Tesla C1060

On Tuesday I received my brand new Tesla C1060 :) right on the day it was promised by Armari.

A quick word about Tesla suppliers... Although Armari was efficient to deal with, some of the others were absolutely terrible. Possibly because I am a private individual only buying one unit or maybe just because their overall service is terrible. Their names are not mentioned here as I have complained already and it's only fair to give them a chance...

For those not familiar with the specs it has 30 multiprocessors giving 240 stream cores with 4GB of 512bit wide DDR3 memory. It is a rather large double slot 10.5 inch long PCI card.



I was a bit nervous about installation. I have big computer case and large power supply but my entire pc is 3 1/2 years old. I put on the old grounding strap just to be safe and opened up the box the tesla came in. Armari had done a very good job of packing in the oem PNY box inside their own box. Inside the oem box it was again very well packaged with a combination of antistatic bag, bubble wrap and foam with a cd in a black cover on top.

My motherboard is an  Asus A8N-SLIR which for its time was rather advanced with 2 pci express slots. As I have two 8800GT's in there already I removed the PNY 8800 in slot two then removed the backplane cover next to it as the tesla takes up two slots. When trying to seat the C1060 into slot 2 I immediately had a problem: my motherboard has some capacitors towards the side of the motherboard in line with slot 2. For most long cards this wouldn't be a problem as the capacitors are not particularly tall but the Tesla fits really low against the motherboard for the entire length of the card. This is something worth looking at if you are buying a motherboard to fit some tesla's in as a lot of them  seem to have heatsinks / capacitors in line with some of the pci slots.

My only other option was to try and fit it in slot 1. I had wanted to keep my Asus 8800 GT in the machine as it has a very nice cooling fan on it and tends to run much cooler than my PNY. The problem is that the Asus only wants to work in slot 1 and refuses to work in slot 2. It could be something to do with the motherboard but I'm not entirely sure. Once I had removed the Asus 8800 GT I attempted to remove the Tesla from slot 2. Little problem here: the pci express retention clip sits very tightly under the Tesla as it is such a low card. I had to remove my other PCI cards in order to get to it comfortably. The C1060 then fitted very happily into slot 1 and the PNY 8800 into slot 2. Now to connect the power...

The Tesla requires one 8 pin power connector or two 6 pin power connectors. Furthermore it has a warning sticker on it stating that if you use the two 6 pin power connectors they must come from the same rail. My power supply has two 6 pin connectors but I needed 3 in order to power the 8800 too. After a bit of checking to make sure the two 6pin connectors were on the same rail, I hooked them up to the Tesla. Then I used a twin hard drive power connector to single 6 pin adaptor to connect to the 8800. Plugged it in at the wall and started my computer.  The Tesla's little led was green indicating it was happy with the power and the machine proceeded to boot.

The fact that the three and a half year old Asus motherboard accepted the brand new Tesla and that the Tesla dropped its pci express speed is a tribute to both Asus and NVidia engineering.

 Windows XP professional picked up the new card and prompted me to load the driver disk. I rather stupidly followed it's advice and did so, it then happily loaded the old drivers onto the system causing a bit of confusion later when I checked the versions. If you do add a tesla to a system with newer drivers / cuda 2.2 etc rather point the windows driver installer to their location than loading from the older cd.

Once running I tested all the CUDA samples and got very impressive results. The C1060 is at least 2x faster than a 8800 GT. In my setup the Tesla is limited by the bus transfer speeds and my DDR400 memory but in pure processing performance it is blindingly fast once you have the data in its huge 4GB memory.

I did have a problem with one of the samples and some of my apps refused to work. This I traced down to some OpenGL interoperability issues but I'll post more on this once I have more information.

 

In some of my earlier posts I indicated that I didnt see the point of getting a Tesla over a consumer graphics card apart from the increase in memory. How wrong I was! In long running kernels the Tesla only goes up about 6 or 7 degrees celcius whereas the 8800 GT goes up over 30 degrees. It seems more stable in very long runs although to be fair I haven't had a lot of problems with the consumer cards either. The decrease in temperature though is very welcome and the card's huge fan makes very little noise, much quieter than the 8800's fan in fact.

Now that I'm moving into bigger data sets, double precision and longer running kernels, the impressive thermal management and huge memory of the C1060 makes it a very good purchase.