Hi, I have been crawling the web for information on which mobile GPUs will have 1.3 Compute Capability.
All the GTX 2XXM series are based on the 94 architecture and so that’s a no-go.
The best I have found are the GTS 250M / GTS 260M / GT240M / GT230M / GT210M (which do not appear to have been released in notebooks yet), but the information not definitive; along the line of “based on the new 200 series architecture”.
Would anybody know whether these support 1.3, or whether there are any mobile GPUs that will be released soon with 1.3 compute capability?
Hmm, the Dell / Alienware M15x appears to have a GT240M, upgradable to GTX260M at extra cost.
Cheaper are models from ASUS, e.g the ASUS G60VX-JX006K equipped with a GTX260M also
Still far from “affordable”, so I am sticking with the previous generation mobile GPUs.
Hi, I just bought a new ASUS G51VX with NVIDIA GTX 260M 1GB inside.
I used “deviceQuery” appliaction from CUDA SDK 2.3 and found that its compute capability is 1.1
I am wondering, is it a real compute capability of GTX 260M?
GTS 260M is compute 1.2. They should be available “real soon now” from OEMs… likely they’re all waiting for Windows 7 launch to fire off the new models.
This page is often useful… not always complete, but perhaps the best reference. The 40 nm GDDR5 chips are compute 1.2. It’s unclear about the GDDR3 40nm chips, but probably.
Dunno, soon I hope. Double-precision hardware apparently adds a lot of extra circuitry to the chips though, which saps power, etc., which is why you don’t see any mobile versions. I’ve been keeping an eye out for a double-precision mobile chip, but neither nVidia or ATI looks like they’ll have one until at least the NEXT generation of cards (ATI 6xxx or post-Fermi).
I don’t think we will see mobile double precision hardware until one of the mainstream APIs (ie a DX11 successor) demands it, and even then mobile parts usually trail their desktop parts by half or one generation in API support. Like profquail says, it is a lot of extra logic and die area to tie up for little return as things stand. I would expect that mobile GPUs of the future will start looking more like Tegra, ie. lots of dedicated fixed function hardware for multimedia acceleration as well as reduced function desktop derived 3D GPU stuff, than the parts we have to day. It makes more sense from a power consumption point of view to do so. A lot more so than double precision, at least for the mainstream market.
Well thanks for the clear answer! It would help if the CC level could be routinely printed in the product web pages and literature.
It is an odd business though - there is a mobile workstation market, and while it is small it could be well served by some DP-capable GPUs, perhaps with a smaller number of cores or technology like turbo boost to power down when not in full use. Right now I have to put my big desktop onto a tea trolley to go give a lecture, and I cannot do that if I travel to a conference or go to see a business. I do not care much about battery life either as I almost always plug in - I have not even bothered to replace some batteries that no longer work or charge.
Your best bet regarding Compute Capability is the latest programming guide. The device you mentioned is not listed there, though; but I think there is no mobile GPUs with 1.3 capability, so as device specification mention it is DX 10.1 compatible, I’d guess it’s CC 1.2.
Also: +1 for listing CC in the device specifications. Programming guide is always lagging behind the latest products, so oftentimes there is simply no way of knowing CC of some devices; as NVIDIA product lines naming schemes are extremely confusing, even guessing anything is rather hard.
So you are suggesting that since the GTS 360m is compatible with DX10.1, it might be a CUDA 1.2 capable GPU card ? This is still a speculation.
I have not found a single official NVIDIA specification listing GTS 360m as a CUDA 1.2 capable device.
The naming schemes are much more confusing and misleading for notebook GPUs. For example GTX260 desktop version is a CUDA 1.3 capable card, but GTX260m (mobile version) is only a CUDA 1.1 capable card.