1GB vs. 1.5GB? That’s a pretty big difference in price for 512MB.
The Mac Pro has 4 PCI-E slots. Since these adapters are dual slot, I should be able to fit one in slot #1, another in slot #3.
The FX4800 is 150 watts (according to the nVidia site). I can’t find a clear number on the GTX 285’s power usage. HotHardware.COM claims it’s 185 watts, but I don’t know if the PC version is different other than EFI.
Running kernels on my Macbook Pro (laptop with 8600M GT/512MB VRAM) certainly does get the fans on quickly. I think it sucks more power than the CPU!
The GTX 295 pulls 289 watts, which explains why nVidia didn’t bother with making a 295 for Mac…what would you plug it into? Maybe nVidia can add a 120V plug directly into the side of the PCB…
I have two 285s in a 2008 Mac Pro and both are working under CUDA already. E.g. MonteCarlomultiGPU. I power one card from logic board and other from external PSU. One is in slot 1, the other in slot 2.
In slot 1 I have Mac 285 with 1G RAM. In slot 2 I have Palit 2G card loaded with same drivers but post boot injector. You can get 2G cards from both Palit and EVGA now and I regard it is affordable baby Tesla.
I have not done much with OpenCL yet but some demo progs seem to see both cards under bootcamp. Both my cards together cost less than Quadro but of course there might be stability/driver differences that I cannot speak to. The one issue (almost certainly an early adopter problem - I am sure it will be fixed) is that the current Nvidia drivers give the Mac 285 low speed operation by default. If you run deviceQuery it says the card is at 0.6Ghz. If you run nbody and then dQ again you get 1.48Ghz. I am hoping this is being sorted out…
A 285+295 is also possible. While 2x285 could be powered from motherboard+molex in optical bay (I did this with 285+8800, a 3 connector rig, 2x285 needs 4), I think a 285+295 would really need external PSU. I have a Toughpower box on standby for this next experiment, though I want to see 285 power management sorted first.
No arguments from me. The Quadro series tends to be targeted at a more upscale market, and charges a lot more for similar specs to the GeForce cards. I just wanted to be clear that the GTX 285 was not globally better than the FX 4800. (Although you can get now a special 2 GB version of the GTX 285 which has more memory and better floating point performance, but is still cheaper than the FX 4800.)
That sounds like a winner. Is the GTX 285/2GB available for Mac (i.e. with an EFI Video OPROM)? I’m not much for hacks, especially since most of my code looks like one big hack. SOMETHING has to be dependable… :">
This is going into a Mac Pro 8-core Nehalem. The 295 doesn’t have an EFI BIOS and my preference is not to start introducing hacks. Dual 295’s sounds just dreamy…nearly 1000 cores(!), but I would certainly need to rig an additional power supply, possibly even another circuit breaker. I can see it now. The dual 295’s come out of low-power mode and the lights go dim in Northern California.
I know the Mac Pro 8-core is only a $5K machine, but why did Apple choose to provide such little power in their ONLY machine with PCI-E exposed?
How hard is it to power a card from an external PSU? I’ve never tried anything like that before. Are there external PSU’s with the appropriate 6-pin (?) connectors, and you just plug it into your card? I’m looking to run a GTX 285 with a MacBook Pro, via an ExpressCard-to-PCIe expansion module. Magma makes one that holds one card and isn’t too crazy expensive, but only has a 60W PSU. The GTX 285 needs 204W. Do you think this might work, or have any suggestions?
Regarding the slow transfer rate of ExpressCard (PCIe x1, I think), it shouldn’t be a major issue for my project, because I’ll be able to do 99% of my data transfer to the GPU only once during initialization of a computational run.
Me thinks you are hell-bent on spending many nights barking at the moon. Too many things could go wrong and just about nobody will provide you much support if it doesn’t happen to work.
Either get a cheap Win/Lin-box with PCIe x16 and a big PSU or sign up for Apple’s Developer Connection and buy a low-end Mac Pro. BTW, the ADC discounts are HUGE! http://store.apple.com/us_adc_38839 . Base Mac Pro is $1999.
To be clear I do this with the card in a Mac Pro 08, so there might be other issues with the “portable” rig you propose. But otherwise this is EASY. I tried it out with a really cheap ATX PSU that had a couple of 6-pin connectors already connected, and it worked fine. Now I have Toughpower with oodles of 6 and 8 pin outputs for a future 295 experiment.
There is a `small matter’ of sticking a paper clip (metal!) between 2 of the sockets to fool the PSU into thinking it is in a computer, but this is not as strange as it sounds at first and I have been running my Mac for weeks like this now.
This is all fine with a Pro, but for going around doing presentations I am holding out for a Laptop (Mac pref) with a proper GPU (cc 1.3). I thought the 260/280M might be getting us there and then felt let down. I share the other poster’s concerns about this all working, but power is not the problem here.
Yeah, I’m feeling wary too, it’s becoming more of hack. But it’d be nice to be transportable with my setup. Here’s what I’ll do if I decide to try it:
Magma EB1 bare board set (no enclosure or PSU) - $600 from Magma
Add a PSU
Add an enclosure
The EB1 can’t be used as is because it’s only got a 60W PSU and isn’t wide enough for the double-width GTX 285. Magma says the EB1’s been tested with the MacBook Pro for non-GPU applications. Hopefully CUDA would find it easily as a PCIe device. It’s only x1 speed, but that’s ok for my needs, at least now.
The savings quickly rises above the cost of the ADC membership if you get the octo-core Mac Pro. In my case, I saved $1700 by spending $500. Meanwhile, access to Snow Leopard (and OpenCL) is a big plus.