Hi, I plan to buy a laptop which can run and develop cuda programs. The budget is $1500-2500 and I hope this laptop can support cuda “COMPUTE CAPABILITY 2.0”. I am not very familiar with CUDA, but I have some experience of running GPU code on several windows 32-bit desktops for my research. (I bought a 9500GT on my desktop and it worked for my old cuda program downloaded from internet. However, I found this card can’t support matlab 2010b gpu function, it is really frustrating because this card was not cheap but out of date so fast.)
After doing some research on the internet, I found it seems many people have support/compatibility issue about their Nvidia GPU of notebook version. The worse case is they can’t run cuda on their laptop even it is cuda enabled, and in some case they have to wait for the update of the OEM driver for a long time even though there have been an official nvidia driver for laptop.
My question is:
Is there any laptop which just uses official nvidia GPU notebook driver such that I can always get driver update from Nvidia like in the case of desktop GPU? If not, which laptop on the current market have the best support/compatibility for cuda development(It means the manufacture will keep updating their GPU driver)? This laptop is a huge investment to me, so I really need to make sure I can run/develop the cuda code on it for at least 3 years.
I am also considering external GPU like “ViDock 4 Plus”. It seems the best solution on support/compatibility, cuz it installs desktop GPU card on laptop. However, I am still not sure about if it can really works like an desktop GPU on all the respects especially on cuda programs. There is not much information/review about this product on the internet. This company seems not a huge company so I am still concerned about the future support. The most useful review I found is: http://www.lenovox201.com/2010/11/vidock-4-unbox-hardware-setup-software.html If anyone have more information about running cuda code on ViDock, I am willing to hear it.
If there is any suggestion about my questions, I will appreciate it.
There are quite a lot of notebooks that do support CUDA. I am currently using a Samsung R580 which was about 650â‚¬ last june, and it features a 310M GPU with CUDA 1.2 capability. I am using this device for GPU development currently. However, as technology progesses there should be lots of 2.0 enabled notebooks out there.
Also I might have to mention that I am only using linux on that notebook(there is a windows xp copy installed on another partition which I hardly use which has the CUDA driver + SDK from nvidia installed and I ran the samples and they worked
also I would not recommend those external GPUs for something like GPGPU as they are not able to provide the needed bandwidth for HostToDevice transfers.(this is pretty much a big bottleneck!)
But theres some special shops that do sell minipci based GPU devices - I cannot provide any recommendations here, but I suppose that those could be a better solution.
I’m going to purchase dell xps17 laptop with 3gb gpu, supports cuda 2.0. I’m just waiting for sandybridge cpu update for that model. I’ve never had problems using notebook drivers on dell laptops (currently using older version of xps laptop).
if this info can help you in choosing a laptop for cuda development take in mind that many laptops integrate a primary graphic board (Intel chipset) and the nvidia one. This second board switches automatically if the workload requires it (under windows).
Cuda applications do run under windows but it seems it’s not possible to exclude the intel graphic board and set the nvidia as the primary one.
Under Linux it seems it’s possible to activate the nvidia graphic board only with nasty tricks and after spending energies and hacking.
I would like to develop under Linux cuda apps on my laptop but cannot do it for software/hardware limitations - there is no BIOS switcher to select hybrid mode - primary integrate intel GB - secondary discrete nvidia GB - and it seems the nvidia drivers cannot figure how to manage this strange hybrid situation.
Dell xps17 laptop with 3gb gpu sounds fitting my need. I will consider this machine!
I am wondering why doesn’t Nvidia just produce a laptop by themselves which guarantee the compatibility. Just like Google has a Google phones for android. The compatibility issue of notebook GPU is really annoying…
I think maybe a mini desktop is also an option as a n portable machine which is necessary for some application such as program demo. But since my old dell desktop (Inspiron 530,Intel Pentium) suffered from the size and power supply issue when installing a GPU card, I am still not sure if a mini(portable) desktop could install a powerful GPU card such as GTX 460.
I am also looking for a laptop with a good config to start with CUDA. However, after going through several threads from different forum, I am still a little bit confused on the right choice to make:
Going for one of the new Fermi card (like the 400 series, GT420/435/445 in Dell configs for example), as they feature CCC 2.x, but most of them come with Optimus, which seems to remain a big pain for linux users.
Going for older high-end graphics (like GTX260M, GTS350M) which won’t feature Optimus, but also be CCC < 2.0
Going for older mid-range graphics (like GT330M, GT240M) which would be definitely cheaper but of less potential.
I will mainly perform numerical computation on images/volumes (reconstruction from MRI data). Every opinion is welcome, as well as advice on what TO or NOT TO buy. Thank you guys.
Not all xps laptops have optimus - only those who have integrated intel video, and currently those are only the ones with Corei5 model. Corei7 variants don’t have it currently. New sandybridge cpus will have optimus on all cpu versions.
I’m actually looking forward to integrated stuff accompanying discrete gpu, cause then I’d be able to play with them both at the same time (I hope intel will come up with an Opencl driver for it), but I’m a microsoft fan External Image.
I’d really NOT buy older card just because they only support older cuda standards (<=1.3), and even if you think that now you are not going to make use of fermi, some time later you will regret that you didn’t buy cuda2.0 capable solution.
I’d go with at least 445m with 192bit memory interface - it has more than 2 times better memory bandwidth. Also, I’ve managed to write cuda bugs which only showed themselves on GPUs with big number of SMs. GT435 afaik is for 15’ laptops. GT435m and further is for 17’ laptops.
I know gtx 460m is much more powerful than gt 540m, but im not pretending to game that much, so maybe i really dont need the ASUS, my main concern is about learning CUDA…A nice pro for y needs would be linux support, but if that is not possible I will develop under windows…
On the other side, I have read there are issues with optimus technology. Is that true? does the gtx 460m implements that technology?
Thanks a lot