Running an FX-8150 at work now

[just brew it!]

Needed an upgrade, as the mobo in my existing work box was maxed out at 8GB and this was not enough to run multiple Quartus II (FPGA synthesis tool) compilations smoothly. My current workflow involves doing multiple FPGA compilations at once. The compilations take upwards of an hour or more, and don't take good advantage of more than 2 cores; so I need to work multiple issues at once in multiple instances of the Quartus II environment.

At the end of the day the FX-8150 looked like a great bang-for-the-buck in spite of its known issues. The 8 cores should allow me to do many compiles at once without bogging down, given enough RAM (the compiler is a massive resource pig, consuming multiple GB).

So I've put together a system consisting of:

Asus M5A97 EVO
AMD FX-8150
Coolermaster Hyper TX3 tower heatpipe cooler
16GB ECC DDR3-1333
Seasonic 520W PSU
Corsair Force 3 240GB SSD (60 GB boot partition, 32GB swap partition, the remainder used as staging area for FPGA compilation)
Pair of 1.5TB Caviar Blacks in RAID-1 (storage)
Some cheap-ass Visiontek passively cooled Radeon card with dual digital out
Cheap DVD burner
Inexpensive NZXT mid-tower case with plenty of openings for extra fans
Cheap-ass USB 2.0 PCI card with a VIA USB controller on it (for some reason Quartus II's FPGA programming tool seems to work best with VIA USB ports, go figure)

Total price tag came in at just under $1,200.

So far, so good. I've been using it for about a week, and I really like it. Individual compiles aren't much faster than my old Phenom II X4 box (not that I expected them to be), but I can run 4 builds at once and the only indication anything out of the ordinary is going on is that the TX3 intermittently becomes audible as it spools up to keep those 8 cores cool. The system remains incredibly smooth and responsive throughout (the old system got laggy as all of the CPU cores got pegged, and became all but unusable as things intermittently spilled over into the swap file at certain points in the compilation).

The only weirdnesses I've hit so far all relate to power management and monitoring in Linux:

1) Enabling Cool-n-Quiet on this system crashes the Linux jack audio stack. But, disabling CnQ doesn't seem to increase idle power draw, temps, or noise by a huge amount; apparently the CPU is pretty good at managing its own power usage based on load even when the core clocks aren't being throttled down. It also helps that this motherboard has good thermal CPU fan speed control (which works even when CnQ is off).

2) The CPU temperature and fan speed sensors on the M5A97 EVO weren't supported out-of-box; I had to chase down (and compile from source) a driver for the monitoring chip Asus used.

3) For some strange reason, Linux is reporting that the cores are all idling at 3.9GHz (with CnQ off... with CnQ on they were idling at 1.6GHz IIRC). Stock for the FX-8150 is supposed to be 3.6GHz; 3.9GHz is the level 1 "turbo core" speed (level 2 is 4.2GHz). WTF? But as long as the temps and noise are reasonable, I guess I don't really care... with no load temps are in mid-30s and the fan is silent, which is fine by me.

Edit: At full load, CPU temp seems to peak a degree or two above whatever I set as the "100% CPU fan speed" temperature threshold in the BIOS; given that the max temp for the FX-8150 is 61C, I've set the threshold to 55C to minimize the amount of time the fan runs full throttle while also keeping the CPU temps within spec.

[anotherengineer]

Right on, cool stuff JBI.

I guess since it is a production machine you don't have the time to see how well it can undervolt?

I have a 955BE at stock 3.2 ghz in a GA-990xa-ud3 with CnQ & C1E disabled and have it at 1.225V perfectly stable.

I would be curious what that chip could undervolt to?

edit- it helps my power consumption and temps at idle and full load, which keeps my fan from ever going over 600rpm on my noctua cooler

[kc77]

Congratulations!

Personally I went ahead and bit the bullet on a dual Operton 4284 setup (Bulldozer). To tell you the truth BD's performance is remarkably better in Linux than it is in Windows. Handbrake itself is easily 25 to 30% quicker in Linux and everything just feels snappier. To me at this point BD's are actually a pretty good deal.... IF you are honest with yourself on why you are buying it. I primarily run Linux where BD runs better. I also run VM's normally day to day and I play the random Linux game or two.

For me it performs quite well. I've run a game while having my normal 2 VM's up and didn't even notice they were running. This wasn't possible on my i7 workstation before. Overall I like the performance of it considering the price I paid for the whole system which was 900 for procs+MB and another 200 for the memory.

Regarding Turbo and C6 states.. I have noticed the peculiar behavior as well. It seemed to idle at 1400 and clock up to stock @ 3000Mhz. However, I was using system-monitor and some widget to view my clock speed... these didn't report it right. I had to use cpufreq-aperf to physically see it break past the stock clock.

[just brew it!]

I guess since it is a production machine you don't have the time to see how well it can undervolt?

Oh, I suppose once the workload eases off a bit (we're in a crunch time right now) I might have some time to play around with it a bit. Not much motivation to undervolt though since the fan only becomes audible when I really push it, and I'm not paying for the electricity.

Personally I went ahead and bit the bullet on a dual Operton 4284 setup (Bulldozer). To tell you the truth BD's performance is remarkably better in Linux than it is in Windows. Handbrake itself is easily 25 to 30% quicker in Linux and everything just feels snappier. To me at this point BD's are actually a pretty good deal.... IF you are honest with yourself on why you are buying it. I primarily run Linux where BD runs better. I also run VM's normally day to day and I play the random Linux game or two.

Yes, I run VMs as well, both to build (other) projects that need specific build environments, and to run the obligatory Windows VM to deal with the corporate stuff. Since it's a work machine I'm not gaming though...

For me it performs quite well. I've run a game while having my normal 2 VM's up and didn't even notice they were running. This wasn't possible on my i7 workstation before. Overall I like the performance of it considering the price I paid for the whole system which was 900 for procs+MB and another 200 for the memory.

Yup, like I said bang for the buck seems to be pretty good. My office mate upgraded just a couple of weeks before me, but went Intel instead. I think he spent nearly twice as much (of someone else's money... ) for roughly equivalent performance, and doesn't have support for ECC RAM (yeah we can argue about how important that is on a desktop) or IOMMU (which could come in handy given the sort of things we do... being able to hand off bare PCI/PCIe devices to a VM could come in handy).

Regarding Turbo and C6 states.. I have noticed the peculiar behavior as well. It seemed to idle at 1400 and clock up to stock @ 3000Mhz. However, I was using system-monitor and some widget to view my clock speed... these didn't report it right. I had to use cpufreq-aperf to physically see it break past the stock clock.

Sounds like a slightly different (but possibly related) issue. It seemed to behave more or less as you describe when CnQ was enabled. But with CnQ disabled (how I'm running it now) it appears to be running 300MHz *above* stock even when it is idling. At least, that's what /proc/cpuinfo is telling me... it could be lying to me, especially given that I'm still running Ubuntu 10.04 LTS (which predates the Bulldozer core).

[just brew it!]

Another interesting side note: The video card I used in this build (http://www.newegg.com/Product/Product.a ... 6814129154) is apparently capable of three-head output (as long as one of the monitors is analog). I'm currently running dual 1600x1200s; guess I need to see if I can scam a third monitor from somewhere. Having a wall-o-monitors is pretty handy when you're juggling a half dozen different tasks at once...

[just brew it!]

Built the Realtek Linux audio driver from source yesterday; now Ubuntu recognizes the onboard 8-channel HD codec (instead of treating it as a generic 2-channel AC97). Not terribly important for a work machine, but I do like the ability to plug the headphones and speakers into separate line out jacks, and have different EQ curves set for each. (Been working a lot of evenings lately; after the office mate leaves I tend to take the headphones off and use the speakers instead...)

Much to my delight, the pair of built-in JMicron eSATA ports just work with Linux out of the box (no special drivers needed). Tested them today.

On a less delightful note, learned that the kernel Ubuntu 10.04 LTS is currently stuck on (2.6.32) is one point revision too old to support automatic TRIM for the SSD. Not sure how much of an issue that really is... guess I'll find out.

[Airmantharp]

Good to see all the pieces falling together!

Given that Sandforce controllers have pretty good garbage collection, you should be fine until you can get a Trim-enabled kernel. I like seeing things like audio working, I just fought through getting a Xonar U3 to output DDLive.

[just brew it!]

Looks like I've got two options for getting the TRIM support:

1. Install a backported 2.6.35 kernel. This is available in the Ubuntu repository, but isn't installed as part of the automatic system update process; you have to explicitly install it.

2. Run a batch TRIM tool once a week as a cron job.

I'm currently leaning slightly towards option #2 since it allows me to stick with the official 10.04 LTS kernel until I finally move on to 12.04 LTS (or to Debian, haven't made up my mind yet). There also seems to be some disagreement regarding how well ext4's "on the fly" TRIM works (i.e. overhead of issuing the TRIMs negatively impacting performance when lots of blocks are getting deleted). Periodically batching up all the TRIMs in the middle of the night when I'm not at work sounds like a better all-around solution to me!

Edit: Another small thing I forgot to mention - the front panel audio jack is surprisingly quiet. No noise or static at all. The fact that the front panel audio cable is routed through a hole near the bottom of the case and around the back of the motherboard tray probably helps, by putting what amounts to a big ground plane between the audio cable and all the EMI-emitting components on the motherboard.

Edit 2: This is the case I used. No-nonsense black mid-tower with plenty of ventilation options. My only gripes are that the sheet metal is a little on the thin side (not unexpected in a $50 case), and the tool-less hard drive retaining clips seem rather flimsy (but you can still use standard hard drive mounting screws if you prefer, so this is basically a non-issue).

[ronch]

JBI: I read your first post on this thread about the chip running at 3.9GHz at idle and remembered some guy at Tigerdirect.com who also bought an FX-8150 who noticed the same exact thing (review by WEK on July 14, 2012).

http://www.tigerdirect.com/applications ... u=A79-8150

Not sure if that helps but for some odd reason the FX-8150 seems to think it has to turbo up to 3.9. With one thread alone running it's supposed to be 4.2, so at 3.9 the chip may have decided a couple of cores should be clocked at that speed to service all your background tasks.

[Airmantharp]

I'd probably stick to using Trim off-shift as well, since it looks like the easiest solution, and will probably affect performance less. Making the upgrade would make sense if you're abusing the SSD though, as you'd want to reclaim that deleted space as quickly as possible.

And that case is sharp, I'm impressed! It looks like a great candidate for budget builds, gaming or otherwise, with the only tweak I can see needed would be to ensure that the front intakes are filtered.

[just brew it!]

I like seeing things like audio working, I just fought through getting a Xonar U3 to output DDLive.

Finally got around to configuring all the audio settings to my liking. Even got a little mic taped to the ringer on my desk phone that feeds through to the headphones (but not the speakers), so I can still hear the phone ring when I've got the headphones on.

[ludi]

Even got a little mic taped to the ringer on my desk phone that feeds through to the headphones (but not the speakers), so I can still hear the phone ring when I've got the headphones on.

I've always considered that one of these is the ideal tool for the job. Not only is it audible through your headphones, it can invoke involuntary excretory functions a good two or three offices away.

[axeman]

This is completely off topic, but what does one use the jack audio stack for? I've seen it mentioned here and there, but no experience with it. Not that I'm a big fan of PulseAudio

[anotherengineer]

Hey JBI

If you don't have trim you should have gotten a plextor M3 or corsair performance pro
http://www.xbitlabs.com/articles/storag ... html#sect0

Sorry I didn't tell you several months ago.

[Airmantharp]

Hey JBI

If you don't have trim you should have gotten a plextor M3 or corsair performance pro
http://www.xbitlabs.com/articles/storag ... html#sect0

Sorry I didn't tell you several months ago.

That's essentially the same for any Sandforce 2 drive. The base firmware these drives use make them the best performance choice for a Trim-less install with their excellent garbage collection.

[just brew it!]

This is completely off topic, but what does one use the jack audio stack for? I've seen it mentioned here and there, but no experience with it. Not that I'm a big fan of PulseAudio

It's a synchronous (audio samples flow through the system in lockstep) audio stack, developed with an eye towards professional audio applications. I originally installed it because I wanted to play around with Rosegarden and FluidSynth (MIDI sequencing and synthesis packages), and they require it. After discovering that it lets you do things like separate EQ curves for individual input sources and/or output jacks, and in general gives you a great deal of control and flexibility in your audio setup, I decided to try it for everyday use.

Unfortunately JACK can't be used as a complete replacement for PulseAudio, since many applications (e.g. the Flash browser plugin) don't know how to talk to it. It also can't run alongside PulseAudio, because they both want to take over control of the lower level ALSA layer. The solution is to run PulseAudio ON TOP of JACK, feeding the output of PulseAudio into JACK as if it was another JACK signal source. In this configuration, both PulseAudio/ALSA and JACK applications can be run without restarting the audio stack.

Sadly, the stability of this arrangement leaves something to be desired. I already mentioned the problems it seems to have with CPU power management. The PulseAudio JACK plugin is also not quite 100% stable; the latest version of the Flash browser plugin seems to give it indigestion, causing videos to stutter quite badly; I've temporarily backed off to an older Flash plugin to deal with this.

I'm hopeful that newer versions address these issues, but I'm still running Ubuntu 10.04 LTS on my primary desktops (work and home), and haven't had much time to play with newer distros (or figure out how to shoehorn a newer version of JACK and/or PulseAudio into 10.04).

[Flatland_Spider]

This is completely off topic, but what does one use the jack audio stack for? I've seen it mentioned here and there, but no experience with it. Not that I'm a big fan of PulseAudio

Think of JACK as a digital patch panel or mixer board.

[just brew it!]

This is completely off topic, but what does one use the jack audio stack for? I've seen it mentioned here and there, but no experience with it. Not that I'm a big fan of PulseAudio

Think of JACK as a digital patch panel or mixer board.

...with a huge library of things you can plug into that patch panel, and support for applications that will do pretty much anything you could want to do to a real-time audio stream.