Precision Fixed Workstations

[WizardOfBoz2]

I have a T7910 with dual 2699A v4 CPUs. 128Gb of Samsung DDR4 2400T ECC RDIMM 2Rx8 memory in 8 16Gb sticks. When I run the passmark performance test, the results are the 30th percentile. So I'm wondering 1) Can this be right? 2400 should be pretty zippy. Not like 3200, but I should be in the lower half of performance 2) If correct, does this memory <Profanity removed> Did I get the wrong stuff? 3) If correct, and the memory isn't bad, is there something I've set in BIOS that perhaps is limiting memory bandwidth? The CPU bandwidth, with 8 sticks of memory, screams. 48,000, in the 90th percentile. What gives with the memory limitation? Thanks.

[mazzinia]

Not going to help, but your post made me download and run passmark.

With a dual 4110  t7820, I get very, VERY poor memory results ( 1480 , with 4x8gb ram stock hynix , 14th percentile ).

Even the cpu is not the best submitted result vs the same cpu ( 16812 vs best being over 19k )

And the nvme ssd has very low IOPS compared to the same nvme unit stored in other benchmarks

[WizardOfBoz2]

I have the nvme in a cheapie PCI-E x4 card, but the diskmark was not horrible (70th percentile?  Can't recall and I dual boot and right now I'm in ubuntu). 

I just wonder if their something wrong in how I've set my BIOS/UEFI settings.  Or if the chipset firmware is wonky.  Dell Command Support indicates I'm up to date on all drivers and firmware, and I have the most recent firmware ( A34).  I have a hard time believing that this Dell is that bad wrt memory.

The passmark sw is nice for setup, but after a while I think one can go off the rails and spend more time trying to obtain that last little bit of benchmark than one could possibly save in actual useful work efficiency.  I was doing fine work (math simulation of disease, so not exactly light computation) on a Thinkpad laptop from several years ago. It's passmark is 7 times slower than the "new" 7910 I've put together.  Worked great, though.  

One interesting point: when my memory was just installed in slots 1 and 2 for each CPU (dual channel) the cpu passmark was much lower, about 10%. Adding in sticks in slots 3 and 4 (quad channel per CPU) was what boosted my passmark scores.   If the analogy holds, you'd see the max cpu passmark if you added another 4 sticks of 8Gb.

[MickKr]

Hi every one,

T5610 2x e5-2650 V2, 8* 16gb 1866Mhz ram, GTX 1080ti Asus Poseidon 11gb, Nvme disk in pcie slot Samsung Evo Pro 500gb as system disk.

Pleae post your results as screenshots. Wilk be much easier to comapre results

 

[mazzinia]

You make a good point, about dual vs quad channel. I didn't really think about it.

About the nvme I'm a bit baffled by the results. It sits in the front disc carrier with the kit and everything, and what passmark gives me is :

14225 disk mark

1974 sequential read

1145 sequential write

854 iops 32kqd20

28 iops 4kd1

Way on the low end compared to all the other samsung pm981 nvme recorded to compare with. Especially the 4kd1 is the lowest score ever recorded

This nvme unit has felt slow from the start, installing file intensive service packs ( windows )

[mazzinia]

Dual 4110 ( 2.1Ghz ) , 32gb ram (4x8) , 1TB Samsung nvme pcie 781 in flexbay 1 , <Profanity removed> quadro P600

[WizardOfBoz2]

mazzania:

>About the nvme I'm a bit baffled by the results. It sits in the front disc carrier 

It may be worth it to buy a $14 pci-e x4 card and install the nvme card directly interfacing the bus.  I think the adaption through the HDD/SDD bay uses the SAS/SATA interface which is, well, at normal SAS/SATA speeds.

But perhaps folks here can verify or disavow this.  

For me I had to remove all disks except the new nvme in the card slot (I think I turned off the RAID controller, too). Then I did a clean Win 10 install (using the Win 7 key on the case, thanks, Dell!), then I installed two sdds and an hdd in the bays and turned the RAID controller back on. (I also installed Ubuntu over the top of Windows so that now GRUB 2 allows me a choice).

As I said, my initial CPU passmark was around 43500 I think, with 4x16Gb.  I added another 4 sticks of identical memory and this boosted the cpumark to 48000.

I'm waiting for either a 2 6 pin to one eight pin PCI-E adaptor, or a Dell 8 pin adapter, to give me 2 8 pin cords to install an EVGA 2080 Ti card.  I also installed a 5.25 inch Blu Ray optical drive (with M-Disc). I should be in good shape to do actual work then (as opposed to mess around with a computer build - fun, but it doesn't pay the bills).

Still, that memory passmark bugs me.  What gives, Dell?

[mazzinia]

> It may be worth it to buy a $14 pci-e x4 card and install the nvme card directly interfacing the bus.  I think the adaption through the HDD/SDD bay uses the SAS/SATA interface which is, well, at normal SAS/SATA speeds.

no, it's interfaced directly to a pcie specific connector. In normal benchmarks I get 3400 MBs reads / 2900 or 3000 writes

but in those normal benchmarks, the iops feel a bit low

[WizardOfBoz2]

The plot thickens...

Purchased memory from ATech off ebay.  Supposedly new Samsung stuff. But look at the memory passmarks.  Very poor results!  So I pulled out my trusty CPU-Z.  Of the eight sticks, six listed Samsung, a model number, and a serial number.   2 showed up with no mfr, no model number, and no serial number. So I'm wondering if I've gotten a batch of seconds.  I wrote to ATech.  If they send me two sticks and they show up with mfr, model, and SNs listed and give me decent passmarks I make keep them.  Otherwise I think I'll just buy from another vendor.  Will advise on how ATech handles this.   

So again, if anyone knows of a very reliable brand of server memory with good specs, please let me know.  I'm looking for 16Gb DDR4 2400 PC4-19200 ECC Registered Dual rank (R2x8) sticks.

[mazzinia]

Side note : warning I would advise not to run Passmark with vmware workstation 15.5 running at the same time. It caused the lock of some resource linked to storage devices and caused a lot of troubles, including troubles to restart windows