The bandwidth tests I tend to use are of the:

dd if=/dev/zero of=/dev/null
followed by CTRL+C after a second or 10
or

dd if=/dev/random of=/dev/null
followed by CTRL+C after a second or 10

variety.
But I get more or less the same results with that as with memtest and other apps.

I need to get me some of that ram you got for the Sempron system :p the mobo can handle up to 6.4GB/s bandwidth at 200Mhz FSB and I know that the only thing keeping that system back from doing 433 or even 466 is the ram.

:) you call that a rant Jan? very good post, it was funny to read.

However, I wonder what's the reliability (when it comes specifically to SETI performance) of the memory benchmark which is built in to the KWSN apps.

This is what I get on an Opteron 2.7 with DDR420 2-2-2-5 1T, GenSSE3:

speed: 2666 MHz -- read megs/sec: L1=14617, L2=7248, RAM=3709

Memtest86+ reported similar numbers IIRC.

---

Dual-channel AMD systems using DDR2 can achieve over 2GBytes/s memory bandwidth.

Only 2 GB/s ? :D I should be glad I'm still running DDR ;D seriously, what's the preferred benchmark to rate the effect of memory bandwidth on SETI performance?

OK, OK, OK... So I'm behind the times a little...

The benchmarks to look for are such as the SiSoftware Sandra 2007 "Memory floating point" and "Memory integer" (links to Toms Hardware CPU charts).

Looks like AMD has the lead with 9.2GBytes/s there.

Happy crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Lol, my Socket A Sempron is actualy pushing close to 3G/s and thats just plain old DDR1 @ 400Mhz with crappy 333-8 timings :p

Current tech hardware like Ahtlon64/FX's and C2D's are pushing between 4 and 8GB/s and I've seen some crazy OC's pushing 10G's with quality ram.

The Intel BX chipset was doing 300MB/s at 100Mhz fsb's for P2's and P3's while its competitors were still trying to get past 150MB/s.

The Ali chipset I'm talking about is a 133FSB revision which still can only pull 150MB/s and I know that the BX was pulling close to 500MB/s at 133FSB and even more at 150FSB.

The Intel BX was the best chipset Intel ever made and to date Intel has yet to deliver a chipset it can call king of performance again. A titel the BX took and held for a long long time. After that Via took the crown of performance and tweakability followed by nVidia's nForce series chipsets, which are still the most feature rich and best overclockable chipsets around.

The Abit BE6 and BE6-2 are still boards of folklore and legend in the overclocking scene, all thanks to the BX chipset.

Dual-channel AMD systems using DDR2 can achieve over 2GBytes/s memory bandwidth.

Only 2 GB/s ? :D I should be glad I'm still running DDR ;D seriously, what's the preferred benchmark to rate the effect of memory bandwidth on SETI performance?

---

Man, I'm a bit bummed here, I got 2 P3 systems I had sitting in the attick, 1 is a P3 933 the other a P3 650. Together they are good for about 3 extra units per day

One has a memory bandwidth of +/- 150MB/s (shitty Ali Chipset on P3 933)

The other has a memory bandwidth of +/-300MB/s (Good ol Intel BX chipset, rocket back then, still rocks now)...

Anyways RANT/Off

Technology and memory have marched on a long long way since those days...

Dual-channel AMD systems using DDR2 can achieve over 2GBytes/s memory bandwidth. However, that sort of memory just wasn't available or needed for your era of hardware.

Then again, I will admit that even today's Intel systems are still memory bandwidth limited by the Intel FSB. Then again again, Intel have put a LOT of expensive effort into fancy fast caching to keep things moving.


Its all a compromise and a question of what best balance can be achieved...

Happy crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Man, I'm a bit bummed here, I got 2 P3 systems I had sitting in the attick, 1 is a P3 933 the other a P3 650. Together they are good for about 3 extra units per day

One has a memory bandwidth of +/- 150MB/s (shitty Ali Chipset on P3 933)

The other has a memory bandwidth of +/-300MB/s (Good ol Intel BX chipset, rocket back then, still rocks now)

I wish I could put the P3 650 on the Ali motherboard and the 933 on the BX, so that the lower memory bandwidth on the Ali chipset wouldn't be to much of a problem since the other CPU is slower too.

Problem is the P3 650 is a Slot1 and the P3 933 a Socket 370.

The 650 is spitting out units fastest of the 2, simply because it has more memory bandwidth, a 650 beating the living daylights out of a CPU more then 50% faster!

/wonder how I'm gona get past this one again. Maybe I can get the Ali system to utilise the graphics cards memory. I've made use of graphics cards extra memory before, but usualy as additional swapspace, don't know if you can actualy map it as system ram somehow.

Best thing is still to come though, I know the 650 can do 866Mhz on that board but didn't OC it to that level yet untill I was sure everything worked.
So it'll probably go from beating the crap out of the P3 933 to beating it to death.

Anyways RANT/Off