Official Sandy Bridge / LGA1155 for HTPCs Thread - Page 64

My Onyx 64GB has worked flawlessly also - so I never expected the reliability of SSDs to be called in to question so much. It really surprised me.

I hadnt heard about it until I started looking at "whats the best Sata3 SSD to get". When I started looking at the various support forums (OCZ, Crucial, Corsair, Patriot, etc etc), I was shocked to see just how many people have issues with the Sandforce 2281 based SSDs. Anandtech even has an article about it ("the infamous SF-2281 BSOD bug"). http://www.anandtech.com/show/4712/t...er-with-fw0009

The problem doesnt seem to affect (although not sure of this) SSDs that use a different controller. For instance the drives you mention do not use the 2281 AFAIK - Crucial M4, Intel 510, Plextor and Corsair Performance Series.

The issue seems to be confined to 2281 based SSDs (and there are a lot of them now).

None of the vendors seem to have a clue about what is wrong, what is the root cause (many are blaming it on the Intel Chipsets), why some have issues and others do not, and what they can do about it.

So since nobody here has seen it happening, just wondering if anyone is running a 2281 based SSD on sata3 and had any F4 BSODs?

I had serious BSOD issues with an OCZ Agility 3 build on a Gigabyte Z68 with Intel I3 2100. I swapped the SSD with Newegg and have been stable since. OCZ has since issued one more firmware update. I would stick with Intel 510 SERIES or Crucial until the Sandforce issues are fully worked out. More $, BUT BSOD issues are a time sink. Maybe noone talks about it here because the deep BSOD discussions are on the OCZ and Windows 7 forums.

From what I have read, it seems that the Z68 and P67 tend to be worse for the BSOD. Im just wondering if it also affects the H67 or H61?

Ive read about people getting these BSODs like every couple of hours. I would hari kari if my machine was doing that :-)

Correct, OCZ just released 2.13 for their Series 3 drives. So far it has made it better for some and worse for others. Tells me they still dont really know what is going on...

Did you swap it for a replacement Agility 3, or another model?

If you are interested in using madVR with Celeron G500 Series/Pentium G600/G800 Series + Intel HD Graphics or Core i3-2100/2120/2130/Core i5-2300-2500 + Intel HD Graphics 2000, here is a successful configuration.

- Processor: Any SNB processor with Intel HD Graphics (every Celeron/Pentium SNB) except for G440 single core, or Intel HD Graphics 2000 (Core i3-2100/2120/2130/Core i5-2300-2500). In the test below I used Celeron G530 2C/2T 2.4GHz, the lowest-end dual-core SNB processor.
- Memory: DDR3-2133 2 x 2GB @2133MHz. For example, G.SKILL F3-17000CL9D-4GBXL, $55.
- Z68 chipset mb. For example, ASRock Z68M/USB3, $95. H61/H67 is no good because it supports only up to DDR3-1333 SDRAM.
- Intel HD Graphics @1350MHz. The default clock is 1100MHz. GPU of every SNB processor should be able to run at this clock without problem.
- Player: Any that supports madVR. I used MPC HomeCinema.
- Video decoder: Intel QuickSync. Right now this is part of ffdshow. If you use Pentium G800 Series or higher, you can also use libavcodec.
- Deinterlacer: yadif (frame doubling) in ffdshow for interlaced contents. In future Intel QuickSync may support GPU's hardware deinterlacer.
- Video renderer: madVR, medium quality (Bilinear/Bicubic 50/Bicubic 50), in full screen exclusive mode. For high quality settings, you will need Intel HD Graphics 3000 such as Core i3-2105/2125.

I used the following 2 minute video clips:

- SD film: Ratatouille (2007) (480i60, MPEG-2, AC3)
- SD video: Die Zauberflote (2003) (480i60, MPEG-2, AC3)
- HD film: Iron Man (2008) (1080p24, AVC, TrueHD/AC3)
- HD i video: La Traviata (2010) (1080i60, AVC, DTS-HD MA)
- HD p video: La Traviata (2010), created from the above by deinterlacing with yadif (1080p60, AVC, DTS-HD MA)

The following two are HD i video in VC-1 format, that is rare and sometimes hard to play back (even 4shared.com can't create thumbnails/previews ):

- HD i video (VC-1): Life - Plants (2010) (1080i60, VC-1, DTS-HD HRA
- HD i video (VC-1): Nature's Journey (2007) (1080i60, VC-1, DTS-HD MA)

To be clear,

 

	Origin	Format	Output to renderer	Frame interval
SD film	film	480i60	480p24	41.708 ms
SD video	video	480i60	480p60	16.683 ms
HD film	film	1080p24/1080i60(broadcast)	1080p24	41.708 ms
HD i(nteralced) video	video	1080i60	1080p60	16.683 ms
HD p(rogressive) video	video	1080p60	1080p60	16.683 ms

 

These five are the major video formats found in NTSC countries (well, except for 720p60 in broadcast, which is much easier to play back than 1080p60).

Results (with Celeron G530)

 

	SD film	SD video	HD film	HD i video	HD p video
Dropped frames	0	0	0	0	0
CPU usage (average)	11%	21%	23%	74%	55%
GPU usage (average)	54%	88%	25%	60%	59%
Rendering time (average)	20.93 ms	13.90 ms	8.87 ms	8.88 ms	8.90 ms

 

Why Intel QuickSync video decoder for Celeron G500 Series and Pentium G600 Series processors?

With libavcodec (ffdshow's default decoder), I got:

	SD film	SD video	HD film	HD i video	HD p video
Dropped frames	0	0	0	4017	3715
CPU usage (average)	11%	18%	50%	93%	97%
GPU usage (average)	55%	88%	25%	30%	31%
Rendering time (average)	20.88 ms	13.89 ms	8.61 ms	9.46 ms	13.56 ms

 

There are lots of dropped frames at HD i/p video playback. The bottleneck is the weak CPU that can't handle AVC decode (libavcodec) and deinterlacing (yadif) simultaneously. Intel QuickSync supports hardware AVC/VC-1/MPEG-2 decode under madVR, hence offloads CPU. If you use Pentium G840 or higher, there should be no such problem with libavcodec.

Why overclock GPU?

With the default GPU clock 1100MHz, I got:

 

	SD film	SD video	HD film	HD i video	HD p video
Dropped frames	0	687	0	0	0
CPU usage (average)	13%	21%	25%	70%	54%
GPU usage (average)	54%	96%	25%	60%	59%
Rendering time (average)	20.85 ms	17.40 ms	9.13 ms	8.96 ms	8.89 ms

 

GPU struggles with upsampling SD at the rate of 60 fps. To reduce 17.40 ms to, say, 14.00 ms, safely below the threshold value of 16.68 ms, it needs to be overclocked by the factor of 17.40/14.00 = 1.24, resulting in 1367 MHz. Every SNB processor's GPU should be able to run @1350MHz with no problem. (Note that 1350MHz is the default GPU clock of Core i7-2600K, the highest SNB processor.)

Why DDR3-2133?

With DDR3-1066, all the other settings remaining the same as the first one, I got:

 

	SD film	SD video	HD film	HD i video	HD p video
Dropped frames	0	627	0	2004	0
CPU usage (average)	14%	25%	29%	91%	76%
GPU usage (average)	59%	95%	27%	59%	65%
Rendering time (average)	21.74 ms	16.41 ms	9.98 ms	12.17 ms	10.13 ms

 

With DDR3-1600 (supported only by Z68 chipset), all the other settings remaining the same as the first one, I got:

 

	SD film	SD video	HD film	HD i video	HD p video
Dropped frames	0	162	0	0	0
CPU usage (average)	12%	23%	28%	77%	61%
GPU usage (average)	55%	94%	26%	64%	62%
Rendering time (average)	21.17 ms	15.52 ms	9.57 ms	9.71 ms	9.66 ms

 

Without enough memory bandwidth, CPU struggles with deinterlacing and/or GPU struggles with upsampling SD at the rate of 60 fps.

Edited by renethx - 6/26/12 at 11:04pm

I did swap for another Agility 3. But this build is for HTPC movie use only in a room we use for movies maybe once or twice a week. No gaming, browsing or office use. Only apps I use are Mainlobby and Arcsoft. So when I say I am "stable," please understand this frame of reference. That said, I went from BSOD's more often than once an hour, to maybe one or two over a month.

Sadly the results do show that it would still fail on 60 fps content.

20ms rendering time is too slow for 60 fps. Run yadif in double frame rate mode (the only mode really worth using) on interlaced sources, and it would start dropping frames. On these "low" quality settings in madVR, a HD3000 would probably manage, though (assuming the CPU is fast enough for double-rate yadif)

Interesting will be hardware deinterlacing results, both in Quality and Performance.

20ms is SD film (23.976 fps), for which 41.708 ms = 1/23.976 second rendering time is allowed. madVR receives 60 frames per second from SD video (480i60 non-film), HD interlaced video (1080i60), and HD progressive video (1080p60, the main source is camcorders right now). yadif deinterlaces 60 fields per second to 60 frames per second. As the the chart above shows, the average rendering time in any of these three cases is < 16.683 ms = 1/59.94 sec. Actually I ran longer clips (30 miniutes) and I saw zero dropped frames.

Intel HD Graphics 3k + DDR2-2133 is good enough for high quality settings (softcubic/Lanczos or spline 4 taps) for every kind of video formats. Using DDR3-2133 is essential in either case. With DDR3-1333, there will be lots of dropped frames with to-be-60fps contents. I tested it thoroughly for several days.

The rendering time is independent of the actual fps, its per frame - and an average of 20ms is just too high for 50p or 60p.
As evidenced by the 1080p material, all formats have about the same rendering time.

It could be possible that the GPU went into a lower performance state on your SD film material, increasing the rendering time that way. Your stats don't show those facts.

PS:
If its film content at 23.976, maybe you shouldn't call it "480i60" and "1080p60" to reduce confusion?

PPS:
It seems rather weird that the memory speed makes such a difference. All previous benchmarks (unrelated to video) i have seen show that at the maximum you would want 1600, anything above just costs money with basically no return. I should re-run some benchmarks with the latest drivers on my HD3000, and compare 1333 to 1600, see how the timings are these days.

madVR's upsampling is non-adaptive so your claim is right. Perhaps CPU and GPU are in low-power state at SD 24 fps playback so that the rendering time could be longer. But practically it does not matter, we almost always turn on Intel SpeedStep, and it's about 24 fps, the frame interval is very long, any decent system can handle it easily. As for 60 fps, I couldn't find 480i60 video-based contents (music concerts, sports events etc.) that the above configuration can't handle. If somebody finds a clip that G530 with GPU @1350MHz + DDR3-2133 + madVR medium quality can't handle, please post it, I am willing to test it.

To be clear,
 

	origin	format	what video renderer receives	frame interval
SD film	film-based	480i60	480p24	41.708 ms
SD video	video-based	480i60	480p60	16.683 ms
HD film	film-based	1080p24/1080i60(broadcast)	1080p24	41.708 ms
HD i(nteralced) video	video-based	1080i60	1080p60	16.683 ms
HD p(rogressive) video	video-based	1080p60	1080p60	16.683 ms

These are the major video formats found in NTSC countries (well, except for 720p60 in broadcast, which is much easier to play back than 1080p60).

I tested Core i2-2105 and Core i7-2600K, both with Intel HD Graphics 3k (12 EUs), but graphics behavior is somewhat different. For 2600K, DDR3-1333 is enough (update: That's wrong, I didn't run SD video). For 2105, I tested DDR3-1333/1600/1866/2133 with madVR high quality settings and only DDR3-1866/2133 can handle HD i/p video (in the above sense) without dropped frames. With DDR3-1866, GPU even needs to be overclocked. I haven't expected such results but it turned out to be true. BTW DDR3-1866/2133 are not officially supported by SNB, but every (afaik) Z68 mb supports DDR3-1866/2133 and every SNB('s memory controller) should be able to handle them easily.

Edited by renethx - 6/26/12 at 11:00pm

I wonder, especially with how obviously the system memory affect iGPU performance, if anyone is planning any motherboard with dedicated graphics memory. I have an AMD 785G motherboard that has 128MB of soldered-on memory for the HD4200 GPU. I am surprised none of the AMD APU motherboards, especially the Zacate, have dedicated GPU memory on-board.

Assuming it did actually go into a lower power state because it had such a easy time decoding and rendering the 480p24 content, i would say the results look promising.

Only thing missing is hardwarer deinterlacing, which either Eric or me will hopefully be able to offer soon.

Ivy Bridge will probably increase the performance by ~60%, plus offer chips with even more EUs then the HD3000. Now they just need to fix the 24p issue...
I'm anxious to get my hands on a IVB CPU for a fresh new HTPC though, too bad it will be quite a while still.

Noticed and upgraded the latest one from W7 update and so far so good.

My understanding is the difference would be huge.

Yup, if I can't get a awesome deal on a A4/A55 combo before Xmas I might just wait and see if the 24p bug is fixed. Ivy bridge specs look pretty amazing.

It may be easier said than done. At a CPU/GPU level, the memory controller may have to be completely reworked to support dedicated video memory. For your old motherboard with a HD4200 GPU, it expects to have dedicated memory, and if you asked "why can't my HD4200 use my main system memory," well, it isn't designed to do that.

Having CPU/GPU integrated graphics support discrete video memory is an issue bigger than what a motherboard manufacturer can address. And because most people use integrated graphics to save a few bucks over buying a discrete graphics card, I don't expect them to make integrated graphics more expensive with discrete video memory.

So you are saying that you are still having a couple of BSODs per month still?

Hmmmm OK. Could I ask what firmware version you are running on your Agility 3?

Are people running 2281 SF SSDs on their H67 or H61 motherboards? Issues?

I'm curious. What is your definition of having the 24p bug fixed?
Exactly 23.976? I'd love it if it could do exactly 23.976!
23.975? My ATI 5770 does this, but not all AMD/ATI cards can get this close.
23.974?
23.973? My i7-2600K does this.
etc.

There is another thread discussing how nobody is doing exactly 23.976 right out of the box. With some tweaking, some cards can get closer than others.

When you're less than 23.976 you get dropped frames, and when you're above you get duplicated frames. Both of them annoy me.

I added results with DDR3-1066 and DDR3-1600 to the post. Even DDR3-1600 and GPU @1350MHz is not enough to upsample SD at the rate of 60 fps.

G530T is supposedly launched but I can't find it for sale anywhere in the US. Is it really out yet?

http://ark.intel.com/products/53415/...Cache-2_00-GHz)

That's not entirely true. There are flavors of HD4200, HD3200, etc. with and without "sideport" memory. More often, you'd find HD4200 using system memory. There'd usually be one board in a series with 128MB sideport memory.

It's a different architecture, sure, but I still wonder if it's possible.

Core i3-2105/2125, Core i5-2500K, Core i7-2600K with Intel HD Graphics 3000 has 12 EUs, twice that of Celeron/Pentium/Core i3-2100 and other SNBs with Intel HD Graphics 2000. So we can use high quality settings of madVR with these CPUs.

- Processor: Any SNB processor with Intel HD Graphics 3000. In the test below, I used Core i3-2105.
- Memory: DDR3-2133 2 x 2GB @2133MHz. For example, G.SKILL F3-17000CL9D-4GBXL, $55.
- Z68 chipset mb. For example, ASRock Z68 Pro3-M, $110.
- Intel HD Graphics 3000 @1350MHz. The default clock is 1100MHz, except for Core i7-2600K whose default clock is already 1350MHz. 1350MHz (~25% overclocking) should cause no stability issue.
- Player: Any that supports madVR. I used MPC HomeCinema.
- Video decoder: Intel QuickSync. As the processor is powerful enough, you can also use libavcodec with no problem.
- Deinterlacer: yadif (frame doubling) in ffdshow for interlaced contents.
- Video renderer: madVR, high quality: SoftCubic (softness: 100)/Lanczos (3 taps)/Lanczos (3 taps), in full screen exclusive mode.

I used the same clips as before.

Results (with Core i3-2105)

	SD film	SD video	HD film	HD i video	HD p video
Dropped frames	0	0	0	0	0
CPU usage (average)	9%	13%	14%	30%	23%
GPU usage (average)	52%	82%	23%	56%	54%
Rendering time (average)	20.66 ms	13.40 ms	8.90 ms	8.79 ms	8.76 ms

Why Lanczos 3 taps instead of 4? Why overclock GPU?

- Lanczos 4 taps: sharper, better aliasing
- Lanczos 3 taps: less ringing

Again, the decision is based on the results of SD video playback, the case most taxing GPU. At the default GPU clock of 1100MHz,

	rendering time (rt)	(rt)/(14.00 ms)	(rt)/(14.00 ms) x 1100MHz
4 taps	20.46 ms	1.46	1608MHz
3 taps	16.30 ms	1.16	1281MHz

The right hand column is the GPU clock with which rendering time would be ~14.00 ms, safely below the threshold value of 16.68 ms. My 2105 can run stably at 1600MHz (46% overclock), but voltage bumping +0.1V is necessary. In general, to avoid stability issue, 3 taps is a good compromise for Intel HD Graphics 3000. In this case GPU @1350MHz is fast enough. IVB with 16 EUs should be able to handle 4 taps with no problem.

Edited by renethx - 6/26/12 at 10:59pm

With Sandy Bridge, i would still be a bit careful before committing to a pure SNB HTPC.

Here are the main points i am concerned about (hoping for them to be fixed with IVB):

1) 24p problem
2) VC-1i decoding is still a bit "wonky" on some clips
3) The ffdshow QuickSync decoder isn't 100% stable and finished.
4) Performance might be enough with overclocking and 2133MHz RAM, but its cutting it really close.
5) Its unclear as of yet how the quality of the deinterlacer is when used via the MSDK, and how using it would impact rendering performance

Remarks to the points above:
1) is the old problem, and we all hope the next PCH will solve it.
2) Is probably a driver issue, and only seems to affect certain clips using so called "field interlacing".
3) Will of course improve over time, and i plan to do my part when adding decoding support to my LAV Video, letting it benefit from my DirectShow knowledge/architecture. The main problem is that Eric is new to DirectShow. He does a great job at implementing the QuickSync parts, but the DirectShow interactions are still a bit off.
4) was discussed alot above, and with the current setup, it might just work, but there is no headroom.
5) on the other hand is still a big variable, and we'll have to see once HW Deint is available. The added GPU usage could impact rendering times, or it could not, we'll just have to wait.

Yes. No.

I am running an Agility 3 60GB on my Gigabyte H67-based mini-ITX board. The SSD came with fw 2.11 (which was the latest available at the time). I have not had a single hiccup, let alone a BSOD. Been going over a month now. Mostly in and out of S3, but turned off and on plenty too.

Drive came with 2.9 as I recall. Updated to 2.11 before OS install. Have not updated to 2.13 and do not plan to do so unless I start having more problems. Have NOT had any BSOD's during movie viewing, so that's what matters to me. Since the Agility 3 makes complete start up so fast, I have started powering down after watching movies to avoid any BSOD while in sleep mode.

Hi jakmal, thanks for the cover piece on Anandtech. I was happily surprised it surfaced so fast

BTW, my project is open source, the source code is released with each build (BSD license - can be used by all). I didn't have the time to properly set up a SourceForge project, that will happen within a week or so. Fixing few bugs was more urgent
My code should change a bit for easy integration to other decoders (other than ffdshow) so hasty integration ATM will cause overhead.

Regarding HW deinterlacing, EVR already deinterlaces using HW and I plan to pull this functionality into the decoder (HW DI) once the decoder is in a more stable shape.

Please bear in mind that the Intel QuickSync Decoder is in early alpha stage and I can't do everything at once.
Feature request and bug reports are more than welcome. Please post them on the development thread so I can find them easier.

FYI - BIOS 131 is up for intel -67 motherboards.

http://downloadcenter.intel.com/Deta...&DownloadType=

Subjectively, i wouldn't say that AMD has the best deinterlacing. It looks alot like they are losing alot of vertical detail in order to avoid artifacting. Look at the overall image of the cheese slice test you linked, the top 2/3 lost all of the horizontal lines. Thats what happens with NVIDIA if you use their lower-quality algorithm, but at the highest quality, the lines stay where they are (and look good too!)

There is also the "Noise Response" image, where at least the angled lines on the right look much better with Intel. The left part doesn't really look good on any deinterlacer, although the top part is worse on Intel, i guess.

Admittedly, Intel seems to not produce a image thats as sharp as AMDs or NVIDIAs.

The only parts where AMD seems to be really better is at moving angled lines, like on the sports image or the sailing ship there, but in that same mode other parts of the image lose quality.
Thats really the problem with AMDs algorithm. Their Vector Adaptive works fine with such constructs, however with other parts of the image their Motion Adaptive is better. They don't have a all-in-one. NVIDIAs is more of a mixture of motion and vector adaptive, it doesn't work as good on some parts where AMDs VA works really well, but instead it doesn't fail on other parts either. For real world content, thats better imho.

Not to forget that Cheese Slices is a very artificial test. It should really only be used to determine a baseline response for the deinterlacer (what kind of algorithm does it use, etc), but not to judge its quality in the utmost detail. It uses patterns and movements that are very much computer generated, and unlikely to happen in real world content.

Anyhow, this is just my own opinion, not measured by any testing scales, just how the image looks to me.
I have not seen enough of Intels DI to judge it yet, either.

TL;DR
It depends on the content which algorithm might be better. On sports, AMD might be better, on movie content, it might not. On artificial pattern - who cares?
Its all subjective anyway.