Need a new computer - obviously budget-limited (who isn't?)

[aedipuss]

i couldn't pass up the "free" o/c of the k unit with a p67. i have been encoding with 8 threads @ 4.0GHZ 100% usage. it's nice. temps stay in the mid 60's c during encoding.

what kind of encoding speeds do you see with that thing?

got a test you would like run? i haven't done any speed testing, just work encodes with vegas pro 10 mostly.

[deadrats]

got a test you would like run?

something similar to what i proposed for QS, use the latest media coder, 32 or 64 bit, doesn't matter, take a 1080p source and transcode it with ac3 audio using x264 on ultra fast, normal and slowest to 8mb/s 720p and just report the fps.

[jagabo]

ivy bridge is rumored to be pin and chipset compatible with sandy bridge and version 2 of QS is rumored to be twice as fast

I wonder. Going from an Intel HD Graphics 2000 to 3000 doubles the number of execution units but doesn't make a 2x change in encoding speed. I suspect the reality for Ivy Bridge will be less than 2x. I've been thinking of getting a z68 motherboard and a i3 2105 just so I can play around with Quick Sync.

[aedipuss]

ok. i'm using a blu-ray rip of lotr fotr extended edition with dts 7.1 es audio for the source. 720p with ac-3 2ch. output. here's what it can do on the slowest "placebo" setting. i'm not going to let it finish as i have no use for it.

[aedipuss]

here's ultra-fast preset.

[aedipuss]

medium(there is no normal)

[aedipuss]

cuda

[aedipuss]

some readings seem weird. throughput of up to 287MB/s? what's it doing adding read + write together? my samsung f3's can do about 140MB/s sustained. ultrafast and cuda do seem to be limited by the drive transfer speeds, unused cpu clocks left over.

[aedipuss]

here's an old x264 benchmark from 2007 i found hanging around.

2600k

---------- RUN1PASS1.LOG
encoded 1749 frames, 286.77 fps, 1849.61 kb/s

---------- RUN2PASS1.LOG
encoded 1749 frames, 289.81 fps, 1849.61 kb/s

---------- RUN3PASS1.LOG
encoded 1749 frames, 287.00 fps, 1849.61 kb/s

---------- RUN4PASS1.LOG
encoded 1749 frames, 288.66 fps, 1849.61 kb/s

---------- RUN5PASS1.LOG
encoded 1749 frames, 289.52 fps, 1849.61 kb/s

---------- RUN1PASS2.LOG
encoded 1749 frames, 97.76 fps, 1834.88 kb/s

---------- RUN2PASS2.LOG
encoded 1749 frames, 96.62 fps, 1834.54 kb/s

---------- RUN3PASS2.LOG
encoded 1749 frames, 97.73 fps, 1834.86 kb/s

---------- RUN4PASS2.LOG
encoded 1749 frames, 97.91 fps, 1834.86 kb/s

---------- RUN5PASS2.LOG
encoded 1749 frames, 97.79 fps, 1834.86 kb/s


q6600 @ 3.2GHZ from a couple years ago

--------- RUN1PASS1.LOG
encoded 1749 frames, 119.59 fps, 1850.94 kb/s

---------- RUN2PASS1.LOG
encoded 1749 frames, 120.50 fps, 1850.94 kb/s

---------- RUN3PASS1.LOG
encoded 1749 frames, 119.46 fps, 1850.94 kb/s

---------- RUN4PASS1.LOG
encoded 1749 frames, 120.62 fps, 1850.94 kb/s

---------- RUN5PASS1.LOG
encoded 1749 frames, 120.11 fps, 1850.94 kb/s

---------- RUN1PASS2.LOG
encoded 1749 frames, 32.59 fps, 1829.14 kb/s

---------- RUN2PASS2.LOG
encoded 1749 frames, 32.56 fps, 1829.55 kb/s

---------- RUN3PASS2.LOG
encoded 1749 frames, 32.53 fps, 1829.39 kb/s

---------- RUN4PASS2.LOG
encoded 1749 frames, 32.60 fps, 1829.06 kb/s

---------- RUN5PASS2.LOG
encoded 1749 frames, 32.54 fps, 1829.55 kb/s

[deadrats]

I wonder. Going from an Intel HD Graphics 2000 to 3000 doubles the number of execution units but doesn't make a 2x change in encoding speed. I suspect the reality for Ivy Bridge will be less than 2x. I've been thinking of getting a z68 motherboard and a i3 2105 just so I can play around with Quick Sync.

you're not thinking about things from a programmers view point, your thinking about it based on the half assed QS tests that you've seen performed so far.

intel has said that they analysed the entire encoding process and discovered that motion estimation was the most intensive part of the encoding process, so they designed SB to perform just the ME portion of the encode on the execution units. everything else is implemented as discrete fixed function units built into the gpu but separate from the EU array. i'm guessing you know where i'm going with this.

i'm betting that if your encode is using SAD and a search range of 16 that the execution units aren't really getting that big a work out and thus the difference between 6 and 12 units isn't all that great but use a sum of hardamad differences and a search range of 64 and i think then you would see a big difference between the HD2000 and the 3000.

that being said, i was ready to make the jump to SB as soon as it came out, unfortunately i lost my job and right now i can't afford squat.

but as soon as i can, i think i think i'll pick up a cheap H61 based board and a i3 2100, see what QS has to offer and if i find it lacking, quality wise, then i can always pick up a high end video card and an Ivy Bridge quad core when they come out.

[deadrats]

some readings seem weird. throughput of up to 287MB/s? what's it doing adding read + write together? my samsung f3's can do about 140MB/s sustained. ultrafast and cuda do seem to be limited by the drive transfer speeds, unused cpu clocks left over.

the first time i saw media coder's throughput readings i had a similar thought but i always thought that the measurement was accurate but that the output was being buffered to ram and then written to disk, thus 287MB/s is possible if you're talking about how fast the file is being created because technically even if it's still in the buffer it's still has been created, it just hasn't been written to disk yet.

but the 108fps for x264 ultra fast going from blu-ray rip to 8mb/s 720p is amazing, just wow.

[aedipuss]

if the output was being buffered to ram than ram usage during that encode would have been increasing at 150MB/s. i can tell you it wasn't, the ram usage was at a constant level, and after a minute the system would have crashed with an out of memory error even with 16GB.

[jagabo]

if the output was being buffered to ram than ram usage during that encode would have been increasing at 150MB/s. i can tell you it wasn't, the ram usage was at a constant level, and after a minute the system would have crashed with an out of memory error even with 16GB.

Yes.

It's the size of the uncompressed video. 1280 * 720 * 3 = 2,764,800 bytes per frame. Multiply that by 108 fps and you get 298,598,400 bytes per second, ~287 MiB/s.

[jagabo]

ivy bridge is rumored to be pin and chipset compatible with sandy bridge

Anandtech seems to indicate it's more than a rumor:

Ivy Bridge is pin compatible with Sandy Bridge, and it will work on current LGA1155 motherboards with the appropriate chipset and a firmware and BIOS update (H61, H67, P67, and Z68 are capable of support IB)

http://www.anandtech.com/show/4318/intel-roadmap-ivy-bridge-panther-point-ssds

[jagabo]

[I've been thinking of getting a z68 motherboard and a i3 2105 just so I can play around with Quick Sync.

Ooh, Fry's has the i5 2500K on sale to day for $180...

[magic612]

Can a moderator split this thread? A lot of it is OT now. I appreciate that others want to discuss speed/processor issues, but I keep coming back after getting email alerts to new posts in the thread only to discover it has nothing to do with what I originally posted about.