PCI Express SLI by Alienware!

Wonder why Nvidia havent announced anything like this ? Considering they make both the cards and the boards / chipsets.

Would be an ideal way for them to grab back some market share.....
 
MoodyB said:
Wonder why Nvidia havent announced anything like this ? Considering they make both the cards and the boards / chipsets.

Would be an ideal way for them to grab back some market share.....

Nvidia don't make motherboards - just the chipsets. If they do have anything in mind as a reference design, I doubt they'll talk about it until they announce whatever their Nforce PCI-E chipset is going to be.
 
MoodyB said:
Wonder why Nvidia havent announced anything like this ? Considering they make both the cards and the boards / chipsets.

Would be an ideal way for them to grab back some market share.....

Same reason ATi haven't I suppose. I think it would most likely wind up being a support nightmare.
 
anaqer said:

That is the Video Array? I find this pic puzzling:
http://www.hardwired.hu/img/hir2/e32004_alienware_7_big_1084501650.jpg
Is there 3 video cards in all? So for the Intel setup, the DB15 outputs connect from the first 2 cards go into the DVI connector of the third, then the third DB15 output presumably connects to the monitor for display?
:?:
 
My guess is, it's two video cards + a custom made adapter.
Note that the video cards have dual DVI and look positively NV to me, possibly Quadro... :?:


Shit... look at the second picture... back of the PCB... try to look for SMD components usually located near RAM chips... what a curious RAM layout...
Conclusion - that, my dear friends, is a dual PCX 5950 setup. :oops:

cebitall01.jpg
 
The board in the alienware setup look too short to be a 5950, they also look to be sporting tsop ram. Perhaps they are rv370 or rv380 boards, they were on green pcb's weren't they?

<edit> pic 1 and pic 2 are clearly different graphics boards. The 2nd ones do look to be 5950s
 
AlphaWolf said:
Perhaps they are rv370 or rv380 boards, they were on green pcb's weren't they?

The shorter one looks like PCX 5700, although that sort of fails to make sense to me ( why use two water cooled weak-ass cards...? ).
 
anaqer said:
AlphaWolf said:
Perhaps they are rv370 or rv380 boards, they were on green pcb's weren't they?

The shorter one looks like PCX 5700, although that sort of fails to make sense to me ( why use two water cooled weak-ass cards...? ).

Didn't know there was a version of the 5700 with tsop ram.
 
anaqer said:
AlphaWolf said:
Didn't know there was a version of the 5700 with tsop ram.

Even the reference card for the 5700 uses TSOP.
You are mistaking it with the Ultra, I guess.

And I don't think there was a pci-e version of that card announced was there? I still don't think 5700 cards are that short are they?

Maybe its a 5300 or whatever the low end pci-e card was. /shrug
 
Something I've been trying to work out is how they are getting a motherboard that can support a total of 32 lanes (or more) for PCI express. If they are using an Intel chipset then the Northbridge has a capability for 16 lanes (PEG16X), however they must be getting another 16 lanes from somewhere else - I doubt any of the new southbridges have the capability for supporting up to another 16 lanes, since the first PCIe motherboard implementations only appear to have 3 1X sockets. If they are using a 3rd party chip for the extra PEG16X slot then I wonder if there could be any performance bottlenecks channelling the data to that since I would guess it would be communicating to the Southbridge.
 
anaqer said:
Note that the video cards have dual DVI and look positively NV to me, possibly Quadro... :?:

Yes, at CeBit there were a number of Dual DVI PCX 5900's with "Quadro" sickers on them running in systems.
 
DaveBaumann said:
Something I've been trying to work out is how they are getting a motherboard that can support a total of 32 lanes (or more) for PCI express. If they are using an Intel chipset then the Northbridge has a capability for 16 lanes (PEG16X), however they must be getting another 16 lanes from somewhere else - I doubt any of the new southbridges have the capability for supporting up to another 16 lanes, since the first PCIe motherboard implementations only appear to have 3 1X sockets. If they are using a 3rd party chip for the extra PEG16X slot then I wonder if there could be any performance bottlenecks channelling the data to that since I would guess it would be communicating to the Southbridge.

Or maybe they are just jumping ahead in the PCIX design release, and have a different northbridge chipset that supports more then one PCI-Ex slot.
 
It said they are using Intel chipsets - at the moment Grantsdale and Alderwood don't support it (AFAIK) and I don't think they would be giving early samples of newer chipsets - I'd be surprised to see Intel going this route at all (at least, just yet).
 
Do you know what the limit of lanes support that Grantsdale will have? All I have seen on it is 1 peg 16x and 4 1x slots. Perhaps it can handle more but Intel have simply chosen not to add that on the reference design?
 
Pete said:
Don't video cards draw from top to bottom? Can they draw another way?
Video cards do not draw top to bottom although they do refresh top to bottom. Immediate mode renderers draw whatever they're given and I imagine tilers can draw in whatever tile order the hardware designers choose.

My guess is Alienware is using a bridge chip to split the 16x channel to the chipset. Theoretically this would mean the graphics cards are running at 8x, but I doubt this would cause any bottleneck. The CPU probably can't even feed data fast enough as it is.
 
DaveBaumann said:
It said they are using Intel chipsets - at the moment Grantsdale and Alderwood don't support it (AFAIK) and I don't think they would be giving early samples of newer chipsets - I'd be surprised to see Intel going this route at all (at least, just yet).

Most likely using a PCI-E hub/switch like the PES48G: http://www.imcsemi.com/products_pci.html

16 Lane upstream interface, 2x16 lane downstream interface.

The first product of IMC’s new PCI Express switch family, the PES-48G, targets the high performance graphics market and will tapeout later this month. Customer silicon of this device will be available in June. IMC’s PCI Express switch family is rounded out with the PES-48S, which targets workstation and server markets and will be available in September, along with the PES-32S and PES-24S, which are designed for riser-card computing applications.

Initial pricing for the 48-lane PES-48G and PES-48S devices is targeted at $41.00 and $45.00 respectively, with the 32 lane PES 32S at $30.00 and the 24 lane PES-24S at $23.00. IMC will begin to support customer designs in March by offering design kits to a limited number of beta customers. The new IMC family of PCI Express chips is fully compliant with the 1.0a specification.

Seems to be exactly what alienware are using.

Aaron Spink
speaking for myself inc
 
Back
Top