tictoc

If TRX50 happens, I will be upgrading my 3960X. 

I can't really justify the Threadripper Pro SKUs, but I would really like to upgrade my workstation.

Hopefully this time around there will be at least two CPU generations.  I was pretty salty when AMD killed the HEDT platform, and only released the 5xxx CPUs on WRX80.

https://videocardz.com/newz/amd-to-launch-threadripper-7000-pro-and-non-pro-zen4-cpus-designed-for-wrx90-and-trx50-motherboards

On 06/09/2023 at 06:29, Northwave said:

Hey !

Im currently rearranging my server in the same case. Would love to see a photo taken from the top, to see how much space is behind the HDDs. Thank you ! I will print the hdd holder with 3d printer and to see this in before, would help me a lot.

Sorry for the late reply, I've been out of town for the last couple of weeks.  Here's a shot from the top with the fans removed.

On 09/08/2023 at 19:00, ENTERPRISE said:

That is a nice build for the purpose, case looks like an awesome choice too. Out of interest, is this not going to be on 24/7 ? Only because I noted you saying you would use WOL.

 

My fileserver is on all the time but it runs more services than just file access. 

 

I know back in the day it was less wearing on HDD's to leave them on rather than powering down/up all the time. Im not sure how 'Hardy' HDD's are these days in that respect, haven't kept up with the data om that.

This is probably still true, if for no other reason than thermal cycling electronic components pretty much always hastens their eventual death.  How much it really matters is up for debate.  Exos drives are rated for 600k load/unload cycles and 50k start/stop cycles, so starting up and shutting down once a day wouldn't really put a dent in the rated lifespan.  

This will likely not be the way I run this.  I was just fiddling around with further reducing power consumption.  The reduced power consumption is not really worth it to me, with the increased chance of a failed backup due to a start-up/shut-down or WOL packet error.

With zero tuning the system idles at 78W.  More than half of that power is the drives, nic, and HBA.      

Storage will be updated fairly soon, since I will be retiring a few 16TB drives out of my main server.

Server will be headless running Arch Linux, and will be mostly asleep using WOL to wake the server idle for backup tasks. 

This build is a smallish backup file server that will be the final piece of kit downsizing from my 42U into a 12U rack and two desktop cases.

There are a number of similar sized or smaller NAS type chassis available, and the main reason for this build was the price of the case ($59 on NewEgg at purchase) and the fact that I thought I could shove 6 HDDs with little to no modding.  The case, motherboard, and CPU are the only parts I purchased for this build, and the rest of the parts are repurposed from other machines.

Hardware:

Case: Sama IM01

CPU: AMD Ryzen 3 Pro 5350GE

Motherboard: AsRock B550M-ITX/ac

RAM: 2x 16GB Micron DDR4 3200 ECC UDIMM

NIC: No name AQC113 based 10GBASE-T

HBA: LSI SAS 9207-8i

SSD: Inland Premium 512GB NVMe

HDD: 4x 16TB Seagate Exos X16, 1x 8TB Seagate Ironwolf Pro, 1x 8TB Seagate Exos 7E8

PSU: Corsair SF750

1 hour ago, Barefooter said:

Wow that is some serious storage there!

 

I wonder how much that bad boy is going to cost?

The 30 TB model is listed at $2300 USD. https://www.mouser.com/ProductDetail/Solidigm/SBFPFWBV307TOF1?qs=1Kr7Jg1SGW%2BNbgRRG1pe0A%3D%3D&_gl=1*20h1q4*_ga*dW5kZWZpbmVk*_ga_15W4STQT4T*dW5kZWZpbmVk*_ga_1KQLCYKRX3*dW5kZWZpbmVk

4 hours ago, Memmento Mori said:

I like the military look of the case.. But idk - i probably got bewitched by fractal design and the Torrent model ...  cant help my self...

 

hmmm that way to small as i want to reuse my Thermalright cooler... and i need a CPU cooler height of 170mm  

 

But hey im just getting in to the whole topic  But its very interesting, so i take m time and just gathering info i can.. If any other advice, please just share it  

 

Thank you very much.

It is a pretty small case.  I was just using it as refernece to show that if you are going to do a bit of custom work, most any desktop case will work just fine.

If you go with the torrent you could get some of these rubber stackers: https://www.amazon.com/SEDNA-Hard-Disk-Rubber-Stand/dp/B07XYBKGWN/ref=sr_1_3?crid=20N040NWGAQVW&keywords=hard+drive+stacker&qid=1690330434&sprefix=hard+drive+stacker%2Caps%2C124&sr=8-3#customerReviews

Those require a bit of additional work if you are going to stack a bunch of drives, but in a situation like the torrent, you can install the drives with the cabling facing the motherboard tray.  That will make for a nice and clean look, allow the lower front intake fan to cool the drives, and dampen the vibrations from the mechanical drives.

20 minutes ago, Cerberus said:

WOO! An upgrade from my 1080ti! Thanks @tictoc!

Just made the team switch from my old OCN account.

Thinking about dusting off some Windows installs and subbing some benchies for the team.

I'll take a look through the hardware pile and see what eligible hardware I have.    

How many drives slots do you need/want, 2.5" or 3.5"? 

It looks like the torrent compact is wide enough that you could stack hdd cages up in the the front of the case. You would just need to make sure that whatever cages you use/print allow for airflow from the side.  I would definitely suggest using rubber dampers at all attachment points, but other than that, if you are willing to do a bit of modding you can stuff quite a few drives into most cases.

My backup server is about to be transplanted into a 22L case (about half the volume of the Torrent Compact), and that machine will have 6x16TB HDDs, mitx board, 10Gb nic, and an HBA.

A little late on the start, but at least one GPU will be folding for the FaT.

--EDIT--

Fired up F@H on the outgoing 2080 Super one last time.

+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.86.05              Driver Version: 535.86.05    CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce RTX 2080 ...    On  | 00000000:09:00.0  On |                  N/A |
|  0%   52C    P2             203W / 292W |    547MiB /  8192MiB |     97%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
                                                                                         
+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|    0   N/A  N/A       638      G   /usr/lib/Xorg                                32MiB |
|    0   N/A  N/A       761      C   ...it/22-0.0.20/Core_22.fah/FahCore_22      510MiB |
+---------------------------------------------------------------------------------------+

I did ease back the clocks a bit for its last hurrah on my bench.

nvidia-settings -c :0 -a '[gpu:0]/GPUGraphicsClockOffset[4]=-120'

Life got in the way over the last many months, and I haven't put this router in to production.  I am currently just running off of my old EdgeRouter, but I will be putting this build online in the next month or so...

Here are some final pics with everything installed and wired up. 

• PiKVM tested and fully functional with the ability to power on/off and access the UEFI remotely.
• Hardware power and reset buttons are installed behind the front mesh, to the right of the NICs, just to make sure I can't accidentally hit the buttons if I am working on the rack.

I am still leaning towards VyOS, but I might ultimately go with OPNSense.  I'll post some final pics after it is up and running in the rack.

The final piece to complete my cool and quiet homelab infrastructure rebuild.  This case is for my backup server, which will have an ASRock B550M-ITX/ac with an AMD Ryzen 3 Pro 5350GE, 32GB DDR4 ECC UDIMM, 6x 16TB HDDs.  Pretty imporessive little case for $53.

SAMA-IM01

12 minutes ago, damric said:

 EVGA RTX 2080 SUPER BLACK GAMING

In Q_Cruncher 

Out on prizes

I have a few things up and running, so...

IN Q_Cruncher

You can pull me out of the prize draw.

After battling with some high temps on the HBA, optane drives, and the X550-T2, I finally swapped the fans around on the radiators.  Previously the fans were running as intake, and now I have them running as exhaust. 

In the original configuration, I was never able to get enough front->back airflow to exhaust all the heat that the radiators were dumping into the case.  Results from the swap in fan direction are pretty impressive. Temps dropped by 13°C-20°C on the hottest components, and I was also able to drop the speed of the 180mm intake fan by 200rpm.   

I did new pads on my 6900XT, but it also got a full cover HeatKiller block at the same time.  The pads I used were the same pads I've been using for a very long time on all my cards, which are the Fujipoly ultra extreme.  I am just always on the look out for deals, since they are expensive pads.

The only card where I saw some major temp drops was with a bunch of highly OC'd 7970's.  The workload they were running was basically FurMark on the VRMs. I was able to lower VRM temps by more than 20°C after swapping in the higher quality pads.

Added a few GPUs to the mix.  They are not great for F@H, but it will bump up the ppd by 4-5M.      

6 minutes ago, mx500torid said:

Is anyone folding with a 6800 or 6900 radeon? Seems my ppd has dropped 2 million from the older driver.

 

I am folding on a 6900 XT, but not in Windows. For me there is pretty huge range in ppd on various WUs. I see anywhere from 2.7M-6.4M ppd depending on the project that is running.

*Edit* Currently folding at 2875core|2150mem

1 hour ago, neurotix said:

Tictoc, what command line arguments do you pass to stressapptest to check your systems? There's a lot and I'm not sure what to use, what size, etc

Pretty much what Pook posted, although I forgo the --pause-delay, since I am building from source and that workaround has been fixed in the upstream version.

*Edit* Depending on how you're running it, you might need/want to allocate a different amount of memory, but what Pook posted (leaving 3GB available for other processes) should be more or less universally good unless you're doing some other crazy memory intensive things while running stressapptest.

10 minutes ago, neurotix said:

I totally forgot about stressapptest! I haven't heard that in like 10 years. Yeah, I'll try that, thank you. How long should I run it?

Depending on how close to stable your memory is, it will usually error out in less than 30 minutes, but I usually give it an hour.  If that checks out then I'll usually run through a few other stress tests like mprime (Prime95) large FFTs and Blend.

In Q_Cruncher.  Out on prizes.

20 minutes ago, The Pook said:

 

I stopped using MemTest all together, I've ran it overnight and passed and then failed in GSAT within an hour. GSAT (or TM5) are what most people are using nowadays anyway and you can use either within the OS. 

Longtime stressapptest user.  It almost always uncovers stability issues quicker and more reliably than memtest. 

Additionally, there are some other things you can do to hammer the memory and cache to look for instability.  Feeding pbzip2 a large (100GB+) mix of compressible and incompressible data (text files, pdfs, jpgs, etc.), has also helped me find edge case instability when dealing with large amounts of fast memory on DDR4 platforms.  I haven't played around with DDR5 yet, but the same principles should hold true. 

Vulkan compute on this bench is overall substantially better than Geekbench's OpenCL on Linux.  The Linux OpenCL score is quite a bit slower than Windows, and I imagine my 6900xt running 2950core/2150mem is clocked higher than  the vast majority of 6900xt submissions.

Here's an OpenCL compute run for comparison.