MUSINGS / BUILD: HTPC Rebuild... Skylake / 4K (Part II)

Okay, after laying out the foundation in Part I, let's rip open the boxes and get building the new HTPC!

Right... So I'm reusing the old Antec ATX tower case from 2003(!) for this build. It used to house an old Intel Core 2 Quad Q6600 CPU/board you see on the floor there. I put that computer together back in 2007 and it had been in constant use until 2 weeks ago so I really got my money's worth after 8 years! What I'm really impressed by is that there was still one of the early-model 1TB Maxtor HDs in there in good shape. Time to retire this board in any case... Maybe I can still put it to service at some point for my kids.

And here is the new Intel i5-6500/Gigabyte Z170X-Gaming 7 assembled with the CoolerMaster Hyper 212 Evo heatsink (and Corsair DDR4 RAM) living in the old Antec case. Blue SATA connector attached to the ADATA SP600 256GB SSD:

Here's a look at a few of the chips / features on the motherboard:

Creative Sound Core3D chip to the left. The driver for the built-in audio system is identified as Creative Sound Blaster "Recon3Di". The middle picture is Gigabyte's Turbo B-Clock module which basically allows the BCLK to be adjusted 1MHz at a time. Of significance for the overclockers out there. To the right is a little plastic adaptor they supply for the typical case connectors like the Power LED, Reset Switch, Power Switch, HD LED, etc... Works well especially for an old case like this where the connectors are not well organized; no more fiddling with these little connectors or mess up the polarity. Just plug them all into this part and connect to the motherboard en bloc. Wish they thought of this years ago!

It's alive! Notice the red LED lighting up the ports panel out back. You can change the color in the BIOS to suite your taste / decor / pimping style (blue, green, light green, yellow, white, and thankfully can turn it off)... You can even have it pulsate to music :-).

Of course with no fans going this is completely silent and getting into the BIOS to tune settings was no issue at all. As usual, when trying to find the optimal settings whether overclocking or underclocking does take a fair bit of trial-and-error. Each CPU will have different tolerance in terms of how fast it can be pushed (both in terms of absolute limit and heat tolerance) or how much stability it can maintain with little voltage. Remember, my main goal is to find settings that keep power utilization low so as not to overwhelm the fanless (and not particularly massive) heatsink. Despite trials of many failed settings, at no time did I run into any difficulties requiring that I clear the CMOS - the motherboard detected failures to boot and will get me back to the BIOS screen on the next attempted reboot... Great to see how with each new generation of hardware, things have become easier and more fool-proof for the build-it-youselfers.

BIOS Underclock & Undervolt

It's good practice to get the latest stable BIOS upgrade for the motherboard (I'm running the "beta" F6f which seems quite stable).

Alright, without further ado, here are the final settings I'm going to run with (BIOS screenshots of relevant changes with red arrows):

I'm unclear if setting the "Package Power Limit" to 28W as a target actually makes a difference...

With these settings, I'm running the machine at about 75% of stock speed. Basically a Core i5 (quad core, non-hyperthreading) running 2.4GHz with a small amount of "Turbo Boost" leeway up to 2.7GHz when running non-multithreaded programs (not that this really makes a noticeable difference IMO). Notice the CPU is also undervolted to 0.8V now from around 1.1V stock. (Using similar settings, I actually was able to run this stably at 2.5GHz but I figure it's best to be a little safer especially for the summer when ambient temperatures would be higher.)

Here's what CPUID shows us:

Running Prime95 at the time. Note core voltage "droop" under load to <0.8V.

Power Utilization

Power savings are substantial. In fact as a comparison to my previous Pentium G3220 HTPC, here are my Kill-A-Watt measurements for power usage from the computer:

Pentium G3220 (at 2.8GHz, undervolted)
     Idle: 28W
     Prime95 Torture Test:
          2-core blend mode: 44W

Intel i5-6500 (at 2.4GHz, undervolt)
     Idle: 27W
     Prime95 Torture Test:
          2-core blend mode: 38W
          4-core blend mode: 48W

Nice! Basically going from the old HTPC build to the new one barely increases power consumption when idle and doing CPU computation. In fact, I'm able to run all 4 cores (with no thermal throttling) achieving much faster overall speed with just a +4W increase!

Quick Speed Comparison

"How much faster?" You may ask...

Prime95 benchmark - 4096K FFT

Pentium G3220 machine (dual core, @3.0GHz, no undervolt):
     1 CPU, 1 Worker: 68.97ms - 14.5 iter/s
     2 CPU, 2 Worker: 79.65 / 78.75 - 25.25 iter/s

Core i5-6500 machine (quad core, @2.4GHz, undervolted):
     1 CPU, 1 Worker: 22.6ms - 44.2 iter/s
     2 CPU, 2 Worker: 23.68 / 23.41ms - 84.95 iter/s
     3 CPU, 3 Worker: 27.32 / 27.38 / 27.39 - 109.65 iter/s
     4 CPU, 4 Worker: 37.62 / 37.46 / 43.01 / 34.47 - 105.54 iter/s

As you can see, the Core i5 microarchitecture embarrasses the Pentium in speed (as expected). Single core total throughput is 3x that of the Pentium even at slower clockspeed. This rises to 3.4x with dual workers. Notice that the total throughput with this benchmark is highest with 3 workers  (4.3x over the peak G3220 performance) possibly a result of memory bandwidth limits once we hit 4 threads.

Of course Prime95 is just a single datapoint in terms of all the potential benchmarks out there... But it does give us a look at floating-point performance which likely corresponds to actual DSP processing (I think the JRiver DSP operates at 64-bit floating point).

Finally, before really declaring everything as being ready for operation, it's time to...

Torture Test

So, are we truly stable undervolted and run off a passive heatsink?

Test 1 - IntelBurnTest

One of the most insane tests around! Usually will fail with unstable machines within 5 minutes even with "Standard" setting. This machine was stable for 10 runs at "Very High" settings without evidence of thermal throttling which will cause speed deterioration. Note the consistent 70-71 GFlops result. (As a speed comparison, the old Pentium G3220 HTPC scored 21-22 GFlops, so the new underclocked i5-6500 machine is 3.3x faster.)

Test 2 - Prime95 for 24 hours

There you go, 25 hours, 55 minutes. No errors or warnings for all 4 cores. CPU temperature based on the motherboard sensor was running around 60-65°C.

Test 3 - Prime95 + FurMark

Now this is punishing! Remember that the total "TDP Budget" for this chip is shared between the CPU and integrated GPU components.This test will challenge stability at full power consumption and heat production. In fact, when running this combo, my Kill-A-Watt is measuring 60W power consumption for the whole machine. After about 5 minutes the machine will throttle down due to heat (CPU starts to slow down once it reaches 72°C I think). Notice the CPU-Z window is telling us that the CPU is running at 1.8GHz at this point after 45 minutes of FurMark.

Given that I do not intend to do 3D gaming on this machine, I'm not concerned. The important thing is that even with all the heat, the machine did not crash and there were no graphical glitches showing up with the OpenGL 3D furry torus. Also of no surprise is that the 1280x720 rendering with 2xMSAA is only running at 8fps. These integrated GPUs could be reasonable for casual gaming with low complexity 3D graphics I suppose. DirectX12 is supported by Skylake so features like lower "draw call overhead" could at least improve performance in DX12 optimized games.

Note that the "spread" in power consumption between sitting idle and with Test 3 is 60-27 = 33W for the passive heatsink to dissipate with the settings I chose.

So how well does this machine run?

So, there it is sitting in back propping up my absorption panel; shiny and black like the rest of my gear :-).

Subjectively, how does the new HTPC feel? Why, fast, of course :-). Speedy boot up and program start-up, snappy GUI (a given). But what really matters is whether the higher processing speed allows me to do more...

Audio:
In JRiver, I can now do this with no crackles, pops or other errors:

Upsampling to 384kHz plus applying 131k taps Acourate digital room correction filter feeding my TEAC UD-501. Previously with the Pentium G3220, this resulted in occasional audible distortion especially when crossfading between tracks adding to the processing load. Notice the CPU meter top right is only registering 7% usage total... Nice!

Beyond this, honestly, digital audio doesn't need much processing power (for the audiophile home listener). Also, realistically upsampling to DXD rates is pretty silly and sounds no different unless one has a neurotic distrust of one's DAC's internal oversampling digital filter algorithm (probably should have bought another DAC then!) or is a NOS DAC and could use some antialiasing.

Video:
This is where the HTPC upgrade really makes a difference... Remember, I'm getting ready for 4K.

One problem is that there is little 4K content to actually watch, and the most popular at this time are just streaming 4K clips off YouTube; basically just demos like this (for other demo content - see here):

The above is an example played back with Microsoft's Edge browser streamed as H.264/AVC 30fps. Total CPU usage of 12%. Excellent. It's not CPU speed that's causing the occasional stutter, it's the Internet speed with my cable modem especially during peak hours. Note that currently Internet Explorer and Edge browsers makes efficient use of hardware video decoding. Sadly, and ironically, Google's own Chrome appears extremely inefficient or does not appear to be using the hardware decode feature.

I do also have NetFlix but not the "4 screens + UltraHD" plan which adds 4K streaming for the time being.

The rest of these quick tests are with the few 4K video files floating around the net. These sample files are all played off my central server computer off my home gigabit ethernet (setup described previously).

Let's start with 4K encoded with the most common CODEC in use these days for video downloads - H.264/AVC. Latest Intel graphics driver installed (Beta 15.40.12.4326). Playback software being MPC-HC x64 player with LAV Filters x64, hardware acceleration through DXVA2 (copy-back). The 32-bit x86 LAV Filters I found to be not as good with frame rates:

This is a high bitrate file at 100Mbps. Notice 0 dropped frames, 1ms jitter, <20% CPU usage... Can't really ask for much more!

But it's the next-generation CODECs we also need to think about to be reasonably "future-proof"... The claim being made is that these new formats will provide equivalent quality at almost half the bitrate.

Let's start with Google's VP9 using a test file found here.

And here's another VP9 video downloaded off YouTube using 4K Download... Original video here:

Qualitatively smooth 24/25fps playback. No dropped frames. CPU use around 20-30%. Bitrate relatively low though at ~14-15Mbps for this file.

And there's H.265/HEVC:

Again, really good performance. This is a 30fps video file, 25Mbps. 0 dropped frames, 1 ms jitter. Like the VP9 samples above, CPU usage around 20-30% with hardware acceleration.

For comparison, check out the results using my i7-3770K machine for the same file but without the benefit of hardware video decoding:

Video quality is still reasonable although I did notice a bit more video "judder" in some parts with more activity and complexity. Notice the amount of CPU usage - 86% with a more power i7 CPU running at 3.5GHz. This high amount of processing demand doesn't provide much head-room at all when the scene gets complex and triggers an increase in processing latency (thus the judder at times).

To end off... Hardware accelerated 4K/30Hz is a reality with Skylake for excellent playback of the latest VP9 and H.265/HEVC CODECs. But with HDMI2.0 specification, we could see videos released at 50/60fps.

At 60fps, H.264/AVC demos have been OK so far - here's Big Buck Bunny (2008, AC3 multichannel audio):

Notice that after 8 minutes of video, there were 31 dropped frames (remember, this is 60fps). Not perfect but also not noticeable when watching. 17% CPU utilization at this instance but varying between 20-35% usually. Also notice the wider audio-visual "Sync Offset". It's not a high bitrate video though.

Already there are a number of video files out there based on Samsung's UHD HEVC demos at 4K/60Hz. As far as I can tell, the i5-6500 Skylake is not currently able to decode these 60fps smoothly at stock speed with LAV Filters 0.67 x64. VP9 at 4K/60Hz is significantly better but there's still judder, some dropped frames, and audio stutter as video catches up. HEVC is still young and I'm sure many code optimizations and driver updates are to come (anyone know if commercial decoders like UHDecode Upgrade improves performance with 60fps?)... Furthermore, other than the Hobbit movies in 48Hz "High Frame Rate" (HFR), I really don't know of many others following suit (maybe James Cameron's Avatar sequels). Would have been nice to see faultless 60fps playback, but not exactly essential at this time in history.

BTW: Here's an interesting read on visual physiology, resolution, and touching on the debate between "cinema" 24fps vs. HFR. First batch of Sony UHD Blu-Rays likewise are not likely to impress...

Well, I guess the only thing left to do now is to actually get myself a 4K TV downstairs in the media room :-). Unfortunately, I will also need a 4K/HDMI2.0/HDCP2.2 receiver to make it all fully ready for UHD Blu-Ray. Great to see 75-80" UHD TV sets finally dropping below CAD$5000 now. I'm unsure if I want 3D capability before plunking down the cash.

I'll be looking out for deals over the holidays in any case...

-----------------------

Last thoughts before signing off... I don't think it's much of a stretch to say that 4K/UHD Blu-Ray will likely be the last major disk-based delivery medium for video. Probably similar to what we've seen in the audio world, "ultra" high resolution formats like SACD and DVD-A will likely not be adopted at the same rate as an already excellent resolution format like CD before (in the case of video, 1080P Blu-Rays). I do expect that UHD Blu-Ray will do much better than the floundering SACD format and dead DVD-A simply because the image quality improvement will be obvious in appropriate situations rather than hi-res audio where benefits are at best vague assuming the quality of the audio mastering is actually of high-resolution caliber to start.

Streaming video is clearly the future and physical formats will more than likely be aimed at the videophile collectors who just needs to have the "best" version. I think it's interesting that in the audiophile world, the "best" physical version to own as a collector is often the vinyl record due to size of artwork and packaging (but not as the undisputed, "highest fidelity" format) whereas potentially in the video world it likely will be the 4K/UHD Blu-Ray disks and packaging which will be sought after plus objectively and subjectively provide the best audio-video quality.

I look forward to 2016 as a big year for the videophiles and HT geeks... Enjoy the videos and music everyone as we head into the holidays!