MSP7 and Hyperthreading Benchmarks

[Hulk]

I've posted three 3.06 P4 scores on my benchmark site:

http://www.hyperactivemusic.com/mspro.htm

The scores are all over the road, I've tried to analyze them but I'm afraid we won't get to the bottom of this without more scores. It also seems like there is great potential in the combination of HT and MSP7.

If you have MSP7 and haven't yet ran the test on your system I would appreciate it if you would take the time run the test so we can expand the results table.

- Mark

[Sciascia]

Do you just want scores from Hyperthreading? I have a 2.53B in the box right now, but I have a 2.6C sitting on the desk as we speak ready to be installed in the same box. I could probably get a pretty good comparison of what an older and newer Northwood can do. Only thing that is going to throw it off is the new one is 800MHz FSB.

[Hulk]

That would be great. If you could run the test on the 2.53 and the 2.6, with and without HT on, then we would have some great data.

We could see the difference in effeciency on the same motherboard when going from 333MHz memory to 400MHz memory by comparing the 2.53 with the 2.6 (HT off).

Your data with the 2.6 could also help us get to the bottom of the scores I've already received. Test the 2.6 with HT on and off.

Thanks,

Mark

[stevenscott]

Hye Hulk, is that board and CPU a 400, or 500 FSB, or 800 FSB ? would really like to see that info too. Just a thought...

ooops, I see, you already stated it, it's 533 FSB.

[fberger]

When you test the 2.6c vs. 2.53b on the IC7, also keep in mind that you'll not only get the higher FSB, but PAT will also kick in since you're using the IC7 (i875). PAT will improve perf by about 3%. You may want to deduct this from the improvement you see over the 2.53.

When testing HT vs. non HT, I wonder if one does not have to manually update MSP7's .ini file and make it non dual processor. I believe the 7.01 patch detects dual processors when first launched, but who knows if, when you disable HT, MSP tests again for the presence of the second processor. This may explain the poor performance (a 3.06 non-HT is only 7 seconds faster than a 2.53 over a 25 minute job).

The perfect scenario would be to have a dual-boot HT system:
1. Enable HT, install OS and MSP run test.
2. Disable HT in Bios, install OS on second partition, install MSP and test.

This would be a good clean test, instead of just activating and deactivating HT for the same software environment.

Fred

[Hulk]

Fred,

I think the deduction for PAT can just be a mental note when analyzing scores. After all, PAT is part of the 875 chipset.

Your second point is interesting. Perhaps MSP is thinking that a HT system is dual processor even after turning off HT. I suspect that MSP checks this when the program is launched, and since HT is turned off from the Bios, requiring a restart, it should get this right.

Like I said, we need more data.

- Mark

[Sciascia]

Hulk, you should have my results from the test in your email. Everything was faster with the new chip, which is kind of the point of going with a newer chip

I'll let Hulk post the results on his site, since I don't want to interpret them wrong (pretty hard to do, but still)

I also ran a few other benches:

2.53B at 533MHz FSB
Sysoft Sandra:
-Memory bandwidth
3350-3356
-CPU Multimedia
10026-12209
-CPU Arithmetic
4846-1320/3100

3D Mark 2001
14419
3D Mark2003
4970

2.6C at 800MHz FSB w/ HT'ing
Sysoft Sandra:
-Memory bandwidth
4796-4808
-CPU Multimedia
12214-19411
-CPU Arithmetic
6619-2110//4951

3D Mark 2001
15,719
3D Mark2003
5060

Pretty good boost across the board, not sure which one helped more, the HT'ing, the PAT or the 800MHz FSB. I guess they all helped ultimately. The best part is that it is almost completely silent. The Samsung HD's are a lot slower than my old RAID0 array with a pair of WD's, but it is fast enough and are dead silent. The CPU HSF is incredible, great temps and in silent mode it is silent. Zalman makes really good stuff, I also bought the fanless chipset cooler. But the best part was the Fortron PSU, it is the quietest PSU I have ever seen and was only $31. I ditched my Enermax Whisper series that cost almost 3X as much for this thing. Voltages are solid too.

[Hulk]

Benchmark website has been updated to reflect the new scores.

- Mark

[Sciascia]

Not really related, but I started playing and decided to see what this puppy could do. I OC'ed her to 3G (using the 4/5 memory divider) and she didn't even flinch. Very nice CPU, I won't run it like this all the time, just nice to know that I have alittle "wiggle room" and she isn't running that hard.

Here are the benchmarks:

CPU Mutimedia
15025-23723

CPU Arithmetic
7340-2583/6514

3D Mark 2003
5245 <--link

3DMark2001
17166 <--link

[fberger]

I was planning on getting a similarly configured setup, but am still trying to figure out the memory I should get that would work with the Abit IC7 or IS7. So many people are having problems, it's unbelievable. Anyways, here's my question: you are using Corsair PC3200 memory, could you please tell me exactly which one? Is it the XMS, or the LL? If you had some part number, that would be very helpful. What memory settings are you using? SPD? Manual?
I was planning on getting a P4 @ 2.6 and overcloking it to 3.2 with a 5:4 FSB to RAM ratio.

It looks like you got a very nice rig, and stable too. Congrats!

Fred

[fberger]

Mark,

On your web site you list Sciascia's system FSB as 533 Mhz, is it a typo or isn't he using the full potential of the IC7? With a 2.6c it should be 800 Mhz FSB, should it not?

Fred

[Hulk]

Fred,

You are right, that is a typo, it should be 800fsb.

Thanks,

Mark

[Brian Ellis]

Mark

It is possible that my poor results are due to my using W2k. Intel claim that XP is required to obtain maximum benefit from HT and they do not recomment W2k. However, I'm leaving home in a few hours. I'll try to revisit this in 2 weeks or so, after I return.

May I respectfully suggest that it would be useful to add RAM size and OS to your tables?

[Sciascia]

Originally posted by fberger
I was planning on getting a similarly configured setup, but am still trying to figure out the memory I should get that would work with the Abit IC7 or IS7. So many people are having problems, it's unbelievable. Anyways, here's my question: you are using Corsair PC3200 memory, could you please tell me exactly which one? Is it the XMS, or the LL? If you had some part number, that would be very helpful. What memory settings are you using? SPD? Manual?
I was planning on getting a P4 @ 2.6 and overcloking it to 3.2 with a 5:4 FSB to RAM ratio.

It looks like you got a very nice rig, and stable too. Congrats!

Fred

I am using the PC3200C2 XMS series....here is the part number CMX256A-3200C2. The LL series have a lot of issues and are geared mainly towards AMD CPU's. I am running stable at 2-3-2-5. If you use a "C" series CPU then you can enable some BIOS settings for the perforance mode and the am timings are not adjustable, but even though they are slower timings, there is a boost in performance. I believe it has to do with PAT and maybe some other timings that aren't adjustable manually.

Those are the exact OC'ing settings I used, a 5/4 divider and about a 240MHz FSB. 3.2G and rock solid. I am not leaving it like that as at 2.6G it is still insanely fast, but it will do it. I wouldn't use stock cooling, I have a big old Zalman CNSP-7000Cu that is near silent at full speed and silent at low speed. I also swapped the NB cooler out to a fanless Zalman and it works great.

[Sciascia]

OK, I just did the non HT'ing and the OC'ed tests. Pretty interesting. The OC'ed (3.2G) with HT'ing is obviously very fast, though a dual XEON system still beats it, and it should.

She's back down to stock speeds, it is just way too hot over here with no AC to be running something that hard. I know if I am sweating, the CPU is near melting!