Nexxen wrote on 2024-06-05, 10:51:

Fire Gl is @ 16x 1.1 - works @ 4x 1.1 Other cards 16x 1.1 works @ 8x 1.1 2.0 same @ 1.1...... […]
Show full quote

Fire Gl is @ 16x 1.1 - works @ 4x 1.1
Other cards 16x 1.1 works @ 8x 1.1
2.0 same @ 1.1......

Is this a compatibility issue with old + new? I'll test all these cards on PCIe 2.0 and 3.0 boards on i7 4th gen to check if it is.

Do you have an option in the BIOS to force which PCIe generation is used for a particular slot? Try setting it manually if you do.

mockingbird wrote on 2024-06-05, 12:29:

Nexxen wrote on 2024-06-05, 10:51:

Fire Gl is @ 16x 1.1 - works @ 4x 1.1 Other cards 16x 1.1 works @ 8x 1.1 2.0 same @ 1.1...... […]
Show full quote

Fire Gl is @ 16x 1.1 - works @ 4x 1.1
Other cards 16x 1.1 works @ 8x 1.1
2.0 same @ 1.1......

Is this a compatibility issue with old + new? I'll test all these cards on PCIe 2.0 and 3.0 boards on i7 4th gen to check if it is.

Do you have an option in the BIOS to force which PCIe generation is used for a particular slot? Try setting it manually if you do.

I tried everything but no change. Force Gen 1/2/3... no change, lanes stay the same amount.
With all the changes even on cpu support with newer BIOSes, I wonder if it is indeed a compatibility issue.

Having watched a bunch of videos from reputable reviewers, x16 or x8 doesn't really matter unless it's a lower generation (but even then...).
That's not my issue. Just to clarify 😀

I'll test with a i7 on s.1150 and on 1151 and FM1.
This is strange and interesting.

Mostly because of my ignorance on the subject. Probably dedicated forums will laugh at me.
Does anyone know a good one to go to?

One more thing I checked is the actual presence of lanes on the video cards.

My 1050Ti is (16x 3.0) but only has 8x on the PCB. This is one thing to take into account.
But the others all have 16 lanes. No missing caps or res on the lanes.

Nexxen wrote on 2024-06-05, 12:46:

I tried everything but no change. Force Gen 1/2/3... no change, lanes stay the same amount.
With all the changes even on cpu support with newer BIOSes, I wonder if it is indeed a compatibility issue.

On my ancient Ivy Bridge board, I have to manually set my PCIe to Gen 2 on one of the lanes or else my NVME drive performance is slow.

Having watched a bunch of videos from reputable reviewers, x16 or x8 doesn't really matter unless it's a lower generation (but even then...).
That's not my issue. Just to clarify 😀

Well, it's like this: PCIe Gen 1 gives you 250MB per lane, so if your card is running at 4x on a Gen1 PCIe slot, then multiply that by four and you're getting 1GB/sec throughput. Probably OK for an ancient graphics card, but would slow down an NVMe drive considerably. At PCIe Gen 2, a 4x slot would give 2GB/sec throughput... Still not ideal, but again, more than adequate for an old graphics card.

I'll test with a i7 on s.1150 and on 1151 and FM1.
This is strange and interesting.

Keep in mind that some of the PCIe lanes are offered through the CPU, and some through the southbridge... The southbridge PCIe lanes are usually slow because they're shared with other things like USB3 and the SATA controller. Test with the first PCIe slot, which always goes through the CPU's PCIe lanes.

mockingbird wrote on 2024-06-05, 14:05:

On my ancient Ivy Bridge board, I have to manually set my PCIe to Gen 2 on one of the lanes or else my NVME drive performance is […]
Show full quote

Nexxen wrote on 2024-06-05, 12:46:

I tried everything but no change. Force Gen 1/2/3... no change, lanes stay the same amount.
With all the changes even on cpu support with newer BIOSes, I wonder if it is indeed a compatibility issue.

On my ancient Ivy Bridge board, I have to manually set my PCIe to Gen 2 on one of the lanes or else my NVME drive performance is slow.

Having watched a bunch of videos from reputable reviewers, x16 or x8 doesn't really matter unless it's a lower generation (but even then...).
That's not my issue. Just to clarify 😀

Well, it's like this: PCIe Gen 1 gives you 250MB per lane, so if your card is running at 4x on a Gen1 PCIe slot, then multiply that by four and you're getting 1GB/sec throughput. Probably OK for an ancient graphics card, but would slow down an NVMe drive considerably. At PCIe Gen 2, a 4x slot would give 2GB/sec throughput... Still not ideal, but again, more than adequate for an old graphics card.

I'll test with a i7 on s.1150 and on 1151 and FM1.
This is strange and interesting.

Keep in mind that some of the PCIe lanes are offered through the CPU, and some through the southbridge... The southbridge PCIe lanes are usually slow because they're shared with other things like USB3 and the SATA controller. Test with the first PCIe slot, which always goes through the CPU's PCIe lanes.

Thanks!

I have to look at the block diagrams to check lanes and their distribution.

It's amazing to see how top cards yield at 1 - 4.0 PCIe.
Still I need to look more into it, being new to this PCIe thing. I never bothered to learn anything and it shows 😀

I tested all the cards I have on a 4th gen i5.
Before drivers I got odd link reports: 2 x 2.0, 4 x 1.1...
After drivers all checks.

I thought that speed and links were independent from any drivers.
Unless it's a Gpuz and HW are both wrong at the same time.

Nexxen wrote on 2024-06-06, 22:53:

I thought that speed and links were independent from any drivers.
Unless it's a Gpuz and HW are both wrong at the same time.

When you run GPUZ, there is a "?" next to where it says "bus interface". Click that and run the render test... Does your PCIe speed change when you do this? Mine shows @16 1.1, but then jumps to @16 3.0 when it runs.

mockingbird wrote on 2024-06-07, 12:38:

Nexxen wrote on 2024-06-06, 22:53:

I thought that speed and links were independent from any drivers.
Unless it's a Gpuz and HW are both wrong at the same time.

When you run GPUZ, there is a "?" next to where it says "bus interface". Click that and run the render test... Does your PCIe speed change when you do this? Mine shows @16 1.1, but then jumps to @16 3.0 when it runs.

Generation can go from 1.1 to 3.0, and it happens correctly.
It's the number of lanes that is inconsistent. Thus I'm thinking there are incompatibilities with old video cards on modern computers.

I'd like to further the topic but on a dedicated forum.
Any suggestions?

Nexxen wrote on 2024-06-07, 15:10:

I'd like to further the topic but on a dedicated forum.
Any suggestions?

No. I have been on the internet for 25 years, VOGONS is the only forum I recommend.

mockingbird wrote on 2024-06-07, 17:15:

Nexxen wrote on 2024-06-07, 15:10:

I'd like to further the topic but on a dedicated forum.
Any suggestions?

No. I have been on the internet for 25 years, VOGONS is the only forum I recommend.

Helmet on, let's soldier through this 😀

Is there a program that can benchmark bandwidth? Maybe it's just misreported numbers. 4x or 16x but still Gb/s are right.
I don't really want to go too deep, graphics card aren't my thing. I do realize that I need some basics asap.

Thanks for the support.

There are PCI-e bandwidth benchmarks :
1) First one is in AIDA64 (under GPGPU benchmark, NOT "Cache and Memory").
Results for "Memory Read" and "Memory Write" will show PCIe speed (uplink/downlink), while "Copy" is VRAM test (requires DX10 or later card).
2) Another option is 3DMark (2013) that has feature test made to test PCIe bandwidth.

Side note : My Z490 Dark motherboard has "Detect Non-Compliance" option in PCIe sub menu, which when enabled on legacy cards fixes any issues with link width I usually see. It needs to be disabled on newer ones (PCI-e 3.0) though, as otherwise it creates those link issues.

agent_x007 wrote on 2024-06-07, 20:30:

There are PCI-e bandwidth benchmarks : 1) First one is in AIDA64 (under GPGPU benchmark, NOT "Cache and Memory"). Results for "M […]
Show full quote

There are PCI-e bandwidth benchmarks :
1) First one is in AIDA64 (under GPGPU benchmark, NOT "Cache and Memory").
Results for "Memory Read" and "Memory Write" will show PCIe speed (uplink/downlink), while "Copy" is VRAM test (requires DX10 or later card).
2) Another option is 3DMark (2013) that has feature test made to test PCIe bandwidth.

Side note : My Z490 Dark motherboard has "Detect Non-Compliance" option in PCIe sub menu, which when enabled on legacy cards fixes any issues with link width I usually see. It needs to be disabled on newer ones (PCI-e 3.0) though, as otherwise it creates those link issues.

So, there are issues with compatibility.
Thanks for the suggestions, I'll dl them and test stuff.