old X58 triple channel board... memtest86 memory speed?

[eccerr0r]

Anyone run memory speed tests with an old X58-based triple channel 1st gen or any other >2 channel board? The one I have is a Gigabyte I think.

Anyway I had four 4GB DDR3-1333 modules in the 6 memory slots and got around 11GB/sec memory speeds assuming that it went dual channel since it didn't have enough modules for triple channel. Now I stuck in two more 4GB modules, but they are DDR3-1600. I'm still getting 11GB/sec speeds according to memtest86...

Just wondering, anyone do memory speed analysis on these triple (and higher) channel memory boards? Are there any indications when it's actually using three or more channels and are single thread applications taking advantage of it?

I only have one known working dual channel board with all DDR3-1600 modules reporting 20GB/sec in memtest86. My triple channel (6x4GB modules) and quad channel (8x8GB modules) boards are getting crappy results around 11-14GB/sec, though it may be due to having mixed brand DDR3-1333 and DDR3-1600 modules but they all have the same size and ranks as far as I know.

[dmill89]

I just dug out my x58 system and got similar results 11.6 GB/s with 6 matched 4GB sticks of Timetec (with Hynix chips) DDR3-1333.

Specs as follows:
Motherboard: Gigabyte GA-X58A-UD3R
CPU: Intel Xeon W3680
RAM: 24GB (6x4GB) Timetec DDR3-1333.
GPU: Nvidia (EVGA) Geforce GTX1060
SSD: Samsung 870 Evo 250 GB



I also have a P55 system (so same era, but dual-channel) and it does even lower at 9.52 GB/s with 2 4GB sticks of Hynix DDR3-1333.

Specs as follows:
Motherboard: Gigabyte GA-H55M-UD2H
CPU: Intel I5-661
RAM: 8GB (2X4GB) Hynix DDR3-1333
GPU: Nvidia (MSI) Geforce GTX 770
HDD: Western Digital Veloci-Raptor 10K RPM 500GB


As well as a P45 board (so a generation older) that uses DDR3 and it only gets 3.89 GB/s with 2 4GB sticks of Corsair DDR3-1333

Specs as follows:
Motherboard: Gigabyte EP45C-UD3R
CPU: Intel Core2Quad Q9450
RAM: 8GB (2X4GB) Corsair DDR3-1333
GPU: AMD (Sapphire) Radeon R9-380
SSD: Samsung PM871 256GB



For some newer DDR3 systems here's a quad channel X79 system getting 13.9 GB/s with 4 8GB sticks of Corsair DDR3-1600

Specs as follows:
Motherboard: Asus X79 Sabertooth
CPU: Intel Xeon E5-2690 V2
RAM: 32 GB (4X8GB)Corsair DDR3-1600
GPU: AMD (Gigabyte) Radeon RX470
SSD: Imation X-Pro 256 GB



And a dual-channel Z97 system getting 19.1 GB/s with 2 8GB sticks of Crucial DDR3-1600

Specs as follows:
Motherboard: Gigabyte Z97X-Gaming 3
CPU: Intel I7-4790
RAM: 16GB (2X 8GB)
GPU: Nvidia (Gigabyte) Geforce GTX-1080
SSD: Western Digital Blue SN570 1TB



So I believe your results were probably typical for an X58 board, and it looks like on these older boards the CPU/Chipset were probably more of a limitation than the memory itself.

[eccerr0r]

Thanks for testing! Nice...

Looks like either the triple/quad channel is either a lie, or still just memtest86 doesn't test it properly, but my machines are working as expected.

Yeah I was kind of shocked my dual channel Z68 completely rips my X58 and X79 boards. It's ripping my DDR4 laptop too, so something just didn't seem right.

Incidentally my 4x2GB = 8GB DDR2-800 Core2 Quads also get around 4.4GB/sec ... wonder if it's just FSB limitation...

[lti]

An old i3-2330M is a little faster, but this might not be an accurate comparison since Lubuntu came with a different Memtest86+ version.


I'm a little surprised that I got HDMI capture working, but Windows forgot how to open .jpg files.

[eccerr0r]

Yeah I get ridiculously fast speeds with IMC dual channel 2nd gen (like my Z68, 20GB/sec), but the triple, quad, and limited single channel (my i3 with differing modules) are slower (11, 14, 10GB/sec respectively). Something still fishy going on...

[dmill89]

Yeah I get ridiculously fast speeds with IMC dual channel 2nd gen (like my Z68, 20GB/sec), but the triple, quad, and limited single channel (my i3 with differing modules) are slower (11, 14, 10GB/sec respectively). Something still fishy going on...

For curiosity sake I spun up Memtest 86+ on some DDR4 systems (where you'd expect the speeds to be much higher than DDR3) and the results are definitely a bit "screwy".


First a quad-channel X99 system with 4 8GB sticks of Adata (Hynix chips) DDR4-2400 is showing 12.6 GB/s

Specs as follows:
Motherboard: Gigabyte GA-X99-UD5 WIFI
CPU: Intel Xeon E5-2690 V4
RAM: 32GB (4X 8GB)Adata DDR4-2400
GPU: Nvidia Geforce GTX-1070
SSD: Hynix BC901 256 GB


A dual-channel 6th gen (skylake) Intel system with 2 8GB sticks of Kingston DDR4-2400 showing 18.6 GB/s

Motherboard: Lenovo ThinkStation P310
CPU: Intel Xeon E3-1230 V5
RAM: 16GB (2X 8GB) Kingston DDR4-2667
GPU: Nvidia (PNY) Geforce GT-1030
SSD: Hynix PC711 256 GB


And my most powerful system (which should be absolutely wiping the floor with everything else) a dual-channel AMD X570 system with 4 32GB sticks of Team-Group DDR4-3600 showing 17.3 GB/s

Motherboard: MSI X570 Gaming Plus
CPU: AMD Ryzen 9 5900XT
RAM: 128GB (4X 32GB) Team Group DDR4-3600
GPU: Nvidia (PNY) Geforce RTX-4070 Ti
SSD: Western Digital SN850 2 TB

And User Benchmark is showing 52.7 GB/s on this same system:





So something seems verry "off" with how Memtest 86+ is determining memory speed.

[eccerr0r]

No Windows here so can't run...
For Linux however, tried hdparm -T ...
This resulted in different results. I haven't tried the X58 yet, but my

C2Q with Q45 6.5GB/sec (significantly faster than memtest86+, near 50% faster)
i3 with H81 4.3GB/sec (possibly tainted result as running something in background, much slower than memtest86+, near 50% slower)
i7 with Z68 14.2GB/sec (slower than memtest86+ but in ballpark)

[dmill89]

I just dug out my x58 system and got similar results 11.6 GB/s with 6 matched 4GB sticks of Timetec (with Hynix chips) DDR3-1333.

Specs as follows:
Motherboard: Gigabyte GA-X58A-UD3R
CPU: Intel Xeon W3680
RAM: 24GB (6x4GB) Timetec DDR3-1333.
GPU: Nvidia (EVGA) Geforce GTX1060
SSD: Samsung 870 Evo 250 GB

Here's what passmark (running windows 7 on this system, so my options are limited on compatible "recent" benchmarking tools) has to say about that X58 system for camparison:

It is showing 19.6 GB/s threaded, 13.1 GB/s cached read, 8.8 Gb/s uncached read, and 8.5 GB/s write. (though it is showing the memory speed as 800 Mhz, but the BIOS shows it as 1333 Mhz 🤔)

[eccerr0r]

Hmm. Just tried hdparm -T (aka cached buffer read speed) on my X58A ... got 10GB/sec which is close to the memtest86 speed. Machine is running a RAID5 mechanical HDD so this sure ain't disk read speed...

# hdparm -Tt /dev/md127 /dev/sda

/dev/md127:
Timing cached reads: 20794 MB in 2.00 seconds = 10410.97 MB/sec
Timing buffered disk reads: 856 MB in 3.00 seconds = 285.33 MB/sec

/dev/sda:
Timing cached reads: 20366 MB in 2.00 seconds = 10196.04 MB/sec
Timing buffered disk reads: 232 MB in 3.00 seconds = 77.33 MB/sec

[Maxpower3]

Here's what passmark (running windows 7 on this system, so my options are limited on compatible "recent" benchmarking tools) has to say about that X58 system for camparison:

It is showing 19.6 GB/s threaded, 13.1 GB/s cached read, 8.8 Gb/s uncached read, and 8.5 GB/s write. (though it is showing the memory speed as 800 Mhz, but the BIOS shows it as 1333 Mhz 🤔)

I looked at your posts and was interested in doing some tests too.
Well, I have a slot that's no longer working; I threw a bit of random RAM into it.
After enabling NUMA in the BIOS, performance doubled.


There's dual and single rank. I'll try to find a complete identical batch to retest.