ATI and dual-GPU cards

[KimB]

Just saw this while browsing around:
http://www.dailytech.com/article.aspx?newsid=3016

Personally, I don't see much point unless they're going to do it with their higher-end designs, but then the PCB would have to either be much longer, or they'd have to go with a dual-PCB setup like nVidia.

 

[SugarCoat]

At the risk of pointing out the obvious, thats what we've been talking about for a few months

http://www.beyond3d.com/forum/showthread.php?t=28718&page=7

 

[Parousia]

Just saw this while browsing around:
http://www.dailytech.com/article.aspx?newsid=3016

Personally, I don't see much point unless they're going to do it with their higher-end designs,

Personally, I don't see much point period. More than one chip on one PCB has always been a case of diminishing returns for the IHVs and their AIBs. Targetting folks with the extra must-have money really just don't make much financial sense other than for PR/bragging-rights' sake.

 

[KimB]

Personally, I don't see much point period. More than one chip on one PCB has always been a case of diminishing returns for the IHVs and their AIBs. Targetting folks with the extra must-have money really just don't make much financial sense other than for PR/bragging-rights' sake.

Yes, it is. That's why it's typically only useful for one of two cases:
1. The chips are high-end, and thus this sort of thing is the only way to get the desired performance.
2. The high-end chips have prices which far outstrip the performance increase, which could make a product like this make some financial sense.

 

[KimB]

At the risk of pointing out the obvious, thats what we've been talking about for a few months

http://www.beyond3d.com/forum/showthread.php?t=28718&page=7

Oops, I need to look a bit mor carefully, don't I? Guess I never looked at that thread because I've never been interested in ATI's mid-range chips.

 

[Parousia]

Yes, it is. That's why it's typically only useful for one of two cases:
1. The chips are high-end, and thus this sort of thing is the only way to get the desired performance.

Desired performance for who?

To me, this all simply comes down to PR/I-have-the-bigger-dick. It would be interesting see the costs involved for the IHVs and the AIBs compared to profits derived from the sale of such multiple-gpus-in-one-pcb products. One IHV with such a business practice already deep-6ed.

Until a IHV is brave enough to be a new eyes-wide price leader for value, this will remain nothing more than a PR statement. I personally thought ATi really had a thing going with their R300 (already "SLI-ready"), to really re-define the video card price market. Sadly and probably understandably, its success was probably too new an experience for ATi. If we could have 4 R300s on a single PCB for the same price as 2 of the latest NV "SLI" PCBs like we have right now...

 

[_xxx_]

If we could have 4 R300s on a single PCB for the same price as 2 of the latest NV "SLI" PCBs like we have right now...

...it would change nothing, since 4 R300 would still be much slower and without SM3 support.

 

[Parousia]

I was talking about back then (before we even had SM3) re the R300s.

 

[IgnorancePersonified]

well the only thing useful I got from that article was the concept of 2xgpu for graphics most of the time or one for Grapics and 1 for physics some of the time - probably ineffective in in this generation but this wouldn't be the 1st tech where that was the case.

 

[_xxx_]

I was talking about back then (before we even had SM3) re the R300s.

They didn't need it. R300 was the fastest/best out there, so why should they? Especially after their former efforts were not that successful (MAXX etc.)

 

[Parousia]

They didn't need it. R300 was the fastest/best out there, so why should they? Especially after their former efforts were not that successful (MAXX etc.)

You have a point especially when the R300 had a big lead over anything any other IHV had back then but I don't think they need it now since need seems to be the criteria.

 

[_xxx_]

You have a point especially when the R300 had a big lead over anything any other IHV had back then but I don't think they need it now since need seems to be the criteria.

But now the competition has it, thus they "need" it as well

 

[KimB]

Er, we started getting multi-GPU technologies pretty much from the moment that PCI Express was released. Back then, we were on AGP, and there were engineering challenges related to making two chips look like one.

 

[_xxx_]

You forgot 3dfx Voodoo1/2, methinks

 

[KimB]

You forgot 3dfx Voodoo1/2, methinks

Of course not. I just don't think they're useful to the discussion.