Stock Yonah benches

[mike]

I was able to run a few benches with a stock clocked 2.16GHZ yonah on an Intel ref board - here's are the results with very loose memory :

1m: 0:30:71 secs
[img=http://img208.imageshack.us/img208/1052/yonah1m5yw.th.jpg]

2m : 1.14.56
[img=http://img71.imageshack.us/img71/7411/yoanh2m7nc.th.jpg]

8m: 6:08:56
[img=http://img71.imageshack.us/img71/2285/yoanh8m7vq.th.jpg]

32m 29:28:141 secs:
[img=http://img208.imageshack.us/img208/3532/yobah32m2mi.th.jpg]


Dual 32m is taking a hit due to sharing the L2 cache:
[img=http://img208.imageshack.us/img208/8566/dualyonah32m9so.th.jpg]

3DMark05 with single 7900GTX all stock clocks and default High IQ driver settings:

[img=http://img71.imageshack.us/img71/9350/yonah3dmark055sv.th.jpg]

3DMark06:
[img=http://img71.imageshack.us/img71/4615/yonah3dmark068bo.th.jpg]

PCMark04:
[img=http://img71.imageshack.us/img71/7595/pcmark04yoanh1gm.th.jpg]

4 and 16m can be found on the results on the Superpi records - the 1m and 512kb record runs were done with another cpu
]
The board does not allow any overclocking unfortunately - but on the positive side it also supports cpus with 4MB cache

[M.Beier]

Great mike

Can you please give 01SE a try, I'd truely like to see how it performs with your x850XT

[Gautam]

Nice stuff my friend. What I like the most is how strong it is in 32M. It looks like it does very well in long Pi calculations, unlike Dothan which sorta begins to lose its edge there.

Nice preview of things to come.

[mike]

Great mike

Can you please give 01SE a try, I'd truely like to see how it performs with your x850XT

If irrc 3DMark01 on the GTX was 26000 points all stock - runs without ANY optimization on WinXP

[Gautam]

That ain't what the ORB says.

http://service.futuremark.com/compare?2k1=8931740

[mike]

That ain't what the ORB says.

http://service.futuremark.com/compare?2k1=8931740

oops - mixed it up

well this run was when I still had problems with the cpu throttling down - so it would be more like 25 or 26k

[Charles Wirth]

Putting some fat body D9 up against ya, first up is my stock scores.

Still setting up and downloading stuff

[ibby]

Whats a Fat body D9 ?

[Gautam]

Think BH5 for DDR2 . High volts, tight latencies.

[Charles Wirth]

Specific DDR2 memory.

28s pi 1M at stock for me so far, If I get setmem running on this thing it might hit 27s

[althes]

This is getting good

[VictorWang]

@.@ hi mike,
have you optimized ur sys ?
why your benchmarks seem a little bit slow ?

see this: I run on my Laptop

[MicroNecessian]

What Mb are you using Mike?

[mike]

[QUOTE=VictorWang]@.@ hi mike,
have you optimized ur sys ?
why your benchmarks seem a little bit slow ?

see this: I run on my Laptop




My memory is running on ridiculously loose timings at 266mhz - I think that's where the culprit lies.

Going to try to do a new install, but not sure if that will help.

The motherboard that I am using is a Intel Mobile Reference board - which prolly is the second culprit for the slow benches

[thunderstruck!]

You all are gonna have to start running 2 Superpi instances when using Yonah. Otherwise, it's only stressing 50%. I assume you want a 100% pass right?

[mike]

You all are gonna have to start running 2 Superpi instances when using Yonah. Otherwise, it's only stressing 50%. I assume you want a 100% pass right?

well, this was only a quick set of benches - and I agree with you on dual - while I had not have the time to run all benches in dual yet - in my first post you can see that I had run dual instances of 32m - which should have give and idea on what is going on

[Waus-mod]

that 2 times run is no need.... you also arent going to start 3dmark twice..
1 time is also not "a stress test" because if that core sux it wont run either!

so i really dont see why you should run it twice,

[mike]

Just switched chipsets in the Bios from 945GM to 955XM and got the SuperPI 1M times down by 1 sec - running a full set of benches again - hopefully with more promising results

3Dmark05 went also up by 2-300 points

[mike]

8m is alos going down nicely fomr 6:08 to 5:51

[wwwww]

Let's see 01 benchies.

[VictorWang]

My memory is running on ridiculously loose timings at 266mhz - I think that's where the culprit lies.

Going to try to do a new install, but not sure if that will help.

The motherboard that I am using is a Intel Mobile Reference board - which prolly is the second culprit for the slow benches

@.@ RAM=533, that's it , my ram=667

[thunderstruck!]

that 2 times run is no need.... you also arent going to start 3dmark twice..
1 time is also not "a stress test" because if that core sux it wont run either!

so i really dont see why you should run it twice,

You should run it twice because then you are using both physical cores. 3d is a performance test, not really a stress test. There's obviously no need to run more than one 3d

[mike]

You should run it twice because then you are using both physical cores. 3d is a performance test, not really a stress test. There's obviously no need to run more than one 3d

When you first posted about running 2 instances - I had assumed you were more intersted what will happen with the performance of the PIs - since the L2 cache is shared between the 2 cores - there is obviously going to be a performance hit. While on a single instance the one core can take advantage of up to 2MB of L2 cache - with both cores runinng full steam - each can only utulize up to 1MB.

This hit can be seen clearly in the times of 2 32ms running simulationsly compared to only running one instance.


For stabilty testing running 2 instances can be helpful - but we have to assume that stabilty is not an issue at stock clocks with loose timings.

Running two instances will be much more interesting once we get the clocks up and running

@.@ RAM=533, that's it , my ram=667

still - your memory is sligthly looser than mine - did you use any tweaks? I ran bone stock - NO optimzation whatsoever. Both you and Fugger are about 1 sec faster. But with Fugger I know that he hardly ever runs anything untweaked :-)

[mike]

On a sidenote - I had to use the microsoft hotfix for dual core throttling in order to get rid of annoying jumping back and forth from lower to max multipliers during runs. Drove me nuts and took forever to figure out.

The throttling only occured with a fresh Windows install - on an old XP install that I had used on a Presler/ 975X system and just threw onto the Yonah rig did not show this problem espcially considering that I got the fastest benches on that install with 975x chipset drivers....lol

[metro.cl]

cant you lower ram timings??