As I'm waiting for my 3900x to arrive (the delivery date keeps getting pushed), I decided to try and do some crunching on a couple of old HP SFF towers that I got super cheap. And then I did some value comparisons between those towers and a Zenbook laptop measuring power draw at the wall.

If my logic is correct (and I'm very unsure about that), it would take just over 3 years of 24/7 crunching on the zenbook to to save the difference in purchasing cost over the less energy-efficient of the two cheap towers.

The math (rounded for convenience):
- The HP 4000 Pro SFF with Pentium E5800 @3.2 GHz manages 250k points in 30 days. Power cost for this period is 500 cents.
- The Zenbook with i7-7500U manages 250k points in 11 days using 120 cents of power.
- So with the Zenbook I save 380 cents of power every 11 days.
- And from that I calculated that I would save the price difference in 3 years and 2 months.

But wait... how can I save 34,5 cents per day when the HP machine only uses 16,7 cents per day and the zenbook about 10,9 cents per day?

Well, I guess my logic here is that I aim to get the same amount of points with both machines in the long run, however long that may be. And to achieve that the zenbook would work significantly less time and at the same time draw significantly less power.

Then considering the additional value the zenbook gives me from actually using it in addition to crunching, it's just so much a better choice.... or is it?

I guess what I'm hoping for is someone to prove me wrong and say it's okay to keep those old towers crunching :D despite the electricity costs. Frankly I'm surprised by the results, I expected the old towers to hold up better.
Then again, I probably won't have the Zenbook crunching for 3 years... nor the towers for that matter.

Machine: Electricity price per 1000 points (prices are in euro cents)
HP Compaq 4000 Pro SFF, Pentium E5700: 2,07
HP Z200 SFF, i5-650: 1,48
Custom desktop build with i7-7700K and non-crunching GTX 1080: 0,79
Zenbook CU580KD, i7-7500U: 0,47

Can't wait to do the same math with the 3900X build :)

Only you can decide if it's worth it or not

If you're not using the waste heat, I would not build dedicated crunchers with inefficient CPUs.

The used market has lots of cheap Ryzen 1700 CPUs, my 1700 does 250k points in 3.3 days, with a 82W power draw.

Back when the 3000 was first announced, there was a push to get rid of inventory of the 1700s. 1700s went down to $135 each, complete with cooler. Built three rigs for a total cost of about $1200. Still the best bang for the buck as far as I've found. Combined, those three rigs equal the result output of my dual Epyc 7301 machine with 32c/64t, but were about 30% cheaper. The only workload that they wouldn't run as well as the epyc would be MIP, because the 1700s have 1MB/thread as opposed to 2MB/thread for the Epyc.

As for mobos, the cheapest mobos won't run the 3000 series, but for $60 they run the 1700 just fine. I'm definitely happy with the combo; encourage others to see if they can find a good deal on 'em!

A 6 core/12 thread 3600 easily outperforms an 8 core/16 thread 1700?

I wouldn't be surprised if it did outperform the old Ryzen 1700! There was, what, like 25% IPC gain between the first and third gen Ryzens, in addition to a higher base and boost clock speed? That said, a lot of that boost potential is based on how cool you can keep the chip. If I remember correctly, the default cooler that comes with the 3600 is solid aluminum without the copper vapour chamber. So, you might have to spring for an aftermarket cooler too. And the 1700s will also run on A320 motherboards which are about $15 cheaper as well. Small additional costs, sure, but if you're trying to create a farm, differences in price per machine add up quick haha.

The one thing I know for sure is that whether you go for the 1700, 2700, or anything 3000, they're all MONSTER chips for WCG