CPU benchmark

E

eeyrjmr

[H]ardness Supreme
Joined
Apr 23, 2002
Messages
4,363
the standard work PC as provided by IT is a:

P4 2.0GHz, 256Meg RAM, 10Gig HD and running windows 2000

This is crap for what I need and was able to get a 1/2 decent lappy last year (Pentium-M,1Gig RAM,Windows 2000)
The RAM made the biggest difference and VHDL synthasis for a 8million gate FPGA can get done in 15min (for a 70% full device), likewise Simulink sims are reasonable


now more and more are doing sim work at work and they only have their desktop and it just grinds!!!
we finally after 2weeks of argueing with IT saying that they are costing our customers in excess of £5000 a month (ie since sims on std desktop takes soo long our time has to be charge for that extra time).


So I got given a "Uber" desktop to try out


Laptop
CPU 1.8GHz Pentium M
RAM 1Gig
OS 2000 Professional

Uber Desktop
CPU 2.8Ghz HyperThread enabled/disabled
RAM 1Gig (lower latency then Laptop)
OS XP Professional

Standard Desktop
CPU2.0GHz P4
RAM256Meg
OS2000 Professional







Simulation times:

Laptop: 980.34sec (16min 20sec)

Uber Desktop: 1561sec (26min 1sec)

Uber Desktop (no HT): 1601.5 (26min 41sec)

Standard: 4911.9 (1h 21min 51sec)



RAM used

Laptop: 430Meg

Uber Desktop: 560Meg

Std : 450Meg (ie into SWAP == slow)
Click to expand...


I was like WTF!!! the speed difference can only be contributed to XP (all OS's were running exactly the same in the background, ie A/V,bar XP specifics) or the CPU
The 2.8Ghz should of shat on the 1.8GHz but XP seems to have turned the 2.8GHz CPU into an equive 1.6GHz!!!


There is a 4Gig partition on the "Uber" machine I am tempted to install Xubuntu and get Matlab to send me the linux binaries so I can really benchmark the hardware


But FFS is XP really crippling hte machine!!!


The only other explanation is.. is the P4 chip really that bad!!! this is pretty bad performance.
I need to spec the right machine for 11off of these machines and they have to come from DELL (lease and all that). IF it is the CPU and since DELL does AMD now I could push for a CPU change
 
How is XP the issue here? I'm not getting your reasoning for such an opinion.

It looks to me like those numbers seem about right. That Pentium M is definitely a better performer than the P4 Hyperthreaded CPU, hands-down. Pentium M has a way more efficient pipeline for both multimedia and straight crunching compared to the P4. In fact, the Core Duo architecture is built off the Pentium M architecture (hence the lower clock speeds but better performance).

You'll have to wait until next month to get an AMD Dell, and if you want the best possible you might want to go with the Core Duo (for now, at least).
 
GreNME said:
How is XP the issue here? I'm not getting your reasoning for such an opinion.

It looks to me like those numbers seem about right. That Pentium M is definitely a better performer than the P4 Hyperthreaded CPU, hands-down. Pentium M has a way more efficient pipeline for both multimedia and straight crunching compared to the P4. In fact, the Core Duo architecture is built off the Pentium M architecture (hence the lower clock speeds but better performance).

You'll have to wait until next month to get an AMD Dell, and if you want the best possible you might want to go with the Core Duo (for now, at least).
Click to expand...


Well on the surface XP was the main difference and the quick-stab was XP BUT if you note I said it is either XP or the CPU (I don't know which hence the question asked)

I never realised the P4 was so bad.
Are Dell selling Core2 now?

IT control what PC's are brought onto site and they have a very restricted "lease" deal with Dell so the machines we get for our dept have to be from dell books

if it is P4 or nothing then we hold out, if it is P4,Core2 or AMD later then I will push for Core2 now
 
Well, it isn't that the P4 is so bad, it is that what you look like you're doing is heavy interger crunching. P4 is more suited to multimedia (SSE3, 3DNow, etc.). All I know for certain is that until the Core Duo, AMD had the clear advantage in pure number-crunching processes (rendering, et al). The only real contender in the Intel camp was the Pentium M, which was amazing in its performace at nearly half the clock speed of a P4.

I say go for the Core Duo now, and make sure you get a Quatro vid card in it (will help with any graphical computation... day and night difference in renderers and 2D apps).
 
Not a problem. I'm actually interested in the Duo's performance with software that will actually make use of its two cores and better architecture (I don't care about games), so let me know how things turn out if you do finally get a Core Duo machine. :)
 
The OP probably want a (large cache) Dual-Core Opteron or Core 2 Duo. Any why so little memory? My gaming machine has 2GiB for christ sake, and it's not bringing in any money from clients :-o

(even if the simulation/routing isn't MT, if you're running it on your desktop it's nice to have some CPU left over to continue design or whatever.)
 
@GreNME
Concider it done. I am looking at the Core CPU for home use as well (Dec rebuild) and this will give me a chance to play with it. Will run exactly the same simulation model to ensure like for like

@eloj]The OP probably want a (large cache) Dual-Core Opteron or Core 2 Duo. Any why so little memory? My gaming machine has 2GiB for christ sake, and it's not bringing in any money from clients :-o

(even if the simulation/routing isn't MT, if you're running it on your desktop it's nice to have some CPU left over to continue design or whatever.)[/QUOTE]

The reason why "so little RAM" is because at my site there is an IT budget and with all the managers and MD's wanting super-slim laptops with fast CPU's and lots-o-RAM just to send email (status symbol) it severly reduces the pool of money available, thus when it comes to use grunts who actually do the work we don't have much ££ to play with, plus our IT likes to waste money (they want to install a citrix server).

I have actually specced 2gig of RAM for these machines.
There were 2 test-model conditions (Simulink model)
both based around a model I made a year ago (SR machine generator model)

1st purely for CPU benchmarking, only 2 "to workspace" ports (RAM used was only 500meg) and model ran for 10sec simulation time (15min real time)
2nd was ~20 scopes dotted around (20 isn't unreasonable) within 1.2sec (simulation time) RAM was at 800meg and still riseing, thus we need more RAM
 
You must log in or register to reply here.
Back
Top