New workstation for home use.

[BravoV]

A nice summary what on earth ECC RAM is for -> Why use ECC?

Point is, memory is "relatively" cheap nowadays, compared to last year, yes, ECC still more expensive, but too overly pricey, but if you valued your memory integrity, ECC is good bet.

Then why not, if the mobo supports it.

This feature was out of reach in desktop computing, until Ryzen.

No, gamer or casual computing do NOT need ECC, its just a big waste of money, imo.

[olkipukki]

This feature was out of reach in desktop computing, until Ryzen.

Back to old good (Intel) days  , entry-level and up Xeons have supported ECC in UDIMM format similar to Ryzen.

I wouldn't say "was out of reach" 

[Karel]

OK. I installed freetype fontconfig and cups dev packages to get that to work.

Much faster, but it seems my 32 core is still a little slower than your 12 core. Those 2nd gen chips are good!

Code: [Select]

real 2m52.332s
user 73m38.355s
sys 9m28.145s
 21:22:27 up  7:53,  1 user,  load average: 37.62, 24.79, 13.60

(user+sys)/real is now 28.9, so it's using most of the cores most of the time.

Doing just "make clean && make" instead of a completely new build directory got me 2m43.4s.

Maybe also the Samsung 970 PRO NVMe plays a role here.

My components:

1TB Samsung 970 Pro M.2
128GB DDR4-2400/2666 PC4-19200/21300
Gigabyte X399 AORUS Gaming TR4

There's plenty of RAM for caching everything so it shouldn't even be hitting the SSD much.

These sites have some interesting info about how cpu's perform when compiling big software projects:

https://openbenchmarking.org/test/pts/build-linux-kernel

https://www.phoronix.com/scan.php?page=news_item&px=Kernel-Bisecting-5.9-Ongoing

[brucehoult]

These sites have some interesting info about how cpu's perform when compiling big software projects:

https://openbenchmarking.org/test/pts/build-linux-kernel

https://www.phoronix.com/scan.php?page=news_item&px=Kernel-Bisecting-5.9-Ongoing

Compiling is a little bit of a conflicted benchmark because at least open-source projects (especially those combining a lot of subprojects) usually tend to alternate phases running "configure" which typically only uses maybe two cores or linking which uses one core, with phases which can use as many cores as there are C files in the current subproject.

So it varies from project to project a lot. A CPU that has a lot of cores but can't turbo one or two cores up to infinity can suffer a lot from Amdahl's law.

Anyway, in the linux kernel builds, my 32 core 2990wx ranks at 40 seconds and the only things faster than it are 2nd gen Threadrippers with 24+ cores, EPYC with 24+ cores, or dual XEONs with 20+ cores each (40+ cores total). Only the 24 core ThreadRipper 3960X is (a little) cheaper. And the "2 x AMD EPYC 7282 16-Core", at least for the CPUs themselves -- the needed motherboard surely is expensive.

The OP's 3900X clocks in at 53 seconds, which is very very respectable, especially for the price.

Incidentally, I just a couple of days ago took delivery of a Lenovo ThinkPad E14 Gen 2 with 6 core (Zen 2) Ryzen 5 4500U, which was around US$735 including tax and shipping. It clocks the Linux kernel build page at 194 seconds.

Both my 2.5 year old NUC and the 6th gen X1 Carbon I had at my last job have an i7-8650U, rated at 256 seconds.

The current entry level 8th gen X1 Carbon and the Intel version of the E14 both have i5-10210U rated at 258 seconds.


The kernel bisecting link doesn't have any comparative CPU figures, but just points at the same kernel build stats and also LLVM build stats (which don't cover many machines))

[iteratee]

If I were building a new high-end workstation today I'd be seriously looking at Power9. I'm not married to Windows, so while the ability to virtualize Windows on my main rig is nice, it's not mandatory - I have other x86 boxen for that, which opens me up to any suitable alternate architecture.

[brucehoult]

If I were building a new high-end workstation today I'd be seriously looking at Power9. I'm not married to Windows, so while the ability to virtualize Windows on my main rig is nice, it's not mandatory - I have other x86 boxen for that, which opens me up to any suitable alternate architecture.

The Linux kernel build times link a couple of messages up has any entry for POWER9 8 core 16 thread CPU taking 298 seconds. As far as I can tell, building any kind of decent system around it (e.g. using Talos II Lite) is going to cost at last $4000.

This is slower than, say, an AMD Ryzen 3 3200G (237 seconds) which I see on Amazon for $99.99. I also see HP offering a complete PC (M01-F0020) with 8 GB RAM and 512 GB SSD for $449.

[iteratee]

If I were building a new high-end workstation today I'd be seriously looking at Power9. I'm not married to Windows, so while the ability to virtualize Windows on my main rig is nice, it's not mandatory - I have other x86 boxen for that, which opens me up to any suitable alternate architecture.

The Linux kernel build times link a couple of messages up has any entry for POWER9 8 core 16 thread CPU taking 298 seconds. As far as I can tell, building any kind of decent system around it (e.g. using Talos II Lite) is going to cost at last $4000.

This is slower than, say, an AMD Ryzen 3 3200G (237 seconds) which I see on Amazon for $99.99. I also see HP offering a complete PC (M01-F0020) with 8 GB RAM and 512 GB SSD for $449.

Yeah the cost-performance is still just a bit much for the other perks to compensate enough for this to be an easy choice. One of the main appealing points to me is for multi-socket systems. Xeon has always been packed with arbitrary limitations that make the cost impossible to justify over a single-socket consumer-grade system especially when considering overclocking is added to the equation.

I haven't researched current hardware for the last few years though. I don't even know all the model naming conventions these days.

[brucehoult]

If I were building a new high-end workstation today I'd be seriously looking at Power9. I'm not married to Windows, so while the ability to virtualize Windows on my main rig is nice, it's not mandatory - I have other x86 boxen for that, which opens me up to any suitable alternate architecture.

The Linux kernel build times link a couple of messages up has any entry for POWER9 8 core 16 thread CPU taking 298 seconds. As far as I can tell, building any kind of decent system around it (e.g. using Talos II Lite) is going to cost at last $4000.

This is slower than, say, an AMD Ryzen 3 3200G (237 seconds) which I see on Amazon for $99.99. I also see HP offering a complete PC (M01-F0020) with 8 GB RAM and 512 GB SSD for $449.

Yeah the cost-performance is still just a bit much for the other perks to compensate enough for this to be an easy choice. One of the main appealing points to me is for multi-socket systems. Xeon has always been packed with arbitrary limitations that make the cost impossible to justify over a single-socket consumer-grade system especially when considering overclocking is added to the equation.

I haven't researched current hardware for the last few years though. I don't even know all the model naming conventions these days.

Don't get me wrong here -- I'd love to have a POWER9 machine. I'm kinda sad I sold my Dual 2.0 GHz G5 PowerMac when the Core 2 Duo Macs came out. I still have a couple of 17" G4 iMacs and G4 Minis.

And I buy RISC-V hardware with right now pretty bad price-performance compared to Intel or ARM e.g. my $999 HiFive Unleashed board that performs only a little better than a $40 Pi 3 and doesn't even have video.

If there was a POWER9 machine for under $1k I'd probably get one...

[Howardlong]

Here's my story about ECC RAM: it can be a double edged sword.

I ran a dual Xeon 24C/48T 64GB beast for a couple of years until suddenly one day I started getting ECC errors that pretty much ground the machine to tortoise pace filling up logs. I'd reboot, and then a couple of hours later they came up again. You couldn't really install updates because it would just go into its unusable tortoise state, serialising with very high CPU on one of the 24 cores. I tried reseating the CPUs & RAM, reducing the RAM to one stick per socket etc etc, but nothing fixed it.

In the end I mothballed the dual Xeon, and switched to an i7-8700K (32GB), and then to an i9-9900K (64GB) which is what I use now.

About 4 or 5 months ago, I decided to power up the dual Xeon again, and sure enough the ECC errors reappeared after an hour or so. I left it running overnight and it finally installed a bunch of updates over the next two days... and I've not had one ECC error since.

So the ECC errors turned out to be an OS feature introduced by auto updates months earlier without me knowing. It was Windows, of course, which I am sure will moisten the multitude of limbering loins of all the Linux fanboys out there for sure!