Meh, these old systems can be fun to tinker with but are _slow_.
E5-2667v2 vs i9-9900k the i9 is 60% faster single core, 44% faster multi-core (both are 8c16t). At lot of use cases, including the majority of development work for most people, is still dominated by single core performance so you're giving up a lot performance to save a couple bucks. The power draw on these server/workstation systems will likely also suck which will offset some of the $ savings.
The xeon is the same price as a Ryzen 2700 or an intel 8400.
They definitely beat it on power usage.
This all makes the argument really depend on how much utilization you think the xeon will have.
If you plan to run it 24/7 100% utilization (or as close as you can get), yes new hardware makes sense. (There's definitely a cost reason that these companies are ditching their old xeons for new ones, it saves them money and rack space).
If you're playing around for a few hours in the evening and on weekends (as suggested), it'll probably be a maximum of $10 per year in power costs.
So I think even with power costs, having run the numbers, these processors do make an attractive alternative to newer hardware. They are comparable and at a comparable price on the market, but if you can get a deal by bargain bin hunting, they can be a great value. It is unsurprising that the $/performance is very similar, that is how arbitrage markets are supposed to work.
There are some pitfalls here that are not mentioned, while the processor itself can be cheap, it is often the case that server grade motherboards are more expensive. This is particularly true with older Xeons, where people are trying to repair things that have broken. There's kind of a curve where demands drops up until a point and then the supply drops (people scrap things for the metal content rather than reselling)... so price starts to creep up again. This shouldn't be a problem.
The other issue is that you end up with hardware compatibility issues. Can I stick my new graphics card in the old xeon box? Probably(?) can I do something more exotic or new? maybe not? Again this shouldn't be an issue.
The hardware compatibility thing almost bit me. I have an older dell t3600 that a bought a new Radeon 480 GPU for. I only verified that the power supply could handle it and the pci slots were compatible. The card “fit” but I couldn’t put side cover back on, because there is a handle riveted to the inside of the side panel that collided with the card. I ended up having to drill out the rivets and remove the handle.
It makes sense if you need much Ram.
DDR3 RDIMMs used from ebay are really Cheap.
And the price difference for the whole system is huge if you buy used.
I payed for my System with a Xeon 2680 V2 (10C20T) 128GB Ram just a little bit more then the asking price for the 9900k alone.
But I bought it almost 2 years ago so the comparable Mainstream CPU at that time was the 4C8T 7700k.
As a student with a limited budget performance per € matters more then performance per watt.
I picked up a Dell T710 with 32 cores and 32GB of DDR3 ECC RAM for $400. I can throw more RAM in it for practically zero cost. It's amazing for batch workloads.
For interactive workloads, there's no need to get anything newer than a 4-series Core iX. The single-threaded performance has not improved much over the last five years. My daily driver is an i5-6500; motherboard and CPU, used, were $120.
I just got a used Thinkpad with an i7. I was eyeballing cheaper stuff with i5. The reason I upgraded is the security problems in CPU's require mitigations that will probably keep slowing them down. I bought a faster core to mitigate those slowdowns a bit. Plus, get security updates for a while longer.
I recommend people buying AMD where they can. Vermaden on Lobsters is a BSD expert. I asked them what hardware runs pretty much all BSD's. Vermaden narrowed it down to a few Thinkpads. All of them on eBay from recyclers had Intels. So, Intel just came with the box.
If building from parts or doing non-BSD, I'd go with an AMD, POWER, or ARM system.
All the ones I saw in good condition on eBay were Intel's. I got a T420 with Core i7 to mitigate potential slowdowns from future CPU vulnerabilities. I've occasionally had to restart it from suspend/resume issues. Otherwise, it's been great.
One more thing: the function key and control key are swapped compared to most laptops. I didn't like that because I'm used to control being far left. duclare told me about a BIOS setting that swaps them back. Everything's fine now. :)
I got an off-warranty Dell Precision from a previous employer. They were cheap and ordere them with the lowest Xeon CPU at the time, so I spent $100 on RAM/CPU to max it out. 6 sticks of 4GB DDR3 and whatever the Xeon equivalant to the i7-950 was, and threw a couple leftover 2TB HD's in there.
Made a great server. But for a workstation it wasn't really that snappy and I just ended up using a 2016 MacBook Pro.
The other thing to keep in mind is that most of the benchmarks you see on the internet for these older systems haven't been redone with the mitigations needed for the new speculative execution vulnerabilities, but the ones for the newer processors more likely have, so you're not comparing like with like.
A cheap eight core processor sounds great until it turns out to be slower at everything than an even cheaper modern quad core. A quad core Ryzen 3 is under a hundred bucks, with a 45W TDP instead of 130W.
my oldest Xeon workstation is a dual E5-2696v2 with 128G ECC RAM. it has a total of 24 cores. Compared to that, I am pretty sure most i9-9900k setups are just for kids. ;)
the cost of the above xeon system (dual cpu + mb + RAM) is slightly more expensive than i9-9900k itself, but much cheaper than a i9-9900k + motherboard + RAM.
not parent either, but your still comparing a single socket of his system with a 8 core ryzen. The latter is nearly double the single thread perf, but the former has 3x the total cores when both sockets are active. Worse the benchmark is likely tilted towards newer AVX workloads/instruction sets that the older cores don't have. So for boring old integer workloads that are easily parallelized (say compiling code) its likely that xeon is at least 30% faster. Possibly more.
I'm in the same boat, I have a 12 core/64GB xeon machine I picked up 4 years ago for $200 (130W compiling). Along the way I put 4 256G SSDs on the RAID5 controller. It likely gets totally thumped by the latest 12 core ryzens, but OTOH, my compile times are ~2 minutes for a fully clean build, on that machine. A similarly spec'ed thread ripper at work only pulls that down to 90 seconds or so. Logically I can't really justify the ~$1800 the new machine is going to cost by the time I get a raided SSD, MB and 64GB ram. I might do it anyway, but its definitely not logical.
Utilizing a 24 core system is harder than an 8 core system. You will have to run multiple compiler instances which eat up memory. If you only change a single file it will take 60% longer than the newer system. Especially if you're using a single threaded linker you will be dependent on the performance of a single core. The 24 core system just doesn't perform well at all. 30% faster than a single CPU means 65% the speed of a dual socket system. Remember the original comment said that the 8 core system is "just for kids" but it's the opposite. The lower multicore performance will be compensated by the significantly higher single core performance and the upfront savings will be compensated by the fact that the newer system consumes less power.
I tend to subscribe the the fastest possible single thread mantra as well and buy low core count, high frequency machines for desktop usage. OTOH, compiled code software development is one of those areas where your better off with a few more cores vs higher frequency/newer machines. A very large number of opensource/automake/cmake/etc based projects have nearly perfect scaling out to at least a hundred threads. This is frequently true even for linker and packaging passes (given a parallel compressors like pigz). Between incremental linking, and runtime/loader linked systems (think linux kernel modules where each module is effectively a few C files linked into a .ko, each linked independently) this can even be true for the link phase as much of the overhead is IO/syscall bound even when cached.
Either way, your still missing the point, these machines cost a couple hundred dollars, and they might be a few percent slower, and 3x the power budget, but where I live that power is going to work out to $20 a year. Its going to take 50 years to make up the price difference in power, and in 5 years i'm going to buy the same machine someone paid $4k for $400 and that person is going to have to spend another $4k to get something better.
E5-2667v2 vs i9-9900k the i9 is 60% faster single core, 44% faster multi-core (both are 8c16t). At lot of use cases, including the majority of development work for most people, is still dominated by single core performance so you're giving up a lot performance to save a couple bucks. The power draw on these server/workstation systems will likely also suck which will offset some of the $ savings.