XPS 630 SLI 8800 GT versus single GTX 280

Chris,

Does Dell have the similar results when the GTX280 is on a full X16 PCIe slot and the 8800GT are on Full X16 PCIe in SLI? That would be a better comparison where interested consumer's were concerned. Also, How about kicking the resolution up a bit to actually take advantage of the GTX280's new architecture and then repost. The GTX280 as all High end cards are never at an advantage with low resolutions. They were designed for 1600X1200 and up. 22" monitor's will be a tossup, below that a 9800Gt may look great compared to a GTX280. Post that comparison with a 30" monitor on Native resolution and see how the 630i or any other fairs in the 9800Gt SLI vs GTX280 matchup. Now That would be much more appropriate. Then do it again X16 VS X16.

Sean

So this proves that a single card setup on the XPS 630 is bottlenecked by the 8 lanes PCIe slot.

What doesn't make sense is how the SLI 8800 GT cards perform worse with 16 lanes PCIe than 8 lanes PCIe slots.

Chris, can you  benchmark Crysis on both machines (Dell Test PC and 630) with the GTX 280 card to see how gaming performance is affected?

 Thanks.

What is the CPU specs on the test machine with 16 lanes, if it is the same as Chris' xps 630 then i would agree with the conclusion.  But 3dmarks scores in not conclusive indicator vs real world performance. 

I have my cpu at 3.2GHz and dual 9800gt in SLI and i can't for the life of me get 3dmark scores as high as someone with the same specs and even with 8800gt. 

madmax2k1 wrote:

What doesn't make sense is how the SLI 8800 GT cards perform worse with 16 lanes PCIe than 8 lanes PCIe slots.

It does if the 8800gt cards are the 8 series GPU's. there is a pretty radical architecture enhancement with the 9 series and even moreso now with the 2XX series GPU's. The 8 series will deal better with less output bandwidth ability of the X8 slot. Give it more to look for and the GPU will have less time to deal with the rendering. Older drivers will probably bring their performance up on a X16 slot. NVidea boxes all their cards into driver updates for the most part. Another reason that updating drivers just cause there's a newer one is not always a better idea.

Sean

It could also be the 64 bit windows has something to do with the GT's on X16. This one's only a Scientific Wild Blank Guess however.

Edit: Just noticed in a reread of Chris' first post that the 8800GT's were the 9 series GPU version. Aside from 64 bit Vista, it looks like it's the 2xx series GPU architecture's where the big change occurs. Provided the drivers are the same in the two tests.

Chris,

If I'm reading the two results correctly, it kinda makes ya wonder if those of US that have been complaining about the X8 PCIe Crippling the 630i for single GPU owners were more correct. It seems that the GPU will not take the 3-5 years to outgrow the limitation of X8 as I was constantly "Bashed" with at DCF. It appears that the issue is not such a small one after all either.
Not all of this difference will be attributed to the X8 PCIe slots however. There is a lot of other Bottlenecks engineered into the Proprietary MOBO/MIO. $60 and a 680i starts looking like it woulda been a better decision for Dell all the time.

Drivers are coming out that enable high end cards to fully utilize all of the 16 datapaths to maximize the efficiency of the GPU's, even if the bandwidth does not fill a full 8 paths. It was the next logical step in GPU evolution. Put that driver in a X8 mode slot and it can't work efficiently looking for something that does not exist. HA! But, what do we know.

Sean

Chris,

This would be some great info for Dell customers to make decisions with. They then can weight their needs for tech against their Dollars to spend. We'll all be winners then. It might be good to give them a 420 test result with the GTX280 (with a PSU to support). For a single GPU customer (me being one), An XPS420 config option with a Little more Beefy PSU to support new high end GPU's (and multi GPU) could end up being a strong demand to go Parallel with a "Light Duty" SLI rig such as the 630i. Even in it's current high end configuration the 420 is an impressive little box. I'd have spent the money on my order for the Better PSU and a couple more fans out of the box if the options were on the Dell order site.

Sean

EDIT: Chris... isn't this heading into the "funner" side of your Liason arena. I love the "Anatomy of Tech". That's why I love my engineering test job. Tinkerin' is fun. I'd love to have the rigs you can get a hold of to play with though! 

Chris,

Just wondering how the next tests are making out.

Sean

I hope you can get benchmarks like this out Chris.

THis will help people and I wish Dell would just run them and show what you should get with a normal percentage variance on all configurations that Dell offers.

I never "liked" the mobo/chipset Dell chose for the 680 but then with the hard wired lock on both slots being x8 was just another knock on consumers.  We can only get so much information about what is what and with Dell typically not disclosing what is what the decision is harder.

I can see where buyers are concerned about loosing performance and Dell should do something about this motherboard.

Not everyone wants SLI or Crossfire and by not giving the BUYERS the CHOICE is the worst option to use.

These are the same old mistakes as the XPS 600, 700,710 and even the two latest 720 and 730 to much lesser degrees.

I keep thinking any time the latest XPS is going to be the one but it looks like that is still a ways off and a benchmark configuration in the web building cart process would be fantastic to help people get what they want.

Thanks a lot for this Chris, we are looking forward to further testing.

I feel that a practice that is of questionable nature is the voluntary "crippling" of machines to encourage buyers to go "higher"... I believe the 8x lane on the 630 falls into that category, but even worse is the fact that those voluntarily crippled features are not marked as such.

For instance, if you were to develop somehow a 15K HDD that is somehow limited to output equivalent to that on 10K drives, then it should be noted as such (I know this is not the same, its just an example)... the PCI-e lane on the 630 should have been reported as a "half" lane when selling the equipment.

In that matter, this forthcoming information from you Chris is most appreciated and we hope this is the first of many forthcoming informations that will allow the customer to not feel cheated by the equipment they bought.

Rafa.

Chris,

It may not be the driver. 1680X1050 is not playing to the 280's strengths in benchmark testing. It may fare better in a game like Crysis at that resolution however. Try again at 1900X1200 res or higher and I'm sure the results will be much more in favor of the 280 card. That card is a waste of money for any monitor under 24"WS.

Sean