It is currently Mon Dec 22, 2014 5:41 pm

All times are UTC - 8 hours




Post new topic Reply to topic  [ 21 posts ] 
Author Message
 Post subject: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Sun Aug 21, 2011 10:06 am 
Offline
*Lifetime Patron*

Joined: Sun Jan 04, 2004 11:35 am
Posts: 536
Location: Chicago, Ill., USA
Hi all,

Last weekend I completed my new home theater PC. It runs MythTV. Specs:

CASE: Lian Li PC-Q11B
MOBO: Asrock H67M-ITX
CPU: Intel i5-2500k
PSU: picoPSU-160-XT w/150W Brick
RAM: Kingston KVR1333D3N9K2/8G
FAN: NoiseBlocker PK-3
GPU: ZOTAC ZT-40601-20L GeForce GT 430
HSF: Thermalright AXP-140 RT w/TY-140 fan
SSD: Kingston 64 GB
HDD: Samsung Green 2 TB

Not as pretty as josephclemente's q11b, and I have to admit I used his build as inspiration.

In fact, this is one of the more frustrating builds I've done, as the original plans called for an almost completely different set of components, and I went way over budget. Original case was the Apex MI-008, original motherboard was the Biostar TH61, then the Intel DH67CF, original HSF was the Noctua NH-U9B SE2. Original plan was to use the built-in Sandy Bridge GPU. I've got a big writeup on why the original plans fell through; I'll try to post that in a day or two.

Anyway, in the meantime, here's a pic. Though I went over budget and experienced some headaches, the finished product works great and is very quiet. It's not silent, but it sits behind our entertainment center, so it's "silent" for all practical purposes.

Image
cesspool - new mythtv/htpc

Without the discrete GPU and only the SSD, idle power consumption is around 24--25 Watts AC. With the HDD and GPU, it's around 35. I undervolted the RAM a bit, down to 1.400 Volts, and that seemed to shave about a Watt off the idle power consumption. Frankly, I'm a little disappointed in the idle power usage - josephclemente's nearly identical build idles at 26 Watts (without the discrete graphics) on a Seasonic X-560. I expected the picoPSU to have dramatically better efficiency at such low loads, but apparently it's not all that great.

For what it's worth, since I went through three motherboard iterations, the worst was the Biostar, which idled at 28 Watts. The best was the Intel DH67CF, which idled at 22 Watts.


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Sun Aug 21, 2011 11:06 pm 
Offline
*Lifetime Patron*

Joined: Thu Feb 13, 2003 6:53 am
Posts: 1544
Location: Sweden
The power consumption is a little high. DH67CF and 22W? Is that on MythTV? Either your software or your brick is the cause I think. I get about 18W on Win7 but about 21W on Xubuntu.

_________________
Main: ASRock B85M-ITX | i3-4330 | 16GB DDR3 | Intel 730 240GB | HDPLEX H1-S | picoPSU | No moving parts | Idle 13.9W
HTPC: ASRock H81M-ITX | Pentium G3420 | 4GB DDR3 | Intel 520 120GB | HDPLEX H1-S | picoPSU | No moving parts | Idle 11.2W
Gaming: Intel DH77EB | i5-3570K | GTX 750 Ti | 16GB DDR3 | Intel 520 120GB | TJ08-E | G-360 360W
Server: Intel DH77DF | i3-2100T | 4TB+3x3TB | picoPSU | Idle 24W AC


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Mon Aug 22, 2011 7:27 am 
Offline

Joined: Mon Jul 16, 2007 2:35 pm
Posts: 808
Location: Greece
It could be the software indeed. I see considerably higher power consumption under linux, though it is a fairly outdated version (ubuntu 10.4).

_________________
My PCs: Workstation : HTPC in a CD player case : Custom Mini ITX projects


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Mon Aug 22, 2011 7:41 am 
Offline
*Lifetime Patron*

Joined: Wed Aug 09, 2006 6:10 pm
Posts: 122
Location: Columbus, Ohio
Actually, the newer Linux kernels have been getting worse with power consumption.
(see http://www.phoronix.com/scan.php?page=a ... ress&num=1 )
I'm sure they are going to eventually fix this problem, but I'm in no hurry to upgrade
from Ubuntu 10.10 to 11.04 myself. (With 10.10 I'm running kernel 2.6.35-30). I might
just do a fresh install after, oh, 12.10 comes out :D

_________________
E7500, Zotac GF9300-I-E, Lian-Li Q03b, picoPSU-150-XT, 4GB DDR2-800, 128 GB Samsung 830 SSD, 500GB 2.5" Hitachi Travelstar, XUbuntu 14.04 64-bit
My blog ...


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Mon Aug 22, 2011 9:22 am 
Offline
*Lifetime Patron*

Joined: Sun Jan 04, 2004 11:35 am
Posts: 536
Location: Chicago, Ill., USA
Vicotnik wrote:
The power consumption is a little high. DH67CF and 22W? Is that on MythTV? Either your software or your brick is the cause I think. I get about 18W on Win7 but about 21W on Xubuntu.


I tested the power consumption of all those boards under both Windows 7 and Linux (specifically, CentOS 6.0). The idle power consumption was the same for both. I didn't do any tweaking of either, basically, just let it idle for five minutes or so after a default install.

Edit: note that I didn't yet have MythTV installed when I took the readings under Linux.


Last edited by matt_garman on Tue Aug 23, 2011 5:05 am, edited 1 time in total.

Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Mon Aug 22, 2011 9:36 am 
Offline
*Lifetime Patron*

Joined: Sun Jan 04, 2004 11:35 am
Posts: 536
Location: Chicago, Ill., USA
For what it's worth, I'm using this 150W AC-DC adapter from Mini-Box.com. It's made my Edac. I can get the model number when I get home.

Given that I had the same idle power usage under both Win7 and Linux, I'm guessing that maybe it is the power brick. The "benchmark" for low power seems to be the missing remote dh67cf + i3-2100t built at 14 watts. That used the picoPSU-XT-150, which has to be practically the same as the XT-160 I have (maybe not, but that seems like a reasonable assumption). The MR build used a 120W AC-DC brick, although I didn't see where it said exactly which one.


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Mon Aug 22, 2011 9:50 am 
Offline
*Lifetime Patron*

Joined: Thu Feb 13, 2003 6:53 am
Posts: 1544
Location: Sweden
I rant a bit about Edac in this post.. I've only tested one Edac brick but the one I own (model EA11203A, 12V 8.5A) is quite inefficient.

_________________
Main: ASRock B85M-ITX | i3-4330 | 16GB DDR3 | Intel 730 240GB | HDPLEX H1-S | picoPSU | No moving parts | Idle 13.9W
HTPC: ASRock H81M-ITX | Pentium G3420 | 4GB DDR3 | Intel 520 120GB | HDPLEX H1-S | picoPSU | No moving parts | Idle 11.2W
Gaming: Intel DH77EB | i5-3570K | GTX 750 Ti | 16GB DDR3 | Intel 520 120GB | TJ08-E | G-360 360W
Server: Intel DH77DF | i3-2100T | 4TB+3x3TB | picoPSU | Idle 24W AC


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Mon Aug 22, 2011 9:51 am 
Offline

Joined: Mon Jul 16, 2007 2:35 pm
Posts: 808
Location: Greece
A comparison of power bricks with similar ratings (100-120W) shows miniscule differences: http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story3&reid=207

Perhaps the larger Edac unit is less afficient at low loads?

Other possible explenations that come to mind: the fans using a little more power than usual. The large amount of installed ram could also play it`s part.

_________________
My PCs: Workstation : HTPC in a CD player case : Custom Mini ITX projects


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Mon Aug 22, 2011 11:49 am 
Offline

Joined: Tue Nov 02, 2004 12:22 pm
Posts: 2064
Location: Guatemala
ntavlas wrote:
Perhaps the larger Edac unit is less afficient at low loads?
Could be, but in my build using 150W brick and it idles at 15W, i think its a combination of the CPU (2500K even though it drops to 1600 as the 2100T but it might have different voltages), the fans (although im using 3 fans), perifericals attached (i idle at 14W without keyboard, mouse or monitor) and the mobo.

_________________
GameMi >> MSI Z87-GD65 | Intel Core i7-4790K | Thermalright Silver Arrow IBE + 3x Noctua NF-A15 PWM @500rpms | Crucial Ballistix Sport 32GB DDR3 1600 | nVidia GTX780 + ARCTIC Accelero Xtreme IV | Samsung SA850 27'' 2560x1440 | Samsung 840pro 512GB | Hitachi 7K1000 1TB | Fractal Design Define R4 + 4x Noctua NF-A14 PWM @450rpm | SeaSonic SS-860XP2
Other builds ---> ServeMi | CamMi | MiniMi | HTPCMi


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Mon Aug 22, 2011 2:08 pm 
Offline

Joined: Thu Jul 03, 2008 4:27 am
Posts: 1730
Location: Switzerland
Are you even using the same measurement device? Have they been calibrated?

Other than that, I would first suspect different efficiencies among power bricks, especially at low loads. Nevermind tests which tested different bricks, especially if they don't have reliable instruments or methods.

Then there are a lot of small differences: 110V vs 220V mains, peripherals, sample variance, drives, BIOS settings, RAM and so on.

Nevermind the voltage: the 2500K has 4 cores and the 2100T has 2. They may also have different GPUs... the IGP does burn power, you know. If not, they must have different features.


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Tue Aug 23, 2011 4:31 am 
Offline
*Lifetime Patron*

Joined: Sun Jan 04, 2004 11:35 am
Posts: 536
Location: Chicago, Ill., USA
Here's the writeup I alluded to in my original post...

This build was somewhat of an exercise in not thinking things through and going over budget. My original goal was: a new HTPC that is as small, quiet, and low-power as possible. Specifically, I use MythTV on Linux as a HTPC platform. I thought I could do everything with an Intel Sandy Bridge CPU, given their low power consumption and high performance.

My original plans were:
CPU: Intel i3-2100 -OR- Intel i5-2500k
MB: Biostar TH61
Case: Apex MI-008
GPU: Built-in Intel HD 2000 or HD 3000
CPU Heatsink: Stock, then Noctua NH-U9B SE2
PSU and RAM: same as final

First problem: I chose the i5-2500k, not for overclocking, but for the HD 3000 graphics. For HTPC, HD 2k vs 3k theoretically shouldn't make a difference, but MythTV uses OpenGL for some stuff, and the HD 3k should defnitely be better for 3D applications. That in mind, perhaps the i3-2105 is the best CPU, but I was buying from Microcenter, and they don't carry that chip.

Because I went with the faster/4-core CPU, I found that the stock cooler was too loud (I suppose it could also be too loud for a lesser CPU). My choices for a CPU cooler in this situation are very limited: basically, I needed something that's no wider than the socket itself, given its proxmity to the PCIe slot. (Note that my original plans didn't call for a GPU, but I was thinking of maybe using a Broadcom Crystal HD card, so I wanted to keep the PCIe slot unobstructed.)

I couldn't find anything that suited my needs, although in hindsite I think the Thermalright AXP-140 would have worked. So I went with the Noctua NH-U9B SE2, which was too tall for the Apex MI-008 case.

So now onto the second problem. The MI-008 case was too short for the new Noctua heatsink. The other problem was the lack of any case fan. In this site's review of the case, it was noted that a 120mm fan could be wedged into the side of the case. Unfortunately, this can't be done with a picoPSU on a motherboard whose power connector is right on the edge (technically, it might be possible, but to me it felt like the fan would put too much pressure on the solder points of the picoPSU).

I ordered the Lian Li PC-Q11B to replace the MI-008. (I was already familiar with this case because I used it for a new PC build for my parents.)

Now it looked as though everything would be just fine, but I found out that Sandy Bridge graphics support on Linux is "bleeding edge". These days, I don't have too much time to mess with getting the software to work; not to mention the poor WAF (wife acceptace factor) of having an unstable HTPC. I wanted something that should "just work". SNB graphics on Linux can be done, and I started to go down that road, but it required too much tinkering to make it work, so I scrapped the idea.

The easiest and most common solution for HTPC graphics on Linux is nvidia, and right now the GT 430 seems to be the GPU to get for this purpose. Of course I wanted it silent, so I bought a passive GPU, specifically the Zotac. That brought on another problem: I didn't notice that the GPU heatsink wrapped around the back of the card, causing it to interfere with the Noctua CPU heatsink! So I ditched the Noctua in favor of the Thermalright (after seeing josephclemente's nice Q11B build).

Later, I saw on Newegg that Asus makes a passive GT 430 card that (at least from their picture) looks like it wouldn't occupy more than one slot's space---meaning it might have actually fit in the Apex MI-008 that I originally wanted to use!

The next problem was power consumption of the Biostar TH61 motherboard. Initially, when I planned the build, I wanted to use the Intel DH67CF, as anecdotally, Intel motherboards have the lowest power consumption. However, the Intel boards have a significant price premium. Ironically enough, I initially wanted to keep this build in the semi-budget realm, so I opted for the cheapest mini-ITX board that met my requirements, which was the Biostar. I was really disappointed in the power consumption though: 28 Watts at idle with the i5-2500k, 8 GB of RAM, SSD hard drive and picoPSU. With the picoPSU, I was really hoping for less than 25 Watts. Normally I wouldn't quibble over a few watts, but at this point, I was getting so frustrated with all my mis-steps that I decided to go ahead and send back the Biostar and get the Intel that I wanted in the first place.

And then I found out the hard way that the DH67CF has a huge chip on the back of the motherboard which interferes with the mounting bracket of the AXP-140. I posted about this here. Again I drew inspiration from josephclemente's build, and opted for the Asrock H67ITX motherboard.

At the end of the day, I wasted a lot of money: I over-bought on the processor (i3-2100 would have been fine), I ate shipping and restocking fees for two motherboards, I ate shipping fees on the case return, and right now the Noctua heatsink is sitting unused (although I do plan to use it for a future build).

If I were to start fresh with the experience I now have, I would have used the i3-2100 (or i3-2105) CPU, Thermalright AXP-140 heatsink, Asus passive GT 430, probably the Asrock H67ITX motherboard, and made the MI-008 case work (by making neccessary modifications to fit a case fan in there).


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Tue Aug 23, 2011 4:51 am 
Offline
*Lifetime Patron*

Joined: Thu Feb 13, 2003 6:53 am
Posts: 1544
Location: Sweden
Damn.. Well at least you got the experience. ;) Things can be tricky sometimes.

_________________
Main: ASRock B85M-ITX | i3-4330 | 16GB DDR3 | Intel 730 240GB | HDPLEX H1-S | picoPSU | No moving parts | Idle 13.9W
HTPC: ASRock H81M-ITX | Pentium G3420 | 4GB DDR3 | Intel 520 120GB | HDPLEX H1-S | picoPSU | No moving parts | Idle 11.2W
Gaming: Intel DH77EB | i5-3570K | GTX 750 Ti | 16GB DDR3 | Intel 520 120GB | TJ08-E | G-360 360W
Server: Intel DH77DF | i3-2100T | 4TB+3x3TB | picoPSU | Idle 24W AC


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Tue Aug 23, 2011 4:52 am 
Offline
*Lifetime Patron*

Joined: Sun Jan 04, 2004 11:35 am
Posts: 536
Location: Chicago, Ill., USA
HFat wrote:
Are you even using the same measurement device? Have they been calibrated?


I am using the same measurement device, a Kill-a-Watt. It hasn't been calibrated. I do have a second one, though, so when I get a chance I'll take some readings with it and see if the two read differently. Although IIRC, MikeC suggested that the Kill-a-Watt is less accurate for low power readings.

HFat wrote:
Then there are a lot of small differences: 110V vs 220V mains, peripherals, sample variance, drives, BIOS settings, RAM and so on.


I'm in the US, so I have 110V mains. I'm fairly sure I took the measurements without any peripherals attached, as I used the "poor man's KVM" when testing (in other words, I don't have a KVM, so I would just temporarily borrow the keyboard+video+mouse from my main system to do the install/boot/login on the new system, then move those things back when I got logged in). Then again, my memory is pretty bad, and now that you bring it up, I'm even less sure of myself. :)

My method for measuring idle power consumption is this: boot into the system, then wait a few minutes to make sure everything has "settled", then I just sit and watch the Kill-a-Watt for about a minute to make sure the reading is stable. As long as it stays within a two Watt range for the observation time, I call that idle power consumption.

However, one thing was definitely not consistent from test to test: the CPU heat sink. I always took measurements with the Lian Li case fan plugged in, and running at max speed. However, the Biostar I tested with the stock heatsink and the Noctua (dual 90mm fans using the "ultra low noise" adapters). Going from memory (read: huge disclaimer), I think the idle PC was the same between both heatsinks. With the Intel board, I only tested with the stock heatsink. And with the Asrock, I only used the AXP-140 and TY-140 fan.

HFat wrote:
Nevermind the voltage: the 2500K has 4 cores and the 2100T has 2. They may also have different GPUs... the IGP does burn power, you know. If not, they must have different features.


FWIW, during all the setup, testing and experimenting I did (i.e. all three motherboards), I took power consumption readings with two cores disabled. At load, it made a big difference (can't remember the numbers but I think it was around 30 Watts), but at idle, it made no difference. Even now, I have two cores disabled just to limit the max power draw (and heat output) of the system. As I suggested before, this CPU is way more than I need for a HTPC.


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Tue Aug 23, 2011 5:36 am 
Offline

Joined: Thu Jul 03, 2008 4:27 am
Posts: 1730
Location: Switzerland
Yeah, I've got no idea why people use such powerful parts for HTPCs. I'd simply use a Zacate board!

Do we know how core disabling is implemented? It may be that the disabled cores are powered and idling. There's no real need to implement a more drastic power-saving feature since few people disable cores and te idle pwoer consumption is pretty low.
Maybe you have really disabled two cores and can't measure it. But with good gear, you could measure the differences between CPUs as SPCR has done.
By the way, if you don't need good single-threaded performance or if underclocking doesn't affect turbo speed, I would recommend drastic underclocking in the BIOS instead of disabling cores: 4 underclocked cores should be more efficient at multi-threaded work.

I don't have decent measuring gear but I've used that ASRock board in a system with a single fan: the one on Intel's stock low-profile heatsink at 1500 rpm or so. The board didn't overheat so the power consumption should be reasonable. Out of curiosity, have you been able to determine how much power it draws compared to other boards?


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Wed Aug 24, 2011 6:10 am 
Offline
*Lifetime Patron*

Joined: Sun Jan 04, 2004 11:35 am
Posts: 536
Location: Chicago, Ill., USA
HFat wrote:
Yeah, I've got no idea why people use such powerful parts for HTPCs. I'd simply use a Zacate board!


I do have some rationale behind it. :) Four cores is way too many. But two fast cores is useful. One, for commercial flagging of broadcast TV. But more importantly, software decoding of bluray content. Most people use modern GPUs for this, and can get away with something like an ION or Zacate. But MythTV was really designed around the idea of software decoding of content; that is the primary use-case, and default configuration. Of course, it has great VDPAU support, but it's one more thing you have to set up. My hope was that I could get a fast enough processor that all content could be software-decoded, and thus have a simpler overall setup (no additional graphics hardware, less software setup and configuration). And based on what I read, you do need a really fast CPU to decode some bluray content, as they are not all created (or mastered) equally. Do a google search on "single sliced high bitrate bluray" to read more about it. That reminds me of why I went with the 2500k over the 2100---higher overall clock rate. I intended to disable two cores from the start; I basically wanted the (almost) fastest dual-core Sandy Bridge with HD3000 graphics.

Of course, all that rationale is rendered moot since I had to use the GT430. Still, I believe that eventually the Linux Sandy Bridge support will mature enough to "just work", and I can lose the nVidia card (and further downsize with the MI-008 case as well!).

Also, as far as Zacate is concerned: Linux support of AMD/ATI graphics has historically been weak. Maybe it's improved recently, but I haven't bothered to check. Last I looked, it was kind of a mess: you had the closed-source drivers provided by AMD, which provided one subset of the hardware's functionality; and then you had at least one open-source, community-provided driver that provided a different subset of functionality. And stability complaints about all of them.

I am a Linux enthusiast, and enjoy playing with it. But these days, I just don't have the time. I need things to "just work". And in my experience, a lot of stuff in the Linux ecosystem doesn't meet the "just work" requirement. A basic PC for web and email? No problem. But a sophisticated HTPC? Still requires some work or careful hardware selection.

HFat wrote:
Do we know how core disabling is implemented? It may be that the disabled cores are powered and idling. There's no real need to implement a more drastic power-saving feature since few people disable cores and te idle pwoer consumption is pretty low.


I don't know how it's implemented. :) It definitely lowers the max power consumption/heat generation, which is what I wanted for my particular situation. I was hoping for (but not counting on) additionally lowered idle power usage as well.

HFat wrote:
By the way, if you don't need good single-threaded performance or if underclocking doesn't affect turbo speed, I would recommend drastic underclocking in the BIOS instead of disabling cores: 4 underclocked cores should be more efficient at multi-threaded work.


I didn't spend a lot of time on it, but I did dabble a bit with underclocking. On the motherboards I tried, none would let me underclock below the CPU's lowest idle speed. And none of them let me undervolt. So the best I could do was fix the CPU at the lowest speed/power state, i.e. what it would do naturally when idle.

HFat wrote:
I don't have decent measuring gear but I've used that ASRock board in a system with a single fan: the one on Intel's stock low-profile heatsink at 1500 rpm or so. The board didn't overheat so the power consumption should be reasonable. Out of curiosity, have you been able to determine how much power it draws compared to other boards?


I don't have the equipment to measure the power draw of just the board. I only have a kill-a-watt, for measuring AC power consumption from the wall. I took measurements of three different boards with almost the same components---same CPU, same PSU, same SSD, same RAM, but different heatsinks. So I at least have a decent approximation of the three boards, with the Biostar being the worst, the Intel being the best, and the Asrock in the middle.


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Wed Aug 24, 2011 9:26 am 
Offline

Joined: Thu Jul 03, 2008 4:27 am
Posts: 1730
Location: Switzerland
matt_garman wrote:
you do need a really fast CPU to decode some bluray content, as they are not all created (or mastered) equally.

It might well be so. I will admit I don't understand the point of bluray to begin with...

matt_garman wrote:
I basically wanted the (almost) fastest dual-core Sandy Bridge with HD3000 graphics.

What difference does the HD3000 make? Isn't that only for 3D?

matt_garman wrote:
It definitely lowers the max power consumption/heat generation, which is what I wanted for my particular situation.
...
On the motherboards I tried, none would let me underclock below the CPU's lowest idle speed.

That's the point of underclocking: to lower the actual TDP. It's not supposed to help with idle power consumption (unless you're underclocking something other than the CPU).
What I'm saying is that, assuming a target temperature at load, you should get better multithreaded performance by underclocking than by disabling cores. Underclocking is also more granular (you only have 4 cores).


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Wed Aug 24, 2011 12:28 pm 
Offline

Joined: Tue Nov 02, 2004 12:22 pm
Posts: 2064
Location: Guatemala
matt_garman wrote:
I basically wanted the (almost) fastest dual-core Sandy Bridge with HD3000 graphics.
There was no need to go with 2500k, there are couple of cpus that do offer HD3000, Intel Core i3-2105 and Intel Core i5-2405S, i would probably have gone with the 2105 since you were planning on disabling 2 cores on the quad, and saved close to $100. Eithre way i dont see the need for HD3000, all HD playback can be handle fine with HD2000 (if you wish to read more about it go check AVS Official Sandy Bridge / LGA1155 for HTPCs Thread), the only situations that i a user might benefit is gaming, which still very low end, and encoding/transcoding with quicksync.

_________________
GameMi >> MSI Z87-GD65 | Intel Core i7-4790K | Thermalright Silver Arrow IBE + 3x Noctua NF-A15 PWM @500rpms | Crucial Ballistix Sport 32GB DDR3 1600 | nVidia GTX780 + ARCTIC Accelero Xtreme IV | Samsung SA850 27'' 2560x1440 | Samsung 840pro 512GB | Hitachi 7K1000 1TB | Fractal Design Define R4 + 4x Noctua NF-A14 PWM @450rpm | SeaSonic SS-860XP2
Other builds ---> ServeMi | CamMi | MiniMi | HTPCMi


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Wed Aug 24, 2011 2:29 pm 
Offline
*Lifetime Patron*

Joined: Sun Jan 04, 2004 11:35 am
Posts: 536
Location: Chicago, Ill., USA
HFat wrote:
What difference does the HD3000 make? Isn't that only for 3D?


That is my understanding, that the HD2000 vs 3000 is a matter of 3D performance. On Linux, 3D is done via OpenGL.

MythTV has several ways of rendering video. One of those methods is OpenGL. Also, MythTV can (optionally) use OpenGL for other functionality as well.

So my rational (for my original plans) was, I didn't know exactly how I was going to configure MythTV to work with Sandy Bridge. OpenGL was a possibility, so I wanted the best hardware possible for that scenario.

Of course all that ended up being moot, since I'm using VDPAU (nVidia) to do decoding and rendering.

HFat wrote:
That's the point of underclocking: to lower the actual TDP. It's not supposed to help with idle power consumption (unless you're underclocking something other than the CPU).
What I'm saying is that, assuming a target temperature at load, you should get better multithreaded performance by underclocking than by disabling cores. Underclocking is also more granular (you only have 4 cores).


I see what you mean now. But your suggestion is actually the opposite of what I want. I want fewer, higher-clocked cores; rather than more slowly-clocked cores.


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Wed Aug 24, 2011 3:10 pm 
Offline
*Lifetime Patron*

Joined: Sun Jan 04, 2004 11:35 am
Posts: 536
Location: Chicago, Ill., USA
Abula wrote:
matt_garman wrote:
I basically wanted the (almost) fastest dual-core Sandy Bridge with HD3000 graphics.
There was no need to go with 2500k, there are couple of cpus that do offer HD3000, Intel Core i3-2105 and Intel Core i5-2405S, i would probably have gone with the 2105 since you were planning on disabling 2 cores on the quad, and saved close to $100. Eithre way i dont see the need for HD3000, all HD playback can be handle fine with HD2000 (if you wish to read more about it go check AVS Official Sandy Bridge / LGA1155 for HTPCs Thread), the only situations that i a user might benefit is gaming, which still very low end, and encoding/transcoding with quicksync.


A major consideration for me was buying from Microcenter, as they have the lowest prices on CPUs. The downside is the selection isn't as good. In particular, at least the one by me doesn't carry any of i3-2100T, i3-2105, nor i5-2405S.

I got my i5-2500k for $179.99 ($197 after tax), which is considerably cheaper than anywhere else. On newegg, the i3-2105 is $140, so I really only could have saved $60. But that's just looking at the HD3000 issue. I explained in the post above why I wanted the HD3000 (MythTV can optionally use OpenGL for video rendering.) I actually read through that entire AVS thread on Sandy Bridge, but I didn't really see anyone talking about Linux or software decoding. On Windows, you can use SNB's built-in hardware decoding, and none of this matters. I wish it was that easy on Linux, but it just isn't the case at the moment.

I also wanted the fastest CPU clock speed. The i3-2100/2105 is 3.1 GHz, and no turbo boost. The i5-2500k is 3.3 GHz, and 3.7 GHz with turbo boost. With two out of four cores disabled, and doing a single-threaded task like software decoding, I should pretty much be guaranteed to get the turbo boost speed. So it ends up being a 600 MHz difference. (Of course the i7-2600k is still faster, but that's another $100, so the 2500 seemed to me a better value. On the other hand, 3.7 GHZ @ $197 works out to $53 per GHz, but 3.1 GHz @ $140 is $45 per GHz, which actually makes the i3-2105 a better value. But if I should ever need all four cores, then the value award (probably) goes back to the 2500k.)

The last question is, do I really need that speed? I dunno. What made err on the side of too much power is this thread from the MythTV mailing list. To summarize, someone is talking about how a particular bluray (Transformers 2) is encoded with a single slice for the first three chapters and that software-decoding on a Core i7 920 (2.6 Ghz) is choppy. Single-slice encoding, combined with the way the decoding libraries on Linux/MythTV are written means that only a single-core can be used. I.e., no parallel processing. So a single core from an i7-920 isn't enough to keep up. Perhaps a 3.1 GHz Sandy Bridge core would be enough, but for an additional $60 I get 3.7 GHz (and comparable TDP with two cores disabled).

Really, the more I think about this, the fundamental issue is the current state of Sandy Bridge support on Linux just isn't there---it puts up a lot of road blocks that the Windows people don't even have to worry about.
  • No support for SNB's built-in video decoding
  • Beta-quality SNB graphics drivers that require futzing to make work
  • Limitations in Linux's software decoding libraries (ffmpeg) that demand an uber-fast single core CPU
Using a VDPAU-capable nVidia card takes care of all those problems. But I'm hoping that Linux Sandy Bridge support matures enough in the future to not require a discrete graphics card. All those AVS guys are using Windows and basically building i3-2100 (or 2100T) systems that more or less "just work" because the OS and software fully support the hardware features that are important for a HTPC.

Maybe I should just use Windows? Perhaps, but that's more money for the license, plus I'm not familiar with any free software-based DVRs like MythTV (I'm not saying they don't exist, I just don't know about them, whereas I've been using MythTV for many years now). It's a tradeoff either way, and I just happen to have a lot of "mental inertia" behind the Linux path.


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Wed Aug 24, 2011 3:20 pm 
Offline

Joined: Tue Nov 02, 2004 12:22 pm
Posts: 2064
Location: Guatemala
Good points, i think you made a good decision after explaining all related to your situation. But to much trouble into linux from what i read from your experience, i would just bought win 7 :)

_________________
GameMi >> MSI Z87-GD65 | Intel Core i7-4790K | Thermalright Silver Arrow IBE + 3x Noctua NF-A15 PWM @500rpms | Crucial Ballistix Sport 32GB DDR3 1600 | nVidia GTX780 + ARCTIC Accelero Xtreme IV | Samsung SA850 27'' 2560x1440 | Samsung 840pro 512GB | Hitachi 7K1000 1TB | Fractal Design Define R4 + 4x Noctua NF-A14 PWM @450rpm | SeaSonic SS-860XP2
Other builds ---> ServeMi | CamMi | MiniMi | HTPCMi


Top
 Profile  
 
 Post subject: Re: new htpc - q11b, i5-2500k, axp-140, gt430, picoPSU
PostPosted: Tue Sep 06, 2011 9:18 am 
Offline
*Lifetime Patron*

Joined: Sun Jan 04, 2004 11:35 am
Posts: 536
Location: Chicago, Ill., USA
For what it's worth, I did a "poor man's calibration" of my two Kill-A-Watt meters. My wife has an old lamp with two 25-Watt incandescent bulbs. Both Kill-A-Watt devices took the same measurements:

- With either 25 watt bulb turned on, the Kill-A-Watt read 24 Watts
- With both bulbs on, the Kill-A-Watt read 49 Watts

I purchased these meters several months apart; their serial numbers differ by about 5000. The second letter in the serial number is different as well (SG for one and SJ for the other). Also, the font used for the label on the back is different between the two units. I suspect they came from different factories, or perhaps are different revisions.

Both took the same measurements for the system I built, and based on the "calibration", it seems they should be reasonably accurate, at least within a watt or two.


Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 21 posts ] 

All times are UTC - 8 hours


Who is online

Users browsing this forum: No registered users and 2 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group