I finally got my new machine up and running, thanks to all
those that helped me make my choice. I ended up with the T3400,
E8500, FX1700 2-gigs RAM (2) 10K raptors.
Wow! I can't believe the difference from my P4 machine. I was very disappointed with SW08 until I got this machine what a difference. I just ran the punch holder benchmark file again. 78.93 seconds right out of the box. I was around 7 mins. on the P4.
Craig
Wow! I can't believe the difference from my P4 machine. I was very disappointed with SW08 until I got this machine what a difference. I just ran the punch holder benchmark file again. 78.93 seconds right out of the box. I was around 7 mins. on the P4.
Craig
Here we go again Peter. There must be an issue with Solidworks and dual vs. quad cores. On my one tested Dell T3400 QX9650 3.0 GHz quad, the best time I got was 91.57. This is a 45nm chip and a Core 2 Extreme. The only difference I had was SAS 15K drives. What gives?? I know it is a SLIGHTLY slower clock speed, but you have 2 more cores to use. Is solidworks just not quad core friendly?
I was happy with my purchase, but now I am second guessing.
Any thoughts?
The system was around $2600 that included a 22" ultra sharp monitor, (2) 160 gig 10K raptors, wireless card 2 gigs 800 mhz RAM and quadro FX 1700. I thought it was a pretty good deal. I am real happy with the performace so far.
Craig
Don't know why the numbers are showing this way for you. Craig's E8500 I would expect to be slightly faster than the QX9650 because it is clocked about 5-6% faster, but the disparity between 91.57 and 78.93 is too great to explain by the clock speed difference. You could try turning off 2 of the cores (task manager, Processes... find SolidWorks.exe and right click... then choose "Set Affinity" and set only 2 cores active for SolidWorks) and then running the benchmark again and compare. I would make sure your motherboard, chipset, processor, etc.. drivers are all up to date. It could be that your QX9650 is not running optimally.
I actually finalized my decision to a T5400 with the 3.16GHz quad Xeon. The OVERALL system seemed faster than the T3400, but the punchholder was actually worse at 94 seconds. THAT IS WHY I AM CONCERNED NOW. $4000 vs. $2000.
That is why I was asking all those questions about dual versus quad before. Now I am thinking that I made a bad, expensive decision.
https://forum.solidworks.com/forum/attachments/PunchHolder.jpg
Michael, I wouldn't worry too much about your decision. I personally think you may have made the right one... depending on how long you plan on utilizing this new machine. I think it's clear that as time goes on, more and more software will be optimized for multi-core machines. Solidworks especially, in every new version, will be trying to do this whenever possible. I've already heard that 2009 is one step closer to optimizing code for scalability or multi-threading in more cases.
Long story short, I think as time goes on your system will actually start to seem faster as code gets optimized in the next couple years. However, if you plan on purchasing a new machine in less than two years, you may have spent too much.
Don
Yeah, my time was 59:45.. while running at 3.8ghz...
And that was on fresh new install.. No AV, etc..
Just Windows and SW.. Nothing else installed...
That seemed to make a lil difference.. as Now, currently it's running around 65 with Office, and everythign else loaded on...
So.. Loading on Office, A/V stuff and all that.. That bumped it up 6 seconds..
Stock, this thing is reporting comparable numbers... It ran 87sec at 3.0ghz..
I'll hafta turn it down to 3.0 again and see what it runs now with Office, A/V, etc loaded..
To me, either that data point of 79 sec is bad or Solidworks really takes advantage of DUAL cores somehow
Here is the screen capture, it was actually a little slower this time. This is with office, photoworks and realview running.
Craig
Attachments
3.0 GHz Quad 87 sec
3.16 GHz Dual 79 sec
5% diff in processor speed, 10% diff in test numbers AND the quad had 2 more processors.
Doesn't make sense.
Spec wise the processors are that close to the same that I would guess it to be something else causing the variation. My first guess would be RAM, the T3400 is available with 677MHz or 800MHz and NECC or ECC memory. 800MHz NECC will be the fastest, 677MHz ECC the slowest. Xeon CPU are usually pared with ECC RAM with is slightly slower because of the error checking.
Next I would look at hardware driver versions, the drivers in one may be better optimized for the hardware.
Or it could be as simple as programs such as AV running in the background
It is almost as if Solidworks runs SLOWER with a quad than a dual.
All my tests were done with nothing else loaded or running.
Exactly as Danieal says.. The ECC functions require clock cylces..
Also, some extra info...
my previous system, (Dell 670) with factory software and all.. Ran punch holder at nearly 400... That was with 2 3ghz xeons.. and 4g Ecc ram..
I reformatted, reloading Windows XP then SW.. Punchholder ran around 350!
Same hardware, fresh install of OS & SW.. Knocked off 50 seconds...
For the Dual/Quad stuff... FWIW...
While running the punchholder, Task manager showed all 4 cores pegging out during portions of the rebuild...
I'd imagine, the Patterns may utilize multiple threads... Just guessing though...
---------------
Also, Rich, I just noticed the end of your last post.....the part about the 4 cores pegging at certain times.....
Wouldn't you think that the QUAD would finish faster?? Even at a SLIGHTLY reduced clock speed??
I am comparing Rich's to Craig's directly.....
3.0 GHz Quad 87 sec
3.16 GHz Dual 79 sec
5% diff in processor speed, 10% diff in test numbers AND the quad had 2 more processors.
Doesn't make sense.
Yes I do have 800mhz ECC Ram.
Craig
Now, if you are running a program that uses all available cores, the quads will start to shine.
Someone should propose a new benchmark using PhotoWorks, or Cosmos. I think there'll be a big difference there.
You said you are running right out of the box?? Correct?
If you watch the task manager while running the test, the additional cores ARE used. That is the confusing part.
---------
I guess it comes down to this - why spend $4000 on a workstation that can be outperformed by a workstation costing $2000 based on these test results?
Even though the $4000 machine SHOULD run better.
I'd be willing to bet that even though you may get an occasional spike in the 4 procs, you're not getting a consistent high level of usage. I'd bet your processor averages around 30% for the whole test. People with macs have been forced into buying 4 cores for laptops and 8 cores for desktops, and even though they're forcing the round peg into the square hole, they aren't reporting good performance unless they compare against an old PC.
Throwing cores at SW will only help so much, and only in certain situations.
SW uses multiprocessors, but only for certain functions. This particular single part, single body, with a lot of holes and pattern features, and then the part is not optimally constructed... this is not a great benchmark part. Nothing against Anna, but Ed Eaton showed how he could get a much lower rebuild time than anyone has yet posted on an old P4 I think by just modeling the same part more efficiently.
I wouldn't get bent out of shape about rebuild times on this particular part.
And then Paul Salvador showed that the rebuild time is about 2/3 of the time to regain complete control of SolidWorks, because there is something going on between the end of the time that Feature Statistics records and the time when the cursor does what you want it to again.
Also you need to average several runs, maybe dropping the high and low values. You can easily see a 30% variation in the times reported by Feature Statistics by simply hitting Ctrl-Q and getting the value, then doing it again. The rebuild time changes mysteriously.
Since SW Corp won't do it, we as users should come up with some better benchmarks. A big assembly, an incontext assembly, a complex surfaced multibody part, sheet metal, weldment, casting, etc.
I have parts where my dual core runs over 70%, and spikes to 100%. These are multibody parts, usually. I wish some one from SW would chime in here and tell us something we don't already know.
I understand your points.
This is the crux of my question.....
It seems that even if you have a 3.0 Ghz quad core and a 3.0 Ghz dual core of the same processor generation, all other components being equal, that a single threaded application or benchmark should AT LEAST run as fast on the quad as the dual. But I am seeing in many cases that the dual runs faster. Why is this?? Isn't the quad just 2 duals bridged together?? One would expect the quad to run at the same speed at a minimum and maybe even slightly faster because of the extra 2 cores.
Also, I have never seen a rebuild time differ by more than 1 second.