Motherboard for AMD FX8350

fivemack · 2012-10-23, 10:35

What would a reasonable integrated-graphics motherboard for this chip be?

(on a different note, has anyone got gnfs-lasieve4I15e performance figures for the 8150? I can't see anyone with an 8150 in the stats at http://www.allprojectstats.com/top.p...jekt=91&type=0)

Yes, I know it will not offer good prime95 performance, but eight fast cores with 2MB L2 per pair seems not unreasonable for sieving (although the L1D cache is as far as I can tell down to 16kb, which is not great)

pinhodecarlos · 2012-10-23, 10:44

8150 here (http://escatter11.fullerton.edu/nfs/...p?hostid=27544) but my old core i5 clearly beats it when I running gnfs-lasieve4I14e siever.

fivemack · 2012-10-23, 10:54

OK, scratch that idea

pinhodecarlos · 2012-10-23, 10:59

Quote:

Originally Posted by fivemack

OK, scratch that idea

If you want I can try to run a few 14e wu's so that you can compare to Lionel's 8150 machine. If you go here (http://escatter11.fullerton.edu/nfs/top_hosts.php) you will see my quad-core machine (overclocked to 3.4 GHz) in TOP 20, it is the 28521 host.
I think the best option is i7-3770K, it is tremendous powerful when overclocked, I think it can easily double my machine output with less energy consumption.

Dubslow · 2012-10-23, 20:09

The thing though is that they aren't truly 8 cores; they're 8 integer processors with only 4 FP processors. 2I+1F is called a "module" in AMD terminology. It's really more of "4 cores + 4 half-cores", but they sell it as 8 anyways.

Edit: The Piledriver prices seem much better.

Here's a good description of the architecture.

fivemack · 2012-10-23, 20:56

Yes, each pair of cores shares a single FP unit, but GNFS sieving doesn't use FP so for that particular task it shouldn't be a problem.

On the other hand the L1 caches on each core are small (16kb); whilst lasieve4 can be configured with various L1 cache sizes, I think the executables nfs@home use are configured for 32kb.

(I'm a little confused in that case as to why I get such good performance on hyperthreads on Sandy Bridge - my suspicion is that the small fast per-core L2 caches handle the L1 misses better than the big shared L2 on AMD processors)

2012-10-23, 10:35	#1
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 1913₁₆ Posts	Motherboard for AMD FX8350 What would a reasonable integrated-graphics motherboard for this chip be? (on a different note, has anyone got gnfs-lasieve4I15e performance figures for the 8150? I can't see anyone with an 8150 in the stats at http://www.allprojectstats.com/top.p...jekt=91&type=0) Yes, I know it will not offer good prime95 performance, but eight fast cores with 2MB L2 per pair seems not unreasonable for sieving (although the L1D cache is as far as I can tell down to 16kb, which is not great)

2012-10-23, 10:44	#2
pinhodecarlos "Carlos Pinho" Oct 2011 Milton Keynes, UK 3×17×97 Posts	8150 here (http://escatter11.fullerton.edu/nfs/...p?hostid=27544) but my old core i5 clearly beats it when I running gnfs-lasieve4I14e siever. Last fiddled with by pinhodecarlos on 2012-10-23 at 11:13

2012-10-23, 20:09	#5
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3×29×83 Posts	The thing though is that they aren't truly 8 cores; they're 8 integer processors with only 4 FP processors. 2I+1F is called a "module" in AMD terminology. It's really more of "4 cores + 4 half-cores", but they sell it as 8 anyways. Edit: The Piledriver prices seem much better. Here's a good description of the architecture. Last fiddled with by Dubslow on 2012-10-23 at 20:26

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Motherboard for Xeon	ATH	Hardware	7	2015-10-10 02:13
GPU placement on motherboard	TheMawn	GPU Computing	7	2013-08-17 04:43
Help choosing motherboard please.	Flatlander	GPU Computing	4	2011-01-26 08:15
Motherboard Selection Help	jugbugs	Hardware	13	2004-06-04 15:59
Silicone on Motherboard?	clowns789	Hardware	6	2003-06-23 01:57

2012-10-23, 10:54	#3
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 1100100010011₂ Posts	OK, scratch that idea

2012-10-23, 20:56	#6
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 7²·131 Posts	Yes, each pair of cores shares a single FP unit, but GNFS sieving doesn't use FP so for that particular task it shouldn't be a problem. On the other hand the L1 caches on each core are small (16kb); whilst lasieve4 can be configured with various L1 cache sizes, I think the executables nfs@home use are configured for 32kb. (I'm a little confused in that case as to why I get such good performance on hyperthreads on Sandy Bridge - my suspicion is that the small fast per-core L2 caches handle the L1 misses better than the big shared L2 on AMD processors)