Beating the PS3

Show off your quiet rig.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
johannes
Posts: 53
Joined: Tue Oct 23, 2012 8:13 am
Location: Suisse Romande

Beating the PS3

Post by johannes » Mon Apr 08, 2013 10:39 am

I wanted to build a system which is better than the PS3 slim - smaller, faster, quieter and more energy efficient. I succeeded only partially, but I will try again with Kaveri and in the mean time I'm happy with my little gaming machine.

Comparing the size you can see that my system is smaller - despite the huge power brick.
Image

I wanted a system I could use for playing at 720p with medium to high details, so I went with AMD's trinity APU. A low power intel processor like the i7-3770T combined with a Radeon 7750 would have given me more performance at about 100W TDP, but small cases with support for an half-height expansion card were difficult to source in my part of the world and I had thermal concerns.

The configuration:
AMD A10-5700 (65W Trinity, 3.4GHz, 4 GHz Turbo)
MSI FM2-A75IA-E53 (BIOS v1.2)
2x4 GB G.Skill Sniper Memory (1866 MHz 9-10-8-28, 1.5V)
Samsung SSD 830 128GB
PicoPSU 160XT with 192W external brick (89% avg. efficiency)
Noctua NH-L9a CPU cooler
M350 case

In fact, I ordered the M350 and the PSUs from the US, because it was cheaper than ordering from most European shops. That's also why I got the huge (both physically and in capacity) 192W power brick - it was a cheap option to get a high-efficiency power brick.

Using a high-efficiency power brick paid off, as my measurements show: 23W idle and <90W while gaming are quite good. In idle this is a bit more than 2.5 times lower than the PS3, but for gaming the PS3 uses less energy.

Heavily inspired by the M350 race machine I also built a different front and due to size constraints I had to tape the SSD to the underside of the case. To do this I had to put different feet to increase the distance from the floor.

This little system is up for gaming at 720p at medium to high details and loading times are very short thanks to the SSD. I haven't done a visual comparision, but I will count this as a draw for the performance and a clear win for the SSD equipped system regarding loading time.

For noise, a look at the opened case gives a good idea of potential problems:
Image
The RAM blocks one half of the cooler's exhaust and the other half is directed towards the I/O-shield - not a recipe for efficient exhaustion of hot air. Furthermore the hot air will raise and be sucked in again by the cooler. This is actually no problem with an opened case, but with the case closed this leads to the fan running at 100%. (Thermal target in BIOS is 55°C.)

SPCR has reviewed the intel version of the cooler with the same fan. To my ears it is inaudiable at slightly more than 1000 RPM (the rotational speed at idle), but becomes noticeable at about 1800 RPM, but not badly so. At about 2000 RPM it is still below the noise floor while gaming and has a pleasant sound. At full speed (reported as 2500 RPM) it is loud and distracting.

I tried some things to improve this (reversing the fan, adding fans, turning the case), but in the end I settled for a two zone design:
Image
A cardboard layer separates the cool incoming air from the hot exhausted air. This removes the cooling effect from the upper part of the case, but leads to fan speeds of about 2000 RPM while gaming. I still can get the fan to shortly speed up to 100% while stress testing and I'm afraid in summer I might have some problems with noise.

I wanted to avoid the board cooking in the hot air, so I added a 40mm fan. The noise is OK up to 4000 RPM and clearly audible at 4800 RPM, but not distracting while gaming at a living room distance. With an airflow rating of 8m³/h it is probably not hugely effective at cooling (the 92mm fan of the CPU cooler can do 50m³/h), but some parts close to the fan feel cooler to the touch. Unfortunately I don't get reliable thermal data from the programs I tried, so I have no better data upon which to base my decisions.

Lastly there is coil whine from the PSU at idle. It is worse with C6 activated, but does not disappear when it is disabled. This is no big issue for me, but might be for someone building an HTPC with similar hardware.

Winning against the PS3 on noise is easy - the disc drive has a horrible sound and the fan can be heard in quieter scenes, although not badly so. Neither are silent and in summer this might change, but for now I prefer the noise from my machine to the PS3.

Undervolting is currently not effectively possible (PSCHECK and BIOS do not work. Overdrive does, but installing it disable Wifi and Bluetooth), but there is hope for the quick release of a tool capable of changing voltage.
You do not have the required permissions to view the files attached to this post.

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Re: Beating the PS3

Post by Rebellious » Thu Apr 11, 2013 9:58 am

Isn't paper cardboard a fire hazard?

This PicoPSU has some bulky hardware on the large plug that goes into the 24-pin power connector. Most Mini-ITX mobos have the memory slots right next to the 24-pin connector, but not yours. So it looks like the PicoPSU capacitors might touch the memory modules on most miniITX mobos, or even not fit at all. Is that why you picked this MSI mobo?

johannes
Posts: 53
Joined: Tue Oct 23, 2012 8:13 am
Location: Suisse Romande

Re: Beating the PS3

Post by johannes » Thu Apr 11, 2013 11:08 am

The cardboard doesn't touch any hot parts (it touches the upper part of the fan made of plastic and parts of the metal frame) and since paper needs about 220°C (think Fahrenheit 451) to ignite, I'm not afraid of the cardboard starting to burn. In case that any other component starts burning it is indeed a problem.

The mainboard choice was motivated by supply and compatibility with custom backplates for small form factor coolers. The second criterium disqualifies the Asrock boards (where the first generation was a fire hazard), which left me with the MSI board and the Zotac A75 one (there are only three companies currently offering FM2 mini-itx boards). The MSI board had good reviews, while I couldn't find any information for the Zotac board except that it doesn't support undervolting. So I profited from a visit to Germany to buy the known and much cheaper (75€ vs 100€) board. (In Switzerland only the Asrock A85 board is readily available.)

Regarding the size of the PicoPSU: The components measure roughly 1cm from the middle of the ATX connector. Note also that I put the 4pin connector which is used for SATA and Molex 4-pin power (red, black, black, yellow cables) facing the mainboard, by default it is on the side facing away from the board. Judging from a quick look at the layout of the other boards your concern is warranted for RAM sticks with heat sinks. If you can get some of the Magic low profile Samsung RAM or similar heat sink-less sticks it should be fine.

And final good news: Undervolting works well with AmdMsrTweaker and the power consumption difference from undervolting is significant. If you are fine with some underclocking you can save 20W at 3 GHz or 30W at 2.6 GHz. I've quite some testing and benchmarking to do before I settle on a frequency setting, but this should do wonders for cooling and thus noise.

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Re: Beating the PS3

Post by Rebellious » Sun Apr 14, 2013 9:31 am

Thank you. Yeah the PicoPSU would be a tight fit with adjacent RAM sticks. I don't even know if memory modules are susceptible to EMI interference (from the PicoPSU).

AmdMsrTweaker was just updated to support Trinity and Richland, so perfect timing on your part. There's nothing for Linux yet, k10ctl is for 10h only, so no VID/ FID control.

Instead of cardboard, maybe you could plug some of the case holes with hot-melt glue and then touch it up with a black marker.

johannes
Posts: 53
Joined: Tue Oct 23, 2012 8:13 am
Location: Suisse Romande

Re: Beating the PS3

Post by johannes » Mon Apr 15, 2013 1:47 am

Please ask if you need more detailed measurements or photos for the PicoPSU. I don't think the PicoPSU creates much EMI - all it does is to forward 12V signals and convert 3.3V and 5V signals from the 12V signal.

Changing frequencies or voltage is actually much simpler than I thought. You "just" have to write some magic values to model specific registers. These registers are exposed to the file system by the Linux kernel and the values are explained in the developer guide: http://support.amd.com/us/Processor_Tec ... h_BKDG.pdf (See section 2.4.19) Should you have a need for undervolting on Linux, I could probably help you to at least create an echo command for the desired parameters.

The cardboard serves to separate cold and hot air. I want the holes to be open so the fan can suck in cold air. But I want to avoid that hot air from the case is taken in again by the fan, so I used the cardboard to prevent hot air from entering the zone close to the fan.

xan_user
*Lifetime Patron*
Posts: 2269
Joined: Sun May 21, 2006 9:09 am
Location: Northern California.

Re: Beating the PS3

Post by xan_user » Mon Apr 15, 2013 9:08 am

Rebellious wrote:Isn't paper cardboard a fire hazard?
i did the exact same mod on my 350, but i used clear packing tape on the inside of the cover, never once over the last 5 years has it ever gotten warm.

if flammability is a concern, one could use foil tape made for heater ducts. its rated up to 260* f. (and i bet it blocks RF too. Come to think of it, i wonder if thats the same tape they use in laptops to shield components from heat, and RF....?)
I don't even know if memory modules are susceptible to EMI interference (from the PicoPSU).
OP could always use an ATX extension cable and put the pico elsewhere.
-but, if pico's were causing rf interference, i believe we would have discussed it here on SPCR by now...

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Re: Beating the PS3

Post by Rebellious » Tue Apr 16, 2013 11:44 am

johannes wrote:Please ask if you need more detailed measurements or photos for the PicoPSU. I don't think the PicoPSU creates much EMI - all it does is to forward 12V signals and convert 3.3V and 5V signals from the 12V signal.

Changing frequencies or voltage is actually much simpler than I thought. You "just" have to write some magic values to model specific registers. These registers are exposed to the file system by the Linux kernel and the values are explained in the developer guide: http://support.amd.com/us/Processor_Tec ... h_BKDG.pdf (See section 2.4.19) Should you have a need for undervolting on Linux, I could probably help you to at least create an echo command for the desired parameters.

The cardboard serves to separate cold and hot air. I want the holes to be open so the fan can suck in cold air. But I want to avoid that hot air from the case is taken in again by the fan, so I used the cardboard to prevent hot air from entering the zone close to the fan.

I'm the poster in xtremesystems.org/forums/ referring to k10stat. I guess you can do a lot in Linux but I just want an easy way to set CpuVid/Fid, without turning this into an engineering project. The lack of such tool is holding many people back from transitioning from Phenoms to FX CPUs, so any help will be most welcome.

re. hot melt, I meant to block "some" of the holes to prevent re-circulation. I'd lay it flat on a piece of glass and apply the melt from the inside. Then paint it with a black Sharpie. Hot-melt is one of my favorites for quick prototypes.

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Re: Beating the PS3

Post by Rebellious » Tue Apr 16, 2013 11:55 am

BTW, I have a PcioPSU 150-XT and an Asus P8Z77-I DELUXE/WD, but haven't tested yet. I will sometime next week, and will listen for any coil whine. I have a hard time believing that Pico is equivalent to a $120 ATX PSU, and to trust that it's equally stable, especially during S3.

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Re: Beating the PS3

Post by Rebellious » Thu Apr 25, 2013 10:42 am

Ha! Are you the johannes that posted amdmsrtweaker-lnx in xtremesystems?

Anyway, I've installed the PicoPSU 150XT kit, seems stable even with the i5 3570K clocked to 4.4GHz, S3 is good, and no coil whine. My only concern is whether the mobo/chassis are properly grounded. The 12V power connector has only 2 leads, what happens if you have a static discharge to the case?

johannes
Posts: 53
Joined: Tue Oct 23, 2012 8:13 am
Location: Suisse Romande

Re: Beating the PS3

Post by johannes » Thu Apr 25, 2013 10:19 pm

Yes, I am. I didn't have much time to work on anything the last days, that's why I only posted there after I got the code working on Linux.

Regarding coil whine: I haven't done all the tests I wanted, but the PicoPSU-80 + 60W power brick combination in my E-350 system does not make coil whine noises on idle. I wanted to try different PicoPSU + power brick + mainboard combinations, but I didn't have the time so far.

Just as a data point: The 192W power brick uses a 4-pin connector (similar to the 12V P4-connector) which I think is not grounded either.

Edit: What I can tell you is that the power consumption at idle is a lot worse on Linux with the free video drivers. The lowest I've seen on Windows is 15W AC (with AMD Overdrive installed, which disables the Wifi and Bluetooth adapters), normal is about 23W AC on Windows and on Linux I get about 35W. (Arch Linux 64 bit, Kernel 3.8, radeon/ati open video driver; haven't tried with Catalyst)

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Re: Beating the PS3

Post by Rebellious » Fri Apr 26, 2013 5:25 am

So it's really NOT grounded! What's the solution? Plug a regular power cord into the mains and connect the ground wire to the chassis?

The i5 3570K idles at 18W both Windows & Linux (with the optimized Intel video driver which is built into the kernel). It runs full screen Adobe flash with ~5% CPU utilization.

On my A8 laptop, the AMD proprietary driver with catalyst (and Power Play disabled) makes a huge difference. Your instructions on the amdmsrtweaker-lnx may be beyond the reach of average users, is there a way you could pre-compile the executable so people can download it just like k10ctl?


BTW, I took your advice and used the "magic" Samsung memory

Image

johannes
Posts: 53
Joined: Tue Oct 23, 2012 8:13 am
Location: Suisse Romande

Re: Beating the PS3

Post by johannes » Sat Apr 27, 2013 8:55 am

I added statically compiled binaries for 32bit and 64bit Linux. See the readme for links and instructions.

Regarding grounding: Since the PicoPSU is not designed to be grounded, I don't think it's a safety issue. (For instance the PS3 or my cheap 2.1 loud speakers are not grounded either). In my case both power bricks are grounded - the 60W one uses the "mickey mouse" c5/c6 connector and the 192W uses a c13/c14 one, just as normal computer PSUs do. I know little about electronics, but I suppose 12V DC doesn't have the same need to be grounded as 110/230V AC.

Your cooler looks huge! I guess the fan is only 120mm, but on top of the little mini-itx board it looks enormous. And it appears that there's not much space for the RAM next to the PicoPSU and the cooler - good thing the Samsung RAM is available in your part of the world!

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Re: Beating the PS3

Post by Rebellious » Sun Apr 28, 2013 11:34 am

This is what I get with my A8-4500M, probably similar to your A10

Code: Select all

AmdMsrTweaker v1.1

.:. General
---
  AMD family 0x15, model 0x10 CPU, 4 cores
  Default reference clock: 100 MHz
  Available multipliers: 1 .. 19
  Available voltage IDs: 0 .. 1.55 (0.00625 steps)

.:. Turbo
---
  enabled
  locked
  Max multiplier: 28

.:. P-states
---
  8 of 8 enabled (P0 .. P7)
  Turbo P-states: P0 P1
  ---
  P0: 23x at 1.1375V
      NorthBridge in NB_P0
  P1: 23x at 1.1375V
      NorthBridge in NB_P0
  P2: 19x at 1.025V
      NorthBridge in NB_P0
  P3: 18x at 0.975V
      NorthBridge in NB_P0
  P4: 17x at 0.925V
      NorthBridge in NB_P0
  P5: 16x at 0.9V
      NorthBridge in NB_P0
  P6: 14x at 0.875V
      NorthBridge in NB_P1
  P7: 9x at 0.875V
      NorthBridge in NB_P1
  ---
  NB_P0: 16x at 1.075V
  NB_P1: 13x at 1.05V
  NB_P2: 11x at 1.025V
  NB_P3: 8x at 0.875V

P7 is unused. I guess you can't force the kernel to use P7, so you'd have to modify P6 to get the machine to idle at 900MHz? Might extend battery life by an hour? But "amdmsrt-amd64 P7" works, cpufreq-aperf shows 893000 KHz on all 4 cores.

. . .
I'm not sure whether grounding is needed here. When you touch the case, you can deliver a static discharge of several thousand volts. Without ground that can damage the motherboard and all components connected to it. Until I know for sure, I've temporarily installed a ground wire to the case.

johannes
Posts: 53
Joined: Tue Oct 23, 2012 8:13 am
Location: Suisse Romande

Re: Beating the PS3

Post by johannes » Mon Apr 29, 2013 1:33 am

P7 is unused. I guess you can't force the kernel to use P7, so you'd have to modify P6 to get the machine to idle at 900MHz?
I think so. I will try to have a look at this the coming days on my A10.
Might extend battery life by an hour?
Unlikely. Your CPU should be in C1E power states (wiki, if I remember correctly AMD CPUs are set to C1E and will automatically enter C2/C3 as appropriate.) and mostly powered down. You can of course still reduce the frequency and voltage for P6 and there will be some improvement, but I wouldn't expect one hour. (Fun fact: You can reduce the frequency to 100 MHz and enjoy a really slow computing experience.)

powertop is a great tool which can help you reduce your power consumption on Linux by allowing the CPU to sleep longer (i.e. be in above mentioned C-states) and activating further power saving features. Sleeping longer is due to less wake-ups - if you for instance have a tool running in the background which does some small amount of work 100 times a second, the CPU has to wake up at least 100 times per second. Shutting down the tool or reducing the frequency might give you a longer battery time.

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Re: Beating the PS3

Post by Rebellious » Thu Jun 27, 2013 6:34 am

I tried the PicoPSU 150XT kit with a more power-hungry AM3 Phenom II system and there is some whine at high load. It appears to come from the power brick.

Image

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Re: Beating the PS3

Post by MikeC » Sun Dec 08, 2013 8:18 am

picopsu grounding: AFAIK, if the power brick uses a 3-prong grounded plug, then the picopsu should be grounded even if the DC output only uses 2 wires, as the negative lead is usually the ground, and every motherboard almost always has ground connections to the chassis, even if only through the metal I/O plate. This is a guess. I can check later with a multimeter.

Interesting project, johannes. If I was facing your fan-ramping problem, I'd be tempted to force a bigger heatsink into the M350. A Scythe Big Shuriken2 would make short work of all your heat issues with very little noise if the board and case will allow it to fit with the cover off. If it did fit that way, I'd cut a hole out from the cover just big enough for the fan and/or top of the SBS2 to stick through. Smooth out edges with a rasp, use good black electrical tape to hide the edge so you don't waste too much time with the cosmetics. The fan & top of heatsink fins might stick out by 10-15mm maybe (the SBS2 is 58mm tall, same as the external dimension of the M350), but that look would be OK, to me... though you might want to stick a wire grill on the fan. The idea is sort of like the Antec ISK100 (our sample of which was not great due to some design and mostly poor fit & finish issues).

Post Reply