Saturday, December 3, 2022

Corsair CMK32GX4M2D3600 RAM Micron single-rank kit vs CMK32GX4M2D3600 Nunya dual-rank RAM kit vs Asus Prime X570 Pro motherboard

In November 2021, I upgraded my main desktop computer with an AMD Ryzen 5950X CPU, an Asus Prime X570 Pro motherboard, and a Corsair CMK32GX4M2D3600 2 x 16GB DDR4 RAM kit. I installed the two sticks in the DIMM_B2 and DIMM_A2 slots, per the motherboard manual. Everything functioned just fine with DOCP settings with the memory running at its rated 3600 MHz speed.

In December 2022, the same memory kit went on sale for about 35% off, and I decided to buy another to expand the RAM on this computer to 64GB. I installed the two sticks in the remaining DIMM_B1 and DIMM_A1 slots on the motherboard. The computer started crashing with a Windows BSOD after benign tasks such as Zoom video conferencing. I ran CPU-Z which showed that the old 2021 kit was made by Micron and was single-rank, but the new 2022 kit was made by Nanya and was dual-rank. All 4 DIMMs had identical SPD timings for XMP 3600, and identical voltage of 1.35V.

I then ran memtest86 from a USB stick. I had to force it to run in parallel on all CPUs, as by default it would only run on a single CPU on this motherboard for some reason. I left it running a few hours. I was horrified to find the following result :


 

The test had more than 10006 errors in under 3 hours, and aborted. Something was very wrong with this picture.

I pulled out all 4 sticks of RAM, and installed just the new dual-rank kit in DIMM_B2 and DIMM_A2 slots on the motherboard. I ran memtest86 again in parallel. After about 5 hours, it passed with flying colors.

I then repeated the test with just the old single-rank kit in DIMM_B2 and DIMM_A2 slots. Again, memtest86 passed with flying colors after 5 hours.

Finally, I had the idea to install the new dual-rank kit in DIMM_B2 and DIMM_A2 slots, and the old single-rank kit in DIMM_B1 and DIMM_A1 slots. I ran memtest86 overnight. After about 11 hours, it also passed in full.

TLDR :

1) Corsair changed manufacturer for its RAM kit, and went from single-rank to dual-rank, without changing the part number.

2) When mixing single rank and dual rank memory, the Asus Prime X570 Pro motherboard needs the dual-rank kit to go in the B2/A2 DIMM slots, and the single rank kit to go into B1/A2 DIMM slots. This is the only configuration that works reliably. The reverse does not.

Sunday, April 18, 2021

Top 6 pet peeves about Lightroom Classic import

  1. Lightroom cannot move files from my camera's SD card
    Lightroom will copy files from the SD card, but not remove them. Only files from a hard drive can be moved. The result is one cannot just put the SD card back into the camera and continue shooting.

    If one removes files from the Lightroom catalog after copy from the SD card, the next time Lightroom imports from the SD card, those files will be copied again !

    Adobe recommends formatting the SD card after importing before putting it back into the camera. But this is not practical if the SD card contains other files, such as video files, that I don't want to import in Lightroom. Lightroom should have the option to move files from any volume, including SD cards.

    Because of this, I am forced to first manually move files from my SD card to a directory on my hard drive, then importing in Lightroom from that location. That is a 2-step process instead of one. Computers were built to automate tasks, not make them more complex.
  2. Lightroom import presets cannot remember whether to include or exclude video files
    One can use the Alt key in the Import dialog to check or uncheck video files during an import, but this setting cannot be saved into a preset. One has to press Alt again at every single import. I only want to manage photos in Lightroom, and not videos, and this is very annoying to be forced to do every time.
  3. Lightroom import presets cannot remember the source directory location.
    If I want to have separate presets for my SD card and OneDrive directory, it's not possible. I have to re-select the source directory manually every time during each import. I would like to save the location into the preset, but it can't be done.
  4. Lightroom auto-import cannot move files into a tree structure
    I cannot use auto-import to move files into a tree structure by date, which is something that I can do in the manual import feature. All files that are auto-imported must be moved into a single folder. Why does the auto-import feature have to be at all different from the manual import ? Beats me.
    Because of this limitation, I cannot use auto-import at all.
  5. Lightroom auto-import can only watch a single source directory.
    If I want to, say, automatically files from both my Google Drive and OneDrive, it's not possible. I have to choose or the other.
  6. Lightroom auto-import only works from folders that are initially empty.
    Normally, I manually import my pictures into D:\Documents\Pictures\YYYY\MM . I would like Lightroom to have the option to automatically monitor any files that have been externally added or removed from those locations. One great example is files manually created by Photoshop when using Super Resolution. As it currently stands, the only way to import those files is to manually run "Synchronize folder", if one knows the actual folder that Photoshop created the image in !

Sunday, June 21, 2020

Realtek RTL 8156 USB 2.5 Gbps Ethernet 2.5GBASE-T vs Trendnet TEG-7080ES NBASE-T switch

I purchased a Trendnet 7080-ES switch in 2019 . It has served me well with multiple PCs using PCIe Aquantia AQN-107 NICs at 10 Gbps speed.

Today, I received a USB 3.0 NIC from Cable Creation . This is a 2.5 Gbps NIC using the Realtek 8156 chipset. I installed it on a Windows 10 system. Windows 10 automatically recognized the NIC. However, once I started running into performance issues, I updated the NIC drivers to 10.38.20.117 from the Realtek website, dated 1/17/2020 .

This is the result of an iperf test between two systems. They are directly connected by a 10ft CAT6 cable. IP addresses are manually configured on each system. The client is using the Realtek NIC on Windows 10. The server is using the Aquantia AQN-107 NIC on Ubuntu 18.04. Both sides are configured with 9KB jumbo frames.

D:\Downloads\iperf-3.1.3-win64>iperf3 -c 192.168.1.237 -t 10 -O 10 -i 5 -R
Connecting to host 192.168.1.237, port 5201
Reverse mode, remote host 192.168.1.237 is sending
[  4] local 192.168.1.236 port 61654 connected to 192.168.1.237 port 5201
[ ID] Interval           Transfer     Bandwidth
[  4]   0.00-5.00   sec  1.43 GBytes  2.45 Gbits/sec                  (omitted)
[  4]   5.00-10.00  sec  1.43 GBytes  2.45 Gbits/sec                  (omitted)
[  4]   0.00-5.00   sec  1.44 GBytes  2.47 Gbits/sec
[  4]   5.00-10.00  sec  1.43 GBytes  2.45 Gbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bandwidth       Retr
[  4]   0.00-10.00  sec  2.87 GBytes  2.47 Gbits/sec    0             sender
[  4]   0.00-10.00  sec  2.86 GBytes  2.46 Gbits/sec                  receiver

iperf Done.

This is about as good a result as can be expected - 99% of the 2.5 Gbps link speed. The difference is mostly TCP and IP overhead. I then connected the machines to the 10 Gbps ports of a Netgear unmanaged GS110MX switch, using CAT6 cables. This was the result :
D:\Downloads\iperf-3.1.3-win64>iperf3 -c 192.168.1.237 -t 10 -O 10 -i 5 -R
Connecting to host 192.168.1.237, port 5201
Reverse mode, remote host 192.168.1.237 is sending
[  4] local 192.168.1.236 port 63175 connected to 192.168.1.237 port 5201
[ ID] Interval           Transfer     Bandwidth
[  4]   0.00-5.00   sec  1.28 GBytes  2.19 Gbits/sec                  (omitted)
[  4]   5.00-10.00  sec  1.29 GBytes  2.22 Gbits/sec                  (omitted)
[  4]   0.00-5.00   sec  1.29 GBytes  2.22 Gbits/sec
[  4]   5.00-10.00  sec  1.30 GBytes  2.23 Gbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bandwidth       Retr
[  4]   0.00-10.00  sec  2.59 GBytes  2.22 Gbits/sec    0             sender
[  4]   0.00-10.00  sec  2.59 GBytes  2.22 Gbits/sec                  receiver

iperf Done.
While this is slower than a direct connection, the speed is still much higher than 2 Gbps, and about 88% of the 2.5 Gbps link speed.

Finally, using the same CAT6 cables, I moved both systems to the Trendnet TEG-7080ES switch. This is the result of the performance test :
D:\Downloads\iperf-3.1.3-win64>iperf3 -c 192.168.1.237 -t 10 -O 10 -i 5 -R
Connecting to host 192.168.1.237, port 5201
Reverse mode, remote host 192.168.1.237 is sending
[  4] local 192.168.1.236 port 61997 connected to 192.168.1.237 port 5201
[ ID] Interval           Transfer     Bandwidth
[  4]   0.00-5.00   sec   325 MBytes   545 Mbits/sec                  (omitted)
[  4]   5.00-10.00  sec   312 MBytes   524 Mbits/sec                  (omitted)
[  4]   0.00-5.00   sec   325 MBytes   546 Mbits/sec
[  4]   5.00-10.00  sec   364 MBytes   610 Mbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bandwidth       Retr
[  4]   0.00-10.00  sec   689 MBytes   579 Mbits/sec  10231             sender
[  4]   0.00-10.00  sec   689 MBytes   578 Mbits/sec                  receiver

iperf Done.

As you can see, the card can't even deliver 1 Gbps in this case. We are very far from the link speed - only about 23% !

The Trendnet TEG-7080ES admin console shows the following :


The client with the Realtek 2.5 Gbps NIC is on port 4 . The server with the Aquantia AQN-107 is on port 5. I have tried setting the flow control setting on both. It doesn't fix the issue. Disabling jumbo frames on both sides and on the switch improves the performance to slightly over 1 Gbps, but still nowhere close to 2.5 Gbps. And it shouldn't be necessary to disable jumbo frames, since they work fine when the 2 systems are directly connected, or connected to the Netgear GS110MX switch. The Trendnet switch is running the latest firmware 1.01.04 dated 04/2018.

There is clearly some sort of incompatibility here between the Realtek 2.5GBASE-T and the Trendnet switch. I'm inclined to think this is a Trendnet issue, as the Realtek 2.5 Gbps NIC works fine against the Aquantia 10 Gbps NIC when directly connected, as well as against the unmanaged Netgear GS110MX switch, though it is still possible there is something wrong on the Realtek side too. I don't have any other multi-gig Ethernet controllers to test with besides this Realtek 2.5 Gbps and four Aquantia AQN-107 PCIe 10 Gbps.