Much excitement as I received the AMD EPYC 9354 CPUs for my HL15 builds today and seeing at least some folks start to receive their HL15s. On the edge of my seat waiting for my three to show up so that I can commence with system builds. Woot.
I image that everyone here has been in this mad rush to buy whatever supporting hardware they want for their home-lab builds lol. I am also sitting on a bunch of hardware waiting for my HL15 to get to me as well. I am VERY excited.
Building out three identical HL15 using the EPYCs pictured. Might do a fourth system in Q1 2024 based on AmpereOne but waiting for an ATX-sized mainboard for those to come out.
Use case? Mostly just tinkering. I’ll be migrating everything to these and shutting down everything running today from Haswell/Broadwell (majority of systems) and Skylake or Cascade Lake (minority of systems). Also retiring a few JBODs and some older drives between 6TB - 10TB in size.
Each system will have a good amount of raw storage: ~30TB NVMe, 300TB 7.2K SATA across 15x20TB internally plus 150-240TB 7.2K SATA (reusing some older 12-16TB drives) connected via a single Lenovo SA120 JBOD attached to each system. One system will have a LTO-8 drive for backups.
Outside container-based workloads, I tend to do all my photo/video editing over the network and the systems have 100GbE NICs so there’s no 1GbE bottleneck on the NVMe drives or even the SATA-based arrays. I’m expecting scrubbing through 2160p or even 4320p footage will be liquid smooth.
Doing the same MB and CPU. Throwing vSphere 8 on 4 of them (eventually). Going to do NSX, vSAN, Tanzu, and the vRealize suite. Initially going to have a TrueNAS VM on one of them with a storage controller passed through. May eventually do a second one so I can do TrueNAS Scale in HA. Make patching my homelab a bit smoother. Already have Nvidia ConnectX-5 VPI cards and some 25gbit switches from the current build.
I’ve gotten as far as deploying a Tanzu cluster so I can’t say what it’s like in operation currently. Current homelab setup was built with noise as the paramount concern and, as a result, is extremely RAM limited. vSAN has about a 22GB RAM per VM host consumption in my current config, NSX was pulling somewhere between 15GB and 20GB per VM host, and my VM hosts max out at 64GB. Didn’t leave a lot for running workloads.
In theory; having vSphere, NSX, vSAN, Tanzu, and vRealize all tightly integrated with one another should give some awesome benefits. I live in the VMware world (fortunately or not, thanks broadcom) for work and have dipped my toes (legs, lower half of my body, up to my eyeballs, etc) into other products (Nutanix and Hyper-V specifically, but pretty much everything else as well). Pricing and licensing suck with VMware but nothing can hold a candle to their product stack. VMUG for the homelab is helpful though, $200 a year for licenses to just about everything they make that anyone might want to run at home.
Some day I’ll give AMD another shot. I’ve not owned anything non Intel or Nvidia since circa 2008, and that was a terrible little cheap laptop with an old Athlon that doubled as a coffee warmer.
I am currently on the hunt for a small laptop (preferably a convertible 2-in-1 type) and as it will run some flavor of Linux, I’m looking to branch out of Intel/Nvidia for this one.
Edit: Saw the notes of Tanzu…
I just finished up an NSX training class, Tanzu was a heavy focus. I think it’s really only best if you’re fully enveloped in the VMware ecosystem. NSX has some REALLY cool stuff around load balancing for containers, firewall, IDS/IPS, network segmentation (within a single subnet!).