Quadro with GTX for more CUDA

[nuessli]

Hi Guys,

I Just joined here and be glad to see Octane keep stepping forward in good mood.
I am going to upgrade my graphic card at work, actually would be one or two GTX card topped on though.

I have a Quadro FX4800 comes with Lenovo D20 the other day. Quadro FX4800 have around 190 cudas, for sure it's not good enough either no reason to use GPU based renderer. So, I thought that add up GTX cards to have better cuda boosting up.

My PC have 2 slots of PCIe x16 and 1 slot of PCIe x4 (x16 physical) - I don't know what that means of (x16 physical) - and one more PCIe x1 slot.

Please give me your thoughts what would be good combination with which GTX model or share your experiences with me would be even better.
It has 1060W power supplym hopefully it would be good enough to have two more cards.

Thanks in advance for your information.
J.

Win 7 64 | Quadro FX4800 | Dual Xeon 2.93GHz | 24GB

[FooZe]

Hi Nuessli,

If you take a look at your PCIe slots you will notice they are different sizes. The x16 physical means it is the size of a standard 16x slot even if it may be running at a lower speed.
This allows it to still take a graphics card.

1060W should be enough for 2 additional cards although i think it would be good try to find out the power requirements for each card and do some quick sums. Usually NVidia lists the max power requirements in the specifications page for them on the NVidia website. I notice you have a dual xeon and 24GB ram so you may be pushing it.
At the moment the best performing cards would be the 580 and 590's however these are starting to be hard to get your hands on.
Here in New Zealand we have been trying to source additional 590's but they are nowhere to be found.

The 6xx series is slower at the moment than the 5xx series in octane and even after optimization may stay that way - we can make no guarantees as too what the performance will be after optimization.
But on the up side the 6xx series can have more VRAM and use less power.

If you go with 2 590's or 2 690's then you should definitely check the total power requirements of the cards as these have 2GPU's per card and hence far higher power requirements.
Also these are usually listed with the total memory on board, not per GPU. ie: a "3GB" 590 will have 1.5GB per GPU therefore only 1.5GB available for octane.

Cheers
Chris.

[nuessli]

HI Chris,

Thank you for the information.
It sounds like when it installed in PCIe 4x same size with 16x, whole other CUDA cores seats in PCIe 16x works slower than actual 16x speed.
Am I right? or only one card in PCIe 4x would be slower than 16x guys.

You've pointed my CPUs and RAMs though, would that be helpful for Octane as well?

Definitely, I would go for 590 1.5G if I could find it in the market as you recommended.

Thanks a lot again and have a wonderful weekend.

J.

[FooZe]

Hi nuessli,

The bus speed (4x,8x,16x) doesn't mean much for render speed. It will help with loading the scene but on the most part it doesn't matter much for octane. The cards should render at full speed since all the work is done on the GPU.

Ram and CPU also only matter for loading (parsing and voxelization) and for power usage.

Cheers
Chris.

[nuessli]

Hi Chris,

Thank you for the explanation.
Well,,, as you said earlier, it would be very hard to find GTX 590 in the market.
Any suggestions with GTX 6xx series, would perform similar with 590?

Thanks a lot.
Jay



Hi nuessli,

The bus speed (4x,8x,16x) doesn't mean much for render speed. It will help with loading the scene but on the most part it doesn't matter much for octane. The cards should render at full speed since all the work is done on the GPU.

Ram and CPU also only matter for loading (parsing and voxelization) and for power usage.

Cheers
Chris.

[mib2berlin]

Hi nuessli, the Kepler (GTX 600) series performance is about 60-70% compare to Fermi (GTX 500).
The Kepler cards a high optimized for Gaming.

Cheers, mib.