I've got an AMD Athlon XP 64 3200+ on an ASUS K8V-SE socket 754. This is a full ATX board with 1GB 3200 RAM. It crunches great on E@H. It cuts through a traditional WU in about 20300 seconds.
I've also got an AMD Athlon XP 3200+ (NON 64) on an ABIT VA-20 mATX board with 1GB of 3200 RAM. This one "works" but no where nearly as good as the one above. This second machine completes WU's in about 25100 seconds.
My question is, why the big difference in completion times for these two chips? Sure, one is 64 bit and one non-64bit. One is on a ATX mobo and one is on a mATX mobo. Each have a gig of 3200 ram.
I don't think E@H is a 64 bit process. I'm not attempting to overclock either machine. The VA-20 is set at 200 FSB (I think). The VA-20 BIOS doesn't give me much control on things.
I just recently asked this same question on the ocUK forums and one gent felt it was the difference in onboard caches on the 64 vs the non-64.
I look forward to your thoughts on the performance difference and some potential improvements I could make to the VA-20 (non 64) 3200+ setup.
Thanks
Bruce Richardson
Team Richwood Timber LLC
When the going gets weird, the weird turn PRO. -- Hunter S. Thompson
Copyright © 2024 Einstein@Home. All rights reserved.
AMD XP 3200+ vs AMD 64 3200+
)
The XP shows: Measured floating point speed 1685.64 million ops/sec
Measured integer speed 2872.63 million ops/sec
The Athlon 64 shows: Measured floating point speed 2101.64 million ops/sec
Measured integer speed 3893.02 million ops/sec
I don't know what the cache size is on these (this site says both are 1MB, but I don't trust that value, I've seen it be wrong), but basically the 64 is almost half-again faster than the XP. (BTW, the 64 doesn't say "XP" anywhere, and I don't think they mix those two terms in the model names.)
You can look at the huge chart of all AMD processors at tomshardware.com, and compare the two, but if I recall correctly, the AMD 64's are simply much newer technology, smaller, cooler, faster... The AMD 64 3200+ Socket 754 is a very well-respected chip. It would be easy to overclock it just a _little_ bit and make it do even better. (I have a mildly-oc'd socket 939 3700+ myself and love it.) I've never dealt with an XP chip, don't know anything about it's overclocking capability, or your MB. Short of overclocking (or replacing the CPU, duh!), I think both your systems are doing pretty well; about what should be expected.
Thanks for the reply. I went
)
Thanks for the reply. I went to the Tom's Hardware site and looked at the "mother of all charts". You are right--it must just be a substantive difference between the 64 and the XP. I had equated the two based on both having a 3200+ number. But they seem to be much more different than similar.
I got the 64 at a great price. Starting about a year ago, I started hunting for a nice price vs performance point for CPU's on ebay. I found I really liked the XP 3000+. I could find Socket A mobo's for reasonable. I could find cases with powersupplies sufficient for that chip for cheap.
I've got a few more boxes to finish bringing online. Then I need to coast for awhile. That doesn't mean I don't look longingly at the FX 53 and 55 and other powerful CPUs.
Bruce Richardson
Team Richwood Timber LLC
When the going gets weird, the weird turn PRO. -- Hunter S. Thompson
RE: That doesn't mean I
)
The 3700+ oc's to FX53 level for much less cost. What _I_ am looking at longingly is something like a 4400 X2 dual-core... :-)
[offtopic]In the other thread you mentioned a host named Heinlein. As someone whose cat is named Pixel, I approve...[/offtopic]
I do not know enough about
)
I do not know enough about the mother boards to be sure, but the fact that you have different ones can also contribute to the discrepancy. Check both to make sure they have dual channel memory and that it is enabled.
Check to see if the memory timing is the same (if one is substantially lower that can also cause the problems).
In general, there is no performance difference with the current systems that can be traced to the 64-bittedness of a CPU chip. None of the projects have 64 bit applications, and even if they did it is not likely we would see much gain if any at all.
And, the Cache line seems to indicate 1M for all chips regardless of the actual value.