Lenovo ThinkStation P620 Review: A Vehicle for Threadripper Pro
anandtech.com
anandtech.com
IIRC, these workstations are made for VFX houses where they'll probably render scenes nearly 24/7, weekends included, so Lenovo didn't fuck around with the cooling for such a demanding use case.
If you're on a crunch to finish that epic 30 second shot worth six-figures for that upcoming Hollywood blockbuster then you want the best pro-grade tools you can get that you can leave running for days and sleep soundly knowing they'll do their job and not hang up or throttle. Any major downtime in this racket is worth more than that workstation.
>If you read the enthusiast forums, it's something that most everyone makes fun of
True, because most "enthusiasts" on forums are just well-off gamers or posers showing off their expensive setups for upvotes, that are just used 2-6 hours a day for video games and scrolling through reddit instead of 24/7 workloads responsible for a business's revenue stream.
They render VFX on employee workstations? Why don't they render in a data centre with proper cooling and things? Must make the offices very hot!
https://www.wsj.com/video/lucasfilm-data-center-and-the-real...
Plus, clusters aren't always at your disposal and usually have schedules when you can book them according to priority and not all workflow can be automated and pipelined to fit a cluster. VFX work is not just 3D modelling where you click Render and boom, 2 days later, your scene it's ready to be put on film. There's tons of tasks you need to do pre- and post- production like layout, rigging, motion capture, rotoscoping, comping, etc. and most of these tasks are done manually on the artist's workstation.
I'm sure big houses like ILM, Pixar, Weta, Disney have the money and technical chops to engineer and maintain their own in-house rendering cluster infrastructure but small firms don't and rely on beefy workstations like this to render their project over multiple days.
EDIT: Oops, leave a tab open for a few hours and come back to find your reply's been scooped. I'll leave this up and add some more links for anyone curious about cloud rendering though[1][2], it's an interesting space.
[0] https://press.aboutamazon.com/news-releases/news-release-det...
[1] https://utopiancorps.com/cloud-rendering/
[2] https://www.awsthinkbox.com/studio-in-the-cloud/cloud-render...
I'm actually surprised how incremental the progress was over the last decade, with regards to CPUs and to a lesser extent GPUs. I'd strongly suggest trying to reuse older hardware instead of spending $1000+ on new one
I think you're a bit over-exaggerating here as it really depends on what you do with your PC and where you're coming from.
If you've only been running Intel this whole time and all you do is light workloads then you won't notice much progress since they haven't made much, but if do CPU demanding tasks and were to try a Ryzen 3xxx/4xxx, or better yet, a 5xxx, you'll notice a huge leap forward (if you could find one in stock, that is).
However, your "new" 7700 is a bit old in the tooth now and, despite Intel's lack of major progress, it will still be left in the dust by a newer 10th/11th gen chip.
My old faithful Intel Q6600 still serves my parents just fine after 14 years of age, but I'm not gonna pretend it isn't slow by today's standards on the modern bloated web and with all the electron crap.
You'd expect as the current console generation actually gets off the ground to start seeing more demanding upper settings, and eventually more demanding minimums.
sort of an odd thing to say when you just bought a machine with the last i7 desktop part to have only four cores. i7-7700 is only about 20% faster than an i7-2700k, but the 9700 and 10700 parts are way faster in any multi-threaded scenario.
I kept the 3945WX (and the rest of the base spec), added a W5700 and a second 10GbE card, then installed an AX200/BT5.1 card, second SSD, and three more DIMMs (64GB total) when it got here.
Outstanding machine for the price if you can get it on discount!
My home PC has very little to do with my day job at this time, so I acknowledge it's complete overkill for that. However, I'm looking at needing to transition careers in a couple of years and so this will likely be used as a homelab in a box. I also game, dabble in coding (playing with a game idea in Unity right now), etc.
From my limited perspective, the 2 benefits of TR over Ryzen at low CPU count are the PCIe lanes and memory channels and slots. If you need a high CPU count then there is no other choice but TR.
I chose this system for exactly the reasons you listed at "low" (is 12 low now?) core counts: PCIe lanes and memory bandwidth.
Also, it's just awesome.
I can see if I can scare up some UDIMMs and test, but I don’t think I have any DDR4 around.
This is the one thing I hate about Lenovo. Just show me the price , creating MEGA sales to try and entice me into buying isn't cool
For this price from HP you get the same GPU, 56 cores, 192GB of memory, and 16 TB of SSDs with 6 m.2 slots left open, and thunderbolt which the Lenovo lacks.
Have they got PCIe Gen 4 yet?
Is there any benefit compared to building from individually purchased components?
I wonder what kind of deal made these 5 months of Lenovo exclusivity worthwhile for AMD.
> The vendor, when asked about on-shelf availability, said that they don't particularly plan on doing a full retail packaging, but should AMD push the TR Pro CPUs into the consumer channel, they will prepare the appropriate boxing. In that light, as it stands, it looks like it will depend on AMD to decide if they want to enable the regular consumer retail channel.
https://www.anandtech.com/show/16396/the-amd-wrx80-chipset-a...
If you need it, you need it, and there's no competition in that space, so it's expensive. But, if you aren't doing whatever GV100s are good at (machine learning? raytracing?) then you can probably get a C++ compilation rig and web browser with 64 cores for ~$10,000. (I suppose that's another "if you need it, you need it" case. But at least more straightforward to the software engineers among us -- if you have more than 64 C++ files to compile... the 64 core chip can be useful. Even "go test ./..." in a moderately-sized project uses all 32 of my non-pro threadripper cores. Briefly. Not doing as much work as the C++ compiler. :)
I also bought cheap “eBay memory” instead of the more expensive Lenovo offering and had to reseat DIMMs a few times to stop having ECC errors (the memory scrubbing feature is nice).