CPU frequency and iteration times.

rx7350 · 2006-05-05, 17:56

I had always thought that a higher cpu frequency would directly result in lower iteration times, but I've observed some odd behavior on some of my nodes, if my assumpton is true. I have a Pentium dual-core running at 2.66GHz on a smaller exponent than an AMD dual-core running at 2.376 Ghz, but the interation times on the Pentium are about 2ms longer than those on the AMD. Both are 10M exponents using the 2048 FFT size. The only noticeable hardware difference is that the Pentium only has 512Mb, whereas the AMD has 2Gb.

Your thoughts?

rx7350 · 2006-05-05, 18:03

Clarification - the exponenets are 10 million-digit exponents, 35M+ on both.

Cruelty · 2006-05-05, 22:34

Quote:

Originally Posted by rx7350

I had always thought that a higher cpu frequency would directly result in lower iteration times

The above would be true if you were comparing two identical architectures e.g. P4 with P4 or A64 with A64...

Anyways, I always thought that A64 is nearly equal in performance under Prime95 with P4 on a clock-for-clock basis... what are your iteration times on both A64 and P4? According to this page you should have iteration times of ~0.0844 for A64 and ~0.0758 for P4.

rx7350 · 2006-05-06, 16:58

A64 X2 4400+ Toledo 8% ovrclk (2.376 Ghz) FSB 1 Ghz
L1 cache 64kb L2 cache 1Mb
2Gb OCZ low latency ram
exponent 35385871 .088 sec
exponent 35385941 .088 sec

Pentium D 805 Smithfield no oc (2.66 Ghz) FSB 533 Mhz
L1 cache 16kb L2 cache 1Mb
512Mb OCZ low latency ram
exponent 35037707 .090 sec
exponent 35171009 .090 sec

The AMD was built to be a household productivity pc as the main determining factor, and I just happen to run Prime95 as one of the many things that are done on this pc.

The Pentium was built to run Prime95 exclusively, and since it's 'widely known(?)' that Pentium cpus are better for running Prime95, that's why I choose that architecture.

My intent is to build additional nodes to run Prime95 exclusively, and the above mentioned Pentium (two nodes) was built for only $332. Obviously, I'd like to get the biggest bang for the buck, and I read everything in these forums to appeared relevant to what I was trying to do. So, I'm a bit mystified as to why the iteration times on the Pentium are not MUCH better than what they are. I'm sure it's because I don't fully understand all of the inter-relationships between the architectural factors that ultimately determine iteration times. I had come to the conclusion that after the intial factoring attempts by Prime95 when first starting a new primality test, that cpu frequency was the most importatn factor by far, but that doesn't seem to be case in this comparison between the two architectures.

axn · 2006-05-06, 19:23

Quote:

Originally Posted by rx7350

My intent is to build additional nodes to run Prime95 exclusively, and the above mentioned Pentium (two nodes) was built for only $332. Obviously, I'd like to get the biggest bang for the buck, and I read everything in these forums to appeared relevant to what I was trying to do. So, I'm a bit mystified as to why the iteration times on the Pentium are not MUCH better than what they are. I'm sure it's because I don't fully understand all of the inter-relationships between the architectural factors that ultimately determine iteration times. I had come to the conclusion that after the intial factoring attempts by Prime95 when first starting a new primality test, that cpu frequency was the most importatn factor by far, but that doesn't seem to be case in this comparison between the two architectures.

Even now, it looks like the best bang for buck is the Pentium D (Prime95-wise).

Does the system have integrated graphics? Sometimes that can cause a slowdown (To verify this, try reducing the screen resolution and/or color depth and see if that improves the timing)

Cruelty · 2006-05-06, 19:47

You may also check with task manager if Prime95 gets 49-50% CPU usage (assuming you run only one instance).

rx7350 · 2006-05-07, 00:07

Integrated (on-board) graphics, one of the things I was looking for in a mobo to minimize costs. When I lowered the graphics settings (resolution, color-depth) as far as I could the iteration times went from .090 to .087, a significant improvement. Makes me think an el cheapo graphics card might even reduce it further.

I also tried raising the priority from 1 to 8, but there was no effect on the iteration times, which kinda makes sense, since I don't run anything else on the system. The only two other things that might have any effect that I do run in the background is an antivirus/firewall app, and the ASUS PC Probe app to monitor system temps/voltages. I exited the ASUS app, but there was no effect on the iteration times. I have to run the antivirus (PC-cillin) app, but I'm thinking maybe I can turn the firewall off, since I run a firewall on my broadband router.

Other than that, I don't know what else can be done to reduce the iteration times. The benchmark you mentioned was probably for a single-core, 512Mb L2 cache cpu. This one is a dual-core, 1 Mb cache processor. I've stared and compared the benchmark tables, and it appears that the 512Kb caches have better iteration times than the 1Mb caches. Maybe something to do with the length of the path between tha caches and the memory controller(?).

rx7350 · 2006-05-07, 00:10

Also, system utilization is 100% (checked that long ago) - two instances, one each in each core. Don't think I have any hardware issues, since I ran extensive tests (Prime95, memtest86,etc) when I built the thing.

axn · 2006-05-07, 12:11

Quote:

Originally Posted by rx7350

I've stared and compared the benchmark tables, and it appears that the 512Kb caches have better iteration times than the 1Mb caches. Maybe something to do with the length of the path between tha caches and the memory controller(?).

512 kb L2 means it's a Northwood core. 1 Mb means prescott. For Prime95, Northwoods do have better timings than Prescotts for the same clockspeed (lower latency L2 and/or fewer pipeline stages, I think).

lycorn · 2006-05-07, 16:07

Are you running WIndows 98 on the Pentium box? I´ve noticed a signficant improvement (at least on a factoring task) turning the firewall (Zone Alarm) off on a W98 machine (but not on XP). And Northwoods are in fact much better than Prescotts for the same clock speeds. Another point is that as far as I remember your dual core CPU has a common cache for both cores. There may be some contention due to that, which would also explain slightly higher iteration times than for a single core CPU with the same architecture and clock speed.

E_tron · 2006-05-07, 17:02

Quote:

Originally Posted by rx7350

Integrated (on-board) graphics, one of the things I was looking for in a mobo to minimize costs. When I lowered the graphics settings (resolution, color-depth) as far as I could the iteration times went from .090 to .087, a significant improvement. Makes me think an el cheapo graphics card might even reduce it further.

instruct your OS to turn the graphics adaptor (monitor) off during inactivity. That seems to boost timings on my sempron notebook with intergrated graphics.

2006-05-05, 17:56	#1
rx7350 Feb 2006 AR, US 2⁴·3² Posts	CPU frequency and iteration times. I had always thought that a higher cpu frequency would directly result in lower iteration times, but I've observed some odd behavior on some of my nodes, if my assumpton is true. I have a Pentium dual-core running at 2.66GHz on a smaller exponent than an AMD dual-core running at 2.376 Ghz, but the interation times on the Pentium are about 2ms longer than those on the AMD. Both are 10M exponents using the 2048 FFT size. The only noticeable hardware difference is that the Pentium only has 512Mb, whereas the AMD has 2Gb. Your thoughts?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Iteration times in i5 and i7	Jud McCranie	Information & Answers	53	2013-08-17 19:09
What are your per-iteration times?	LiquidNitrogen	Hardware	22	2011-07-12 23:15
LLR.exe FFT crossovers and iteration times	SlashDude	15k Search	0	2004-01-28 05:47
slow iteration times	PLeopard	Hardware	9	2003-10-29 05:48
Slow iteration times with 23.7	smoffat	Software	13	2003-10-22 22:50

2006-05-05, 18:03	#2
rx7350 Feb 2006 AR, US 2⁴×3² Posts	Clarification - the exponenets are 10 million-digit exponents, 35M+ on both.

2006-05-06, 16:58	#4
rx7350 Feb 2006 AR, US 10010000₂ Posts	A64 X2 4400+ Toledo 8% ovrclk (2.376 Ghz) FSB 1 Ghz L1 cache 64kb L2 cache 1Mb 2Gb OCZ low latency ram exponent 35385871 .088 sec exponent 35385941 .088 sec Pentium D 805 Smithfield no oc (2.66 Ghz) FSB 533 Mhz L1 cache 16kb L2 cache 1Mb 512Mb OCZ low latency ram exponent 35037707 .090 sec exponent 35171009 .090 sec The AMD was built to be a household productivity pc as the main determining factor, and I just happen to run Prime95 as one of the many things that are done on this pc. The Pentium was built to run Prime95 exclusively, and since it's 'widely known(?)' that Pentium cpus are better for running Prime95, that's why I choose that architecture. My intent is to build additional nodes to run Prime95 exclusively, and the above mentioned Pentium (two nodes) was built for only $332. Obviously, I'd like to get the biggest bang for the buck, and I read everything in these forums to appeared relevant to what I was trying to do. So, I'm a bit mystified as to why the iteration times on the Pentium are not MUCH better than what they are. I'm sure it's because I don't fully understand all of the inter-relationships between the architectural factors that ultimately determine iteration times. I had come to the conclusion that after the intial factoring attempts by Prime95 when first starting a new primality test, that cpu frequency was the most importatn factor by far, but that doesn't seem to be case in this comparison between the two architectures.

2006-05-06, 19:47	#6
Cruelty May 2005 2³×7×29 Posts	You may also check with task manager if Prime95 gets 49-50% CPU usage (assuming you run only one instance).

2006-05-07, 00:07	#7
rx7350 Feb 2006 AR, US 2⁴×3² Posts	Integrated (on-board) graphics, one of the things I was looking for in a mobo to minimize costs. When I lowered the graphics settings (resolution, color-depth) as far as I could the iteration times went from .090 to .087, a significant improvement. Makes me think an el cheapo graphics card might even reduce it further. I also tried raising the priority from 1 to 8, but there was no effect on the iteration times, which kinda makes sense, since I don't run anything else on the system. The only two other things that might have any effect that I do run in the background is an antivirus/firewall app, and the ASUS PC Probe app to monitor system temps/voltages. I exited the ASUS app, but there was no effect on the iteration times. I have to run the antivirus (PC-cillin) app, but I'm thinking maybe I can turn the firewall off, since I run a firewall on my broadband router. Other than that, I don't know what else can be done to reduce the iteration times. The benchmark you mentioned was probably for a single-core, 512Mb L2 cache cpu. This one is a dual-core, 1 Mb cache processor. I've stared and compared the benchmark tables, and it appears that the 512Kb caches have better iteration times than the 1Mb caches. Maybe something to do with the length of the path between tha caches and the memory controller(?).

2006-05-07, 00:10	#8
rx7350 Feb 2006 AR, US 2⁴×3² Posts	Also, system utilization is 100% (checked that long ago) - two instances, one each in each core. Don't think I have any hardware issues, since I ran extensive tests (Prime95, memtest86,etc) when I built the thing.

2006-05-07, 16:07	#10
lycorn Sep 2002 Oeiras, Portugal 2⁶×23 Posts	Are you running WIndows 98 on the Pentium box? I´ve noticed a signficant improvement (at least on a factoring task) turning the firewall (Zone Alarm) off on a W98 machine (but not on XP). And Northwoods are in fact much better than Prescotts for the same clock speeds. Another point is that as far as I remember your dual core CPU has a common cache for both cores. There may be some contention due to that, which would also explain slightly higher iteration times than for a single core CPU with the same architecture and clock speed.