CPUmark99 and modern processors

Toggle sidebar Toggle sidebar
Hulk

Hulk

Diamond Member
Oct 9, 1999
5,096
3,596
136
I'm curious as to how modern processors do with CPUmark99. Yes I know it's a very old synthetic benchmark. It's still a decent indicator of integer performance. If you wouldn't mind please take a minute to benchmark your system and report the score.

You can download the program here:
CPUMark99

It's only about 500kb and WILL NOT install anything on your computer. It runs from the exe file. Just add the ".exe" extension to the file and double-click it to run.

Here are some old scores:
Celeron 450A 36.1 for 12.5Mhz/CPUmark
Athlon 550 51.9 for 10.6Mhz/CPUmark
PIII850 76.1 for 11.2MHz/CPUmark
P4 3.06 194 for 15.8MHz/CPUmark

Please post your cpu model and speed. Memory and mobo doesn't really make a difference for this test. It isolates the processor pretty well.

******UPDATE******
This is cool. I thought of a way to test my HT theory.

Open two instances of CPUmark99.

Put them side by side and start both of them. If you're fast you can get them both running in about 1 second or so.

Now this is cool. Normally with a HT or dual core system only one core will be used by this test and total cpu load will be around 50%.

By starting two tests at once you can load both physical and logical cpu, pegging cpu usage at 100%!

With my Northwood 3.06 I'm getting scores of 115 and 117 for a total of 232. I think that is representative of the total work my Northwood core can do in integer performance.

I think some of the dualies here will scream running two at once. I'm encoding some video now so I can't reboot but I would think with HT off my total score for two tests would be the same as for one.

I would also think that 3GHz Prescott would score like my Northwood with two instances running.

Okay guys, check it out and let's see what we can learn.
 
T

TrevorRC

Senior member
Jan 8, 2006
989
0
0
Opteron 165, 2.52 Ghz

310, and only used 1 core.
Assuming if it had used both cores it would have been around ~600ish.
 
M

morgash

Golden Member
Nov 24, 2005
1,234
0
0
lol i feel ashamed to post this buuuut
p4 520 @ p4 520 (2.8ghz 800FSB 1mbL2 cache)

161

lol it burns the flesh

*EDIT*
my 2.8 pressy scored 109 on each when i ran two and 161 when i ran one
so 218 score with 2 runnin, there are definetly LOTSA unused cycles in that big ass pipe
*EDIT2* i OC'ed my 2.8 pressy to 3.55 and got a score of 210 so there ya go :)
*EDIT3* i just tested the FX-60, at stock speeds(2.6ghz) it pulls a 337 one core
AND a 337 and 334 on both cores!! so a 671 TOTAL!! WOOT!
 
Hulk

Hulk

Diamond Member
Oct 9, 1999
5,096
3,596
136
It's interesting to look at the MHz required for 1 CPUmark or MHz/Cpumark. It's a bit of a look into the integer efficiency of the core. This is only a single threaded test so yes, only 1 core will be working.

That P4 520 is a 31 stage Prescott, right?

Let's see, looks like this so far in terms of MHz/CPUmark. Of couse lower is better.

X2 - 8.1 (one core only)
PIII - 13.3
P4 Northwood - 15.8
P4 Prescott - 17.4

We already have seen this in various benches, but in pure integer performance the A64 has a 1:2 clockspeed advantage over the P4 according to this benchmark. Of course all testing doesn't show this since SSE instructions as well as floating point instructions come into play to help the P4 which does well with those. But we have seen a pretty big advantage in business app testing, which is largely integer based.

I am curious how Dothan compares to Yonah and of course A64. I think that will give us some idea of where Intel will land with Conroe.

I know that Yonah most likely has a 12 stage pipe, 3 issue core and that Conroe will be 14 stage 4 issue. Still, in order to beat X2 Intel will have to get integer performance beyond what I think Yonah will show on this test. I wonder if the wider 4 issue core will make up for the longer pipe?

You see I'm thinking that Intel is going to be releasing Conroe at 2.66GHz at the top and I honestly don't see that being the fastest processor available upon launch. It seems reasonable that by then AMD can release X2 at 2.8GHz and Conroe would have to have one impressive core to beat X2 with a 200+ MHz clockspeed disadvantage. In addition Intel has to deal with AMD's inherent advantage of the on die memory controller.

Of course we won't know until Conroe is tested, but we usually don't see miracles in this business and we all studied enough processor releases to predict what will happen with some accuracy.

I have a feeling Conroe will equal or have a small, maybe 5% efficiency advantage over X2. And some of that will probably be due to SSE and other Intel specific optimizations. Best case. So I guess when Intel says Conroe will be faster than X2 while using half the power at launch they expect AMD not to release a faster chip by then.

I think we're going to see another battle like the PIII and original Athlon where the chips were pretty evenly matched but the one with longer "legs" ie ability to scale will win. Of course then the Athlon was the new design with the legs. This time the tables may be turned as the Athlon is the older design.

But then AMD is in a great position now. It is currently on top and has time to refine their current process.

The next 6 to 8 months in particular should be very interesting.



Let's see if we can get an average efficiency number for each core out there.

Anyone else want post a score. It really only takes a minute to download AND run the test.
Just try and make sure you don't have too many cpu intensive background process running while running the test.
 
M

Matt2

Diamond Member
Jul 28, 2001
4,762
0
0
Opty 170 @ 2.7GHZ - 346 CPU marks - 7.8 mhz/CPU mark

Looks like I have the performance crown! I pwn j00!!
 
Hulk

Hulk

Diamond Member
Oct 9, 1999
5,096
3,596
136
Update on average MHz/CPUmark99

X2 - 8.1 (one core only)
Pentium M (Dothan) - 9.23
PIII - 13.3
P4 Northwood - 15.8
P4 Prescott - 17.4

I doubt anybody has a Yonah based system around here.
 
K

KBTuning

Senior member
Mar 22, 2005
357
0
0
Winchester 3000+ 1.8Ghz@ 2250MHz and it pulled a 266.... so thats 8.4586466165413533834586466165414MHz per point :)
 
Hulk

Hulk

Diamond Member
Oct 9, 1999
5,096
3,596
136
That 31 stage pipe definitely hurts Prescott for integer performance.

Prescott at 3GHz does 174 while Northwood at 3.06 does 194 on 533fsb.

Even with a better branch predictor Prescott is about 9% less efficient than Northwood in integer.

There is an interesting trend here. Pipelength seems to be very important for integer performance.

Athlon and Pentium M with 12 stage pipes perform very closely, while Pentium III with 10 stage pipe does okay for an older chip. Makes sense when you realize PIII only has 2 issue core (I think) vs. 3 for Athlon and PM.

Branch stall are a killer for integer performance. Now I can also see why hyperthreading works on P4, all that unused cycles and be used for a second thread. According to these results Hyperthreading should be even more effective with Prescott over Northwood. That is I would think that on a multithreaded app I would think that Northwood and Prescott would perform about the same at the same clockspeed.

Northwood might have less work on 2nd thread (less pipe stalls) but Prescott would have more stall but 2nd thread therefore picking up more of the load.
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom