Intel Optane 905p in CPU or chipset PCIe slot?

[JCoel]

I've read several questions here about installing the Optane in the chipset PCIe slot, but my intentions are actually the opposite.

Is there ANY performance advantage to connecting the 905p to the 16x CPU PCIe slot instead of the chipset's x4 PCIe slot (yes, I know the graphics card will then operate at x8 instead of x16)?

If so, what should I expect to see?

I also know that connecting it to the chipset means it's sharing bandwidth with other devices, but I am asking more in terms of actual latency and throughoutput simply because of the direct connection to the CPU.

[JosafathB_Intel]

Hello JCoel, Thank you for contacting Intel® Memory & Storage Support. As we understand, you need assistance with your Intel® Optane™ SSD 905P Series. If we infer correctly we will appreciate if you can review the following information: • Physical Installation Instructions for PCIe* NVMe* U.2 and Add-In Card Solid State Drives (https://www.intel.com/content/www/us/en/support/articles/000026843/memory-and-storage/intel-optane-memory.html) • System Requirements for an Intel® Optane™ SSD 905P Series Drive (https://www.intel.com/content/www/us/en/support/articles/000028298/memory-and-storage.html) • After following to the instructions and recommendations provided on the previous articles and checking with your motherboard original equipment manufacturer (OEM) regarding compatibility with your Intel® Optane™ SSD 905P Series and regarding hardware and BIOS configuration, the results should be similar to the ones published on https://ark.intel.com/products/series/129835/Intel-Optane-SSD-905P-Series#@Intel-Optane-SSD-905P-Series For example the expected performance of the Intel® Optane™ SSD 905P Series 1.5TB, 1/2 Height PCIe x4, 20nm, 3D XPoint™ (https://ark.intel.com/products/129833/Intel-Optane-SSD-905P-Series-1-5TB-1-2-Height-PCIe-x4-20nm-3D-XPoint-) will be: Performance Sequential Read (up to) 2600 MB/s Sequential Write (up to) 2200 MB/s Random Read (8GB Span) (up to) 575000 IOPS Random Write (8GB Span) (up to) 550000 IOPS Latency - Read 10 µs Latency - Write 10 µs Power - Active 18.6W Power - Idle 6.2W • To answer to your question “Is there ANY performance advantage to connecting the 905p to the 16x CPU PCIe slot instead of the chipset's x4 PCIe slot (yes, I know the graphics card will then operate at x8 instead of x16)?” If you connect your 905p to a PCIe slot with more lanes the Optane will just run as designed (PCIe x4) and the rest of the lanes will not be used. In that case, it will be better to use your graphics card on that PCIe slot instead of the Optane. We hope you find this information helpful. Best regards, Josh B. Intel® Customer Support Technician Under Contract to Intel Corporation

[JCoel]

Hi Josh B.

First of all, thanks for replying. :)

Unfortunately I don't think you understood my question, I know that the Optane will not use more than 4 lanes even when connected to a PCIe x16 slot. I also know that if I connect the Optane to the CPU PCIe lanes, the graphics cards will operate at 8x and the Optane at 4x.

What I am asking is if there is any PERFORMANCE advantage (latency, through-output) to installing the Optane in the 16 PCIe lanes that connect directly to the CPU vs. connecting it to the motherboard chipset/PCH (in other words, doesn't connecting the Optane to the chipset instead of directly to the CPU introduce extra latency and therefore less through-output?)

If so, what difference should I expect to see? Does Intel have any data on this?

[JosafathB_Intel]

Hello JCoel, Thank you for your reply. The drives should work at around the expected performance that is available in ARK.intel.com Regarding the performance latency of communication, it depends on your Motherboard specifications, design, and manufacturer. We advise you to check with your original equipment manufacturer (OEM) regarding compatibility and the expected performance of the drive in your system before installing your SSD. We hope you find this information helpful. Best regards, Josh B. Intel® Customer Support Technician Under Contract to Intel Corporation

[JosafathB_Intel]

Hello JCoel, Thank you for contacting Intel® Memory & Storage Support. This is just a friendly reminder that your case remains open. Should you need our technical assistance please do not hesitate to contact us back. Best regards, Josh B. Intel® Customer Support Technician Under Contract to Intel Corporation

[JosafathB_Intel]

Hello JCoel, Thank you for having contacted Intel® Technical Support. I was reviewing your community post and we would like to know if you need further assistance or if we can close this case. We will be looking forward to your reply. Best regards, Josh B. Intel® Customer Support Technician Under Contract to Intel Corporation

[JCoel]

Hi Josh!

Please feel free to close this case. Let it be noted, however, that I never got a CLEAR and DEFINITE answer to my question(s), which was, once again:

1. Is there any performance difference between connecting the Optane to CPU PCIe lanes instead of PCH lanes (yes/no).
2. If so, what would that difference be in terms of latency and/or MB/s?

Thanks,

Jorge Coelho

[DiegoV_Intel · Accepted solution]

Hi Jorge,

There should not be any difference between the two connections you are mentioning. Although this is a not tested nor validated scenario, theoretically the performance should be the same. Even though the drive is connected directly to the CPU, the drive will not perform better because of its limitations inherited by the hardware itself and the design.

The drive’s interface is PCIe x4, so it will work as a PCIe x4 device. Connecting it directly to the CPU in the PCIe x16 slot will not make the drive to work faster because of the drive’s interface limit. It may be possible that there is some minor difference in latency, however the overall performance should be the same so there is no real advantage of making this connection change.

As I mentioned, this is a not validated nor tested scenario so there is no real data about performance differences that I can share with you. You can try to run the test by yourself and compare the results between the connections, although the expected result is to have a similar (the same) performance in both cases.

I hope this information clarified your inquiries.

Let me know if there is anything else I can help you with.

Have a nice day.

Regards,

Diego V.

Intel Customer Support Technician

Under Contract to Intel Corporation

[JCoel]

Thank you, Diego, that DOES answer my question. :)

My reasoning was that latency *should* be better when connected directly to the CPU since data no longer has to go through the PCH bridge (or the drive share the limited x4 bandwidth with other devices). And if latency is better, then the number of IOPS should increase too (and thus throughoutput) even if only by a little bit (admitedly in the end always limited by the drive's interface max, of course) - it's just that the drive would then be operating under optimum conditions.

Then again, as you explained, this might not matter at all because even if latency is better, IOPS are still limited by the drive's hardware and the PCH might be more than able to handle max load (when not being forced to share bandwidth with other devices), which I guess is what you were both trying to say.

Anyway, proof is in the pudding, so I will have to try this for myself - as you suggest - to find out for sure (but now with very low expectations). :)

Please feel free to close this case, and thank you both for the feedback.

[DiegoV_Intel]

Hi Jorge,

That's right, that would be what we can theoretically expect. However, if you want to know what would really happen, the only way would be running some tests and compare the results.

Anyhow, thank you for your response. We'll consider this thread as closed.

If you ever have another issue or concern, please feel free to open a new thread. We'll try to help you in any way we can.

Have a nice day.

Regards,

Diego V.

Intel Customer Support Technician

Under Contract to Intel Corporation