SLI config w/ 2 X1.5GB PCIe x16 nVidia Quadro FX5600 cards, Dell

If SLI is enabled, both cards are linked together by a special direct link and their power is combined to generate each screen frame. Think of it as 1 card drawing the top half of the screen, and the other one the bottom half. As a result, your system looks to have only one card, and you can connect only 2 monitors.

If you disable SLI, you'll have 2 cards in your system, and you'll be able to connect 4 monitors, but the frame rate won't be doubled.

Check GPUCapsViewer at

Short answer, not really. Last I saw, SLI increased performance in SolidWorks by maybe 1-2%, at most. If you don't want to run 3 or 4 monitors, you may as well have gone with one card and just taped $3000 to the side of the case.

I will check that out for sure, I don't know why my company didn't check into this better, a small increase in performance would not be worth the added cost since they bought a few of these workstations with this config. Is there still a SolidWorks benchmark program out there for 2008, I would be will to test it out with both cards on and one card on to post the difference. Thanks.

i am the one that tested this in SWx 2007 and actually got a slower time, 1-2%, running with SLI than i did with SLI disabled. you should be able to do the same test. do a time benchmark on an assembly that takes 3-5 minutes to load. do this at least 3 time with a reboot between each benchmark. then disable (or enable) the SLI or pull (or install) the second video card. do benchmarks again 3 times with reboot between each one. average the benchmarks times and provide results. you can also benchmark a cosmos, this gives great benchmark results. another item to try is to disconnect network, sometimes this affects results. iQ

"iQ" a écrit dans le message de news: snipped-for-privacy@n1g2000prb.googlegroups.com...

i am the one that tested this in SWx 2007 and actually got a slower time, 1-2%, running with SLI than i did with SLI disabled. you should be able to do the same test. do a time benchmark on an assembly that takes 3-5 minutes to load.

??????

It seems to me that video cards are for better frame rates, not shorter loading time. I am very much surprised that you even find any time difference between the 2 configs, with and without SLI. Maybe there is some processor cycles that goes into SLI processing.

I imagine that there could be a difference (SLI better) rotating a large assy, with some semi-transparent parts...

Now, if you plan on fragging some opponent parts...

I don't understand this either. Loading (hard disk and file system dependent) and Cosmos (CPU dependent) should really not be effected by the video card.

I think the best thing to do might be to download SPECviewperf

Even a single FX5600 is not a great performer for SolidWorks. It's optimized for games development and high frame rates, not for CAD. A $2000 T3400 w/FX1700 graphics would be faster.

It would help to actually read the post. He's talking about the Quadro FX5600 with 1.5GB of memory, not the old GeForce FX5600. It's the highest performing workstation video card nVidia makes.

I'm talking about the Quadro FX5600. It's targetted to Digital Content Creation (games development, special effects, etc.) where high frame rates and Genlock features are more important than accellerating CAD geometry. It performs well on synthetic benchmarks, like Viewperf, and DCC applications, like 3D Studio and Maya, but NOT on CAD application benchmarks, like SPECapc for SolidWorks.

The FX4600 is the top-end card for CAD, but the FX1700 and FX3700 are based on newer technology and are slightly faster.

One thing to note is what type of ram is being used on the card. I found that ram type can make the card faster or slower and it may be that an new ram type on a older card design may make the graphics card faster. when i was looking at this with SLI i found that the newest card (Quadro 5500) had DDR2 memory and the step down card (Quadro

SPECviewperf

I am downloading the SPEC program now. I checked for SLI in the Nvidia Control Panel and don't even have that option anywhere, I updated the drivers, installed a bridge cable between the cards and checked in GPU Caps program for SLI, it just says disabled, I don't know how to enable it. I will try the test with two cards in and then one card in. Also, one monitor on one card, one on another. I will post soon. Nobody from dell knows what I am talking about when I called tech support, I think when the computers were ordered, they were configured for QUAD view, not SLI mode. You have both options when buying two graphics cards. Don't know if I can fix that now without sending the whole computer back. I will run these benchmarks now anyway to see what diff. two are making.

Oops, guess I was the one guilty of not fully reading your post. My apologies.

You might not be able to enable SLI because they don't have an SLI bridge installed. It's a small connector that runs between the two cards at the edges.

The results of the test are below, restart in between test, one card in, test run 5 times, both card in test run 5 times.

Manufacturer : Dell Inc. Model : Precision WorkStation T7400 OS : Microsoft Windows XP Professional OS SP : Service Pack 2 CPU : Intel(R) Xeon(R) CPU X5482 @ 3.20GHz # of CPU : 8 Memory : 2813

WITH just ONE Quadro FX5600 with one monitor plugged in, average was:

Test Total =3D 213.61 Graphics =3D 87.39 CPU =3D 59.35 I/O =3D 66.87

WITH just ONE Quadro FX5600 with TWO monitorS plugged in, average was:

Test Total =3D 220.41 Graphics =3D 93.4 CPU =3D 58.47 I/O =3D 68.54

WITH TWO Quadro FX5600 with TWO monitorS plugged in ONE card, average was (no SLI, second card enabled/plugged in):

Test Total =3D 213.86 Graphics =3D 91.94 CPU =3D 56.53 I/O =3D 65.39

WITH TWO Quadro FX5600, one monitor on one card, one monitor on another, dual monitors, average was (no SLI, second card enabled/ plugged in):

Test Total =3D 247 Graphics =3D 125.29 CPU =3D 56.95 I/O =3D 64.76

FROM what I gather, one card with two monitors is my best for now, having the other card makes no diff. and even slows performance when using one monitor on each card verse both monitors on one card. WOW, I didn't expect these results.

I would run a test with two cards bridged in SLI mode with one monitor BUT when I bridge them together with the bridge cable, I never get any SLI messages or options in the control panel. I have ran two Graphics spcec programs and both say SLI Disabled, but compatible. I have no idea how to turn it on. I have searched forums and google and gather it maybe a motherboard swith for SLI or in fact the bridge may not be needed because its done through the PCI-e bus, but that still doesn't help me in enabling it. I have tried Dell drivers, Nvidia Drivers, but no luck. I contacted about 5 different people from dell, no one knows how to enable SLI or even what SLI is... Well, we might sell these extra cards since Dell isn't in agreement on just sending the cards back without sending the whole computer (3 of them). Time to talk to the boss.

You have two FX5600 graphics cards and only 3GB of memory? That's your problem. Windows has to map video memory even if it is onboard the graphics card. Your two graphics cards are stealing all the memory - and using up to 340W of power. SPECapc is a relatively small dataset, so performance isn't impacted too much, but performance with real world assemblies would be dismal.

Replace the two FX5600's with ONE FX1700 or FX3700 and your performance should improve dramatically. A system like yours should also be configured with 8GB or more RAM and 64-bit OS.

For reference, a single Core2 Duo e6850 (no longer top of the line) with an FX1700 graphics card completes the SPECapc benchmark in 167 secs. A single Core2 Duo e8500 with an FX3700 does it 125 secs. (You can't really compare these two scores and they are done under diffrent testing conditions. You CAN compare scores of various machines tested by Desktop Engineering and scores of different machines submitted to

I have an older computer with a fx3450 QUADRO in it, i took it out and put it in this new computer taking out the quadro 5600's. I updated the driver, rebooted and ran the spec test again, here are the results:

Manufacturer : Dell Inc. Model : Precision WorkStation T7400 OS : Microsoft Windows XP Professional OS SP : Service Pack 2 CPU : Intel(R) Xeon(R) CPU X5482 @ 3.20GHz # of CPU : 8 Memory : 3325 uadro FX 3450/4000 SDI/PCI/SSE2

Test Total =3D 208.17 Graphics =3D 94.84 CPU =3D 51.43 I/O =3D 61.9

Help me out here, are you saying I should update to 64 bit? How come I am not seeing the speeds in those results from the links? Thanks for the advice.

X5482 =A0@ 3.20GHz

If you could spec me out the perfect SolidWorks Computer, what would be in it and where from, this may help if we send this back under the

You just improved your time quite a bit - with a mid-range graphics card that is a couple of years old and NOT a top-rank card for SolidWorks 2008. (The FX3450 gets a "gold ball" from SolidWorks because it will support all 08 RealView features, but it runs an older shader model and doesn't fully accellerate all the new features.) A current generation FX1700 or FX3700 would be even faster. I bet you'd see some improvement with just one FX5600 instead of two.

I can't say if I'd recommend 64-bit for you or not. 64-bit is recommended ONLY for those working with large datasets. (Which doesn't necessarily mean large assemblies. You can have a 10,000 piece assembly that puts little strain on SolidWorks and a single part that will bring it to it's knees. It's the number and type of features that matter.)

Same for optimal computer - it depends on what you do. If you just run SolidWorks with limited rendering and/or analysis, then a single Core2 Duo system is best (64-bit with 4GB+ RAM if your datasets are large.) If you do a moderate-heavy amount of rendering and/or analysis, then a single quad core CPU with 64-bit Windows and 4-8 GB of RAM is worth the extra money.

Only users doing *extensive* rendering and FEA are going to benefit from more than 4 cores and/or more than 8GB of RAM. IMO, those are the only ones who should be spending the money on a Xeon-based system. I really can't see the point in having an 8 core system that is constrained by a 32-bit operating system.

Thanks, i have learned a lot. Just after the fact though, live and learn.