Adding this post a bit later than normal here, but alongside my RSA Factorisation project using Python, I decided to purchase a computing GPU to use alongside the build. I’ve always been wanting to use a computing GPU to leverage the faster parallel processing power using CUDA cores, in either Machine Learning projects or Python projects where I could import modules and libraries.
Not that I know how to use these GPU’s however. I mainly came across it when purchasing the RX580 GPU for my current computer. I saw an RX580 on eBay for £40 and decided to purchase it. I have a few games on my PC that I’ve been unable to play for many years for many reasons. One of them is that my home computer is underspecced and a load of rubbish. However the GPU on the computer is also a pile of poop. My existing home computer is an Nvidia GT240 with a whopping 1GB of RAM on it.
Whilst I don’t game much anymore on the computer. It’s incredibly frustrating when I have s library full of steam games that I’m unable to play because the GPU is an inferior piece of poop.
Anyway, I purchased an RX580 because it was a cheap enough card, it fits in my case (just) and my existing CPU supports it. Plus I could also play some games now in a good FPS speed, one of them being Frostpunk which I’ve always wanted to play. Whilst I could leverage steam gaming using my internet through Youfibre as it’s fast enough. The problem with the online gaming is you’re limited to slots of 1 hour. So I wanted something locally. Anyhoo. I purchased the GPU and had around £60 left over, and was eyeing up a computing GPU.
So here I am. I decided to purchase the Tesla K80 to get my teeth wet with some projects. It has 24GB of GDDR6 memory, so great for smaller projects and over 4000 CUDA cores for parallen computing. Not that I actually know how to use accelerated computing GPU, or code in CUDA, Python or anything else, but it’s a learning game and my PC should at least support it because it has an 16x slot and fits into my computing. I already had a GPU so it was just a case of configuring it right? Wrong… so..so..wrong
It’s a big long winded why my current motherboard won’t support using the Tesla K80 GPU. Well maybe it can, but I couldn’t be bothered to look. There’s a long winded post on reddit about this, but basically it’s to do with accessing more than 4GB of memory. Unless you enable this option it’s limited to 4GB and doesn’t utilise the full amount which doesn’t unlock the full potential of the GPU
So rather than upgrade the BIOS on my existing motherboard to support to support the computing GPU. I wanted to get a new one (plus my case didn’t have a lot of room to fit it).
As there are various other motherboards that I could find on eBay that could do this (I had an excuse to spend some money on a project). It’s not just about buying a PC with a slot though, you need a computer motherboard that supports above 4G decoding (and also resizable BAR support if possible). Most people get the B550M motherboards. One of the main problems which you really need to consider are the available PCI-express lanes for the CPU. Basically to run a modern computer with a 16X computing card, you need in the region of 36 PCI express lanes minimum.
16 PCI-express lanes for the CPU
16 PCI-express lanes for the GPU
4 PCI-express lanes for an NVME drive
So I started looking on eBay to find a motherboard that supported this, I’d buy a Sage X99 motherboard next if I had the money. They’re absolute monsters, they support above 4G decoding. They’re designed to be monsters though, up to 2TB of memory. Latest threadripper GPU supported, 7 PCI-express slots for multiple GPU. 4 x PCI-e 5.0 NVME slots, with 2 x 10GB NIC’s and a dedicated management NIC with onboard VGA. I can drool.
Still. I’ll find one to use…to be continued….
K80 Accelerator, updated pictures
Some further pictures added to the post on this. I received the GPU cooler from China and fitted it onto the K80 Accelerator card. Was a bit of a headache at first because the fan wouldn’t fit slightly into the housing.
I went through disassembling the entire card (initially thinking front panel). However you have to remove the rear cover of the card and the fans inside, then the cooler slots inside it.
You then secure it together using the screws supplied. I found the screws didn’t align fully on the accelerator card housing however it was secure enough to support it.
Power testing the blower unit is quiet, I powered it with my bench tester and power consumption was 0.22A @ 12V of power. So the fan was very low power consumption.
The final few things I noticed on the cooler when disassembling was a RAM chip was damaged. I don’t know until testing if this will affect the card, so I’ll have to wait to find out when it comes to replacing it if needed. Given that theres a RAM shortage, this will be even more annoying.
The second thing I discovered, is that the K80 Accelerator card is powered by an EPS12V power socket (what is used for the CPU) instead of a PCI-express power socket. So I’ve ordered an adapter on eBay to power it with.
Just waiting for it now then I can power up my system
Disclaimer: This page contains affiliate links. If you choose to make a purchase after clicking a link, I may receive a commission at no additional cost to you. Thank you for your support!
