "Victor de Hollander" Aug 2011 the Netherlands 498₁₆ Posts	Could you do the Prime95 benchmark, or is that not included in Mprime? Electricity is quite expensive in the Netherlands (it is heavily taxed), €0.22/kwh if I'm remembering correctly. So running one of these might just be more economically efficient than a 4770. 35W x 24h x 365days = ~307kwh or ~€68 a year 110W x 24h x 365days = ~964kwh or ~€212 a year