WinSMP 3064 on Yorkfield test

Cob · 10 May 2008 at 09:29

I'm near the end of testing this Q9450, so I thought I'd post a little benchmark of one of the much loved (ahem) 1763 pointer 3064 WinSMP units.

'WinSMP1' is the single 3064 running on this machine -

Code:

Project : 3064
 Core    : SMP Gromacs
 Frames  : 100
 Credit  : 1753

-- WinSMP1 --

 Min. Time / Frame : 7mn 14s  - 3489.84 ppd
 Avg. Time / Frame : 7mn 19s  - 3450.10 ppd
 Cur. Time / Frame : 7mn 29s  - 3373.26 ppd

I should have some VMWare'd linuxSMP 2605 unit comparisons to add later this w-end

*note - the abnormality at 0700-0714 is due to my daily antivirus scan.

booyaka · 10 May 2008 at 12:05

stunning speed!! -

Min. Time / Frame : 14mn 59s - 1684.75 ppd
Avg. Time / Frame : 15mn 50s - 1594.31 ppd

to comapre on a Q6600 @ 3.15ghz!

theheyes · 10 May 2008 at 13:19

Is this where I kick myself for not waiting?

Cob · 10 May 2008 at 13:38

booyaka said:
stunning speed!! -

Min. Time / Frame : 14mn 59s - 1684.75 ppd
Avg. Time / Frame : 15mn 50s - 1594.31 ppd

to comapre on a Q6600 @ 3.15ghz!

Are you running two WinSMP clients at once?

booyaka · 10 May 2008 at 16:34

Cob said:
Are you running two WinSMP clients at once?

yeah - had loads of issues with 3064's & 3065's so they might not be the most accurate benchies!

Cob · 10 May 2008 at 16:41

I didn't try running two together.

I thought about it, but I wanted the unit to finish quickly so that I could start my VMWare clients before I left the house this morning.

booyaka · 10 May 2008 at 17:23

tempted to get a Q9450 now i've seen that!!!

Cob · 10 May 2008 at 17:34

I think it's only about 10% faster than a Q6600 clock-for-clock.

The main difference will be when the clients are optimised for SSE4.1. And god know how long that will take.

theheyes · 10 May 2008 at 19:12

Probably long enough for Nehalem to arrive so it's a catch 22, as always.

booyaka · 10 May 2008 at 21:17

damn - sods law talking about 3064's and i've just picked up 2 of them!

Pilgrim57 · 10 May 2008 at 23:13

V interesting, Cob. What mobo, HSF & RAM are you using?

Cob · 10 May 2008 at 23:21

Abit IP35-Pro, water, and 4x2gb OCZ PC2-8000 Platinum

Cob · 11 May 2008 at 10:15

Four LinuxSMP clients via VMWare running 2605 units in x64 Vista comparison -

Q6600 @ 3.9ghz -

Q9450 @ 3.78ghz -

Note the latter screenshot was taken in x64 XP, but the VM's are running on my Vista machine.

Pilgrim57 · 11 May 2008 at 13:08

Cob said:
Abit IP35-Pro, water, and 4x2gb OCZ PC2-8000 Platinum

I suspected you were running under water

at least it wasn't LN2 :eek:

I wonder what would be realistic with air?
Nice shots: so for 120MHz less you get about 500ppd more about 2 free 1760 pointers a week

& less leccy used as well I presume.

Cob · 11 May 2008 at 13:54

It's only using about 0.04v less than the Q6600 was at 3.9ghz (tho it's a very very good Q6600). And then there's the fact that it cost well over twice as much as the Q6600 did.

It's probably not a worthwhile upgrade atm, atleast not until linuxSMP incorporates SSE4.1.

Pilgrim57 · 12 May 2008 at 21:35

Cob said:
It's probably not a worthwhile upgrade atm, atleast not until linuxSMP incorporates SSE4.1.

& when is that likely to happen?

Stelly · 12 May 2008 at 21:51

Cob can you tell me what settings your using to get that speed on your Yorkie

Stelly

Cob · 12 May 2008 at 22:11

Nope

Vcore - 1.37v under load
MCH - 1.52v
VTT - 1.32v
GTLRef - 71%

And a little bit of the magic touch

Cob · 12 May 2008 at 22:11

Pilgrim57 said:
& when is that likely to happen?

God knows.