$$$ Climbing - Feb $400, April $500, June $600. I took the leap, 32GB V100 SXM2 for LLMs (#RACERRRZ)
Everything is moving cloud-ward - PC hardware prices keep on creeping up – and that is likely to continue for as long as AI data centres are in demand. That’s where it is a rare discovery when something of value comes along that has not yet skyrocketed – the end-of-life Tesla Volta V100 SXM2 AI accelerator that is. In saying that, the card in question has seen a stable price climb (~$350 in 2025 to $600 present day). That brings us back to the problem we are trying to solve. We want to experiment with AI compute in the home lab – but with GPU prices skyrocketing – it really closes up opportunities to get “smart” LLMs loaded into memory. AI accelerators bring in the edge – they were designed around LLM inference first – not frames per second. I am still searching for the answer to "what AI actually has on offer". Inspired by Tony Stark's J.A.R.V.I.S. – I took it upon myself to "vibe-code" my version of this. This ‘character’, A.I.D.E.N., was forged out of a desire to make AI actually useful - with its 'final form' being so much more than a ‘ChatBot’. You’ll see A.I.D.E.N. make his presence known whenever relevant. Responses are his own – but he remains a glorified LLM. --- Hardware: So, how do you add cost effective AI compute to your home lab? You could try to run DGX server hardware, specifically a Telsa V100 SXM2 (Server PCI Express Module 2). So how do you connect your SXM2 GPU to your workstation? A SXM2 to PCIe adapter of course! I was tempted for the 32GB AMD Radeon Mi50's (half the price of the 32GB V100), but they are less well adapted to AI workflows and only function in Linux (not that that is a bad thing - but it does present additional complications). For now - I have assembled the PCIe adapter - applied Thermal Grizzly Kryonaut (32GB #1) and Duronaut (32GB #2; future video!) to the core (with the world's best thermal paste application method, sure to bring a smile). I then stacked thermally conductive silicone for effective contact between the power chokes and mosfets with the copper heatsink. All that is left is for us to push the limits of AI inference. --- Universal LLM Benchmark Version 1.0: Paste this universal prompt into your favourite LLM to test its intelligence. Since we use the same prompt, you can compare your results with the ones I published – allowing comparison across systems! https://chatgpt.com/share/69b8a659-8e... --- V100 Data summary: 16GB V100 SXM2 to PCIe vs RTX 3090 Ti: https://racerrrz.com/wp-content/uploa... 32GB V100 SXM2 to PCIe vs RTX 3090 Ti: https://racerrrz.com/wp-content/uploa... 32GB V100 SXM2 to PCIe vs RTX 3090 Ti Summary: https://racerrrz.com/wp-content/uploa... --- Related posts I have made: Day 1 - 16GB V100 SXM2 Unlocked: • Post Video 1 - 16GB V100: • Post Thermals - • Post Day 30 - 32GB V100 SXM2 Unlocked: • Post ComfyUI: • Post Video 2: 32GB V100 runs A.I.D.E.N: • I Build My Own AI - It Called Itself A.I.D... Take 1 or do you need 2? • Post Video 3: 32GB V100 fitment - you are here. Future content - TBC --- 🛠 Amazon / Ebay / AliExpress Affiliate Links (Please note- links will direct you to Amazon.com or Ebay.com or AliExpress.com. I will receive a commission if a purchase is made through these links. Thank you.) Tesla V100 SXM2 16GB: https://ebay.us/neSXwl https://s.click.aliexpress.com/e/_c4N... Tesla V100 SXM2 32GB: https://ebay.us/MIIWkZ https://s.click.aliexpress.com/e/_c4N... Tesla V100 SXM2 32GB (one I just bought): https://ebay.us/PRk7AZ Fan Cooled SXM2 to PCIe Adapter: https://s.click.aliexpress.com/e/_c3k... Fan Cooled SXM2 to PCIe Adapter w/backplane: https://ebay.us/pEqdb2 Thermal Pad: https://amzn.to/4qZnvtv Thermal Grizzly Kryonaut: https://amzn.to/45Qqlsa 6-pin GPU to 8-pin GPU Power: https://amzn.to/4qvzo9y ------ For more updates and detail on the build: 💻 / discord 💻 HSPEC DataBase (for parts, upgrades, video summaries and more): / hspecworkstations 💻 Discord Server link (Name: HSPEC Computing): / discord 🎯 Support: https://buymeacoffee.com/RACERRRZ #NVIDIA, #V100, #localLLMs, #AI, #ArtificialIntelligence, #MachineLearning, #LLMs, #GenerativeAI, #Computereducation, #Computer, #Gaming, #Gamingcomputer, #Homeserver, #zworkstation, #RACERRRZ

HW News - DRAM Companies Hit Trillions of Dollars, Bambu Open Source, NVIDIA Spark Concerns

I Tested the Cheapest Path to 96GB of VRAM

NVIDIA RTX Spark Explained {Computer Wednesday}

Is the Nvidia Tesla V100 still good for AI? - Inspur DGX V100 vs RTX 5090

I Bought the Trump Phone

Tesla V100: главная ошибка при покупке.

Samsung's 990 Pro SSD warranty policy is a scam; I'm taking them to court.

Android 17 sucks. So I put Linux on a phone.

Three months wrong about why my 4-node AMD cluster was slow

Once You Understand it, You Will Think Everything Else is Silly - Toyota E-CVT

Why Building AI Data Centres Isn’t Working Anymore

Microsoft Just Released Their Own Linux Distro: Should You Be Worried?

How do you cool an nVidia Tesla GPU?

This Ridiculous $200 AI GPU Shouldn’t Be This Good

I Went to COMPUTEX 2026. Here’s What Actually Stood Out

The Local AI Hardware Mistake Everyone Makes

This is why Monitor Specs are complete LIES!

Cooling my Gaming PC with No Fans

I Build My Own AI - It Called Itself A.I.D.E.N. (32GB V100 SXM2 + HP Z8 G4 #RACERRRZ)

