best cuda gpu on HPZ620

[fracara]

I actually have a teska K40 in my Hpz20 that is very good when coupled with specific software that support it: I was wondering what's the best of old cuda card that is possible to install inside.

K40 is officially supported by Hp manual so my purpose is to find a better and compatible card (not new) to replace without spending too much.

I know that someone in this forum mounted a GTX 1080 and it's a very good card, but I'm finding a card of Tesla family.

thanks and happy new year

[Brian1965 · Accepted solution]

With regards to Tesla cards for the HP Z620, I would say that you currently have the best/fastest 'Tesla' card based on the following;

1)  HP recommends not exceeding 225W on the secondary GPU/compute card slot. The K40 is a 235W card.

2)  The newer generation Tesla cards, e.g. M40, P40, etc., only have passive cooling so are only suitable for rack type (forced-air) servers.

3)  Newer Tesla cards are still very expensive.

I put a GTX 1080Ti in my Z620 (PCIe slot 5) to primarily use as a secondary GPU/CUDA/compute card for 3D animation and video rendering. Due to the way I have this configured, I can select either my Quadro P2000 or GTX card to be the primary GPU without the need to reboot. I have ran render projects lasting >24 hours with the P2000 as the primary GPU and the GTX card as a compute card without any issues. The GTX is a very fast card and also considerably cheaper now that the newer range RTX 2000 and 3000 series cards have been released. I also took the added step of converting the GTX card to liquid cooling.

GPU                      CUDA       Memory     Bandwidth (GB/s)    TDP

Tesla K40            2880          12GB                   288                   235W

GTX 1080Ti        3584          11GB                   484                   250W

Unlike the other GPU cards in the GTX 10 series, the GTX 1080Ti card uses the same chip as the Titan X and TitanXp cards.

[fracara]

Thank you, I think this will be my next upgrade because some software I use can benefit of 1080Ti also if it isn't a "professional" line card. I think will follow your setup to try to use gtx as secondary and try to leave existing k2000 as primary, the only concern I have is that i connect to PC via remote desktop. When I'll buy, I'll ask for your help 😀. Thank you for the informations.

[Brian1965]

If you are planning on simply using the GTX card as a secondary CUDA compute card then I don't think you should have any issues using a remote desktop app. I only use a single display/monitor with my set-up, (1x Samsung 34", 21:9 aspect ratio curved display). Will the remote Z620 be running in headless mode? (This is a setting in the BIOS that allows the Z620 to boot without a monitor attached).

[fracara]

I checked headless mode is disabled, and it starts normally without an attached monitor.

Does gtx fit in slot 5? Because it's impossible for me to insert tesla in slot 5

[Brian1965]

You can place the K2000 card in PCIe slot 5, and the GTX card in PCIe slot 2. There is a setting in the BIOS that lets you select the primary GPU PCIe slot, after it detects two GPU cards that have monitor output ports. This setting is not visible in the BIOS if you are using a GPU compute card, as compute (i.e. Tesla) cards do not have any monitor output ports.

I would also recommend you go for the GTX 1080 Ti Founders Edition, (GTX 1080 Ti FE), as certain gaming or overclocked versions are generally bigger in physical size, (wider and/or taller), and may cause issues with the computers side panel. Prior to converting my GTX to liquid cooling, I ran it in PCIe slot 2, (with the P2000 in PCIe slot 5), with the standard GTX air blower installed without any issues, though the computer was considerably louder compared to the liquid cooling now installed.

I would recommend you follow the method I used to install the GPU drivers;

https://h30434.www3.hp.com/t5/Business-PCs-Workstations-and-Point-of-Sale-Systems/SOLVED-Quadro-P200...