x264 utilization of cores concern

[dvd_maniac]

Looking into a new PC build and not sure which way to go.
Benchmarks seem to have these 2 CPU's pretty close in encoding using X264.
The main purpose for this PC will be video encoding. Some of the benchmark reviews I have read are saying that X264 is not using all the cores. Is this correct for 6-core CPUs? If so, is it something that is in the process of being updated in future release of X264 or should I just go with the i7930?

[burfadel]

If the Intel and the AMD perform pretty much the same in encoding using x264, however in the tests x264 doesn't use all the cores well, then looking at it objectively the AMD is actually faster! In the tests you have to consider the version of x264 used, and if avisynth was used was it the multithreaded version? Did they use ffmpegsource, and if so, did they use the multithreaded ffmpegsource and ffmpeg? (test versions I know).

With x264, you can also play with certain setting to potentially improve threading performance that they wouldn't have done in the tests. There are also patches like the thread pool patch which may prove useful when using 6 cores.

[Dark Shikari]

x264 uses lots of cores just fine. In fact, if x264 didn't use all the cores, it would benefit AMD, because Intel chips have more cores than AMD chips (once Hyperthreading is considered). If the review thinks x264 isn't using all the cores, they probably have no idea what they're talking about (or they have a very bad test), so odds are they're not to be trusted on results anyways.

Unfortunately, even when considering the effect of multithreading, the Core i7 quad trounces (http://www.anandtech.com/show/3674/amds-sixcore-phenom-ii-x6-1090t-1055t-reviewed/6) the Phenom x6 pretty effectively in x264 encoding as far as I've seen. The only saving grace is that the Phenom is absurdly cheap, which might make it worth it despite each core being a lot weaker.

[turbojet]

At stock speeds 1090T is ~10% faster with x264, costs less for mobo/cpu/memory, uses less power and runs cooler then i7 930.

It's not as fast as intel's $1000 i7's (980 and 975) but is that really a fair comparison?

After overclocking the 1090T can almost achieve the same performance as the i7 975 while using less power and running cooler.

[Blue_MiSfit]

I'm interested on the server side comparing the 12 core AMDs versus the 6 core Intel CPUs. Any benchmarks of these (the AMDs)?

~MiSfit

[CpT]

I'm interested on the server side comparing the 12 core AMDs versus the 6 core Intel CPUs. Any benchmarks of these (the AMDs)?

~MiSfit

No clue m8, but I rent 4 dedicated boxes and recently upgraded the hardware "there was a sale I couldn't resist" from quad phenom 955 BE's to intel quad 9550's + diff mobos and 8gb of ram on centos5.
The difference was massive via apache and mysql benchmarks - I use nginx as the main web server and it hits over 8k requests per second on dynamic data 15k+ for static content. The amd setup was around half of that / 4500-5200 requests per second on dynamic and 9k max for static.

Regarding x264 cpu usage - I run an intel 980x clocked to 4.5ghz 24/7.
http://valid.canardpc.com/show_oc.php?id=1107202
X264 uses 100% of all 6 cores nonstop with no issues at all... Whoever ran those tests did something wrong.

Although its worth noting I see 0 fps increase when hyperthreading is enabled.
But that's to be expected, there's very few apps that can make good use of hyperthreading.
Generally I leave it disabled because most the games I play can't make use of the 6 cores much less 12 threads...

[Blue_MiSfit]

That's odd.. most of the benchmarks I've seen with x264 show a good increase when enabling hyperthreading... Oh well :)

~MiSfit

[Dark Shikari]

Although its worth noting I see 0 fps increase when hyperthreading is enabled.Then you did something wrong; x264 gains about 25% from hyperthreading.

[dvd_maniac]

I'm using the latest version of Staxrip and currently have the Intel Q6600. The only thing this PC is generally used for is recording TV and encoding video using Staxrip.

What would the experts suggest for a complete new build, Intel i7930 or AMD 1090T? No real budget in mind, just best dollar to fps encoding ratio

[Blue_MiSfit]

I'd say the i7 930, personally.

~MiSfit

[Groucho2004]

I found this review particularly interesting:
http://www.techspot.com/review/269-amd-phenom2-x6-1090T-and-1055T/page7.html

It's obvious that the Tech Arp x264 HD benchmark doesn't ulilize all cores as it should. Handbrake looks OK.

[LoRd_MuldeR]

I'd say the i7 930, personally.

~MiSfit

Why would you get a Socket-1366 ("Bloomfield") version?

I would prefer an i7 860 instead, i.e. the Socket-1156 ("Lynnfield") version. It's cheaper at the same clock speed. And most important the boards are significant cheaper.

Also from what I read, the "Turbo Boost" works more efficient in the Lynnfield. Probably not relevant for x264, but certainly doesn't hurt either.

So the only advantages of spending more money for a Bloomfiled are the QPI interface (which is only relevant if you really need a lot of PCIe lanes) and the third memory channel.

(From what I read, the impact on performance of the third memory channel is negligible. It probably is outperformed by the more efficient Turbo Boost for most applications)

[Blue_MiSfit]

A 1366 setup isn't THAT much more expensive, and I like having 6 RAM slots instead of 4, but that's just me :)

~MiSfit

[Assassinator]

I'm using the latest version of Staxrip and currently have the Intel Q6600. The only thing this PC is generally used for is recording TV and encoding video using Staxrip.

What would the experts suggest for a complete new build, Intel i7930 or AMD 1090T? No real budget in mind, just best dollar to fps encoding ratio

Based on cost/fps ratio, AMD 1090T is better. Especially considering the fact that your other parts (eg. motherboard/RAM) are going to be much cheaper.

... unless you're overclocking, then I don't know enough about the overclocking potential of the AMD to make an accurate comment. But I do know the i7 series are traditionally better than the PII series at overclocking.

I found this review particularly interesting:
http://www.techspot.com/review/269-amd-phenom2-x6-1090T-and-1055T/page7.html

It's obvious that the Tech Arp x264 HD benchmark doesn't ulilize all cores as it should. Handbrake looks OK.

The TechARP x264 HD benchmark DOES utilize all the cores.

It's just that... for some unknown reason... they (benchmark you linked to) decided to only show the 1st pass results, instead of the 2nd pass or ideally both.

This is extremely stupid if you ask me, obviously the 1st pass won't use all the cores. And that resulted in the Phenom II 965 beating i7 930, lololol (for those who don't know, i7 930 is easily faster).

A 1366 setup isn't THAT much more expensive, and I like having 6 RAM slots instead of 4, but that's just me :)

~MiSfit

Spend that extra hundred in your CPU, and you'll see significant increases in encoding performance.

I find 2 RAM slots to be all I need (2x2GB), but that's just me :)

[burfadel]

Yeah the socket 1366 processors are actually quite old now, technology wise. They've been out a very long time! Sure they're fast, but they do have some disadvantages, such as no on chip PCI-E chip (all socket 1156 processors have it onboard, not to be confused with onchip GPU). The socket 1156 Core i7's have the disadvantage of using the slower DMI over the QPI. Realistically, a core i7 860 with QPI would have only been a few dollars more than with DMI, but since the 9xx are meant to be performance chips Intel decided to hinder the performance a little (it wouldn't look good an 860 beating a 930 for example, although it comes close even now!

I was thinking of upgrading myself from a q9400, but in reality it really isn't worth it, the cost is quite a bit considering RAM, CPU, motherboard. I'd much rather wait for the new AMD's coming out next year, I'm sure AMD are striving for it to trounce the Core i7's, plus they'll have all the new instructions such as SSE5 and AVX etc.

So:
- Value for money spending on a system now vs next year
- In each case a whole new set of components are required (not a simple plop in like a q6600 to q9550 for example)
- Many new cores, may be interesting to see how x264 balances over all the cores
- New instructions, maybe x264 will require significant additions of instructions, and able to use that to improve encoding even further

Reason I bring this up is that it is a consideration, even though its a year away it will be (by the sounds of things) a significant change that may warrant looking for value for money now and performance next year. In that case, if the 6 core AMD is only a few percent slower than the i7-930, but is much cheaper, than it may be a wise option!

[Groucho2004]

It's just that... for some unknown reason... they (benchmark you linked to) decided to only show the 1st pass results, instead of the 2nd pass or ideally both.

Yes, I realized this as well after I actually looked at the numbers. :rolleyes:

Yet another crowd of reviewers who don't have a clue, I guess...

[CpT]

Then you did something wrong; x264 gains about 25% from hyperthreading.

I've tested it now on 3 different os's, same hardware and every up2date gui here. 0 fps increase on the 980x with hyperthreading. I can't even get it to use a constant 100% of the cpu with hyperthreading on.

If you know a trick or can recommend a gui to test this I'm open to suggestions.

This chip w/o hyperthreading x264 uses 100% of the cpu nonstop with no issues at all. And its fast as hell. I run placebo with -me 28 and still get really great fps.

Regarding the I7 930, every one I've seen hits 3.9+Ghz on air cooling. Unless that amd can clock extremely well its not all that impressive especially clock for clock. I mean jezus the amd can barely beat a quad lol...

[Boolsheet]

If you know a trick or can recommend a gui to test this I'm open to suggestions.

Feed x264 a raw stream. Execute something like this:

ffmpeg (http://ffmpeg.arrozcru.com/autobuilds/) -i your_favourite_720or1080_source.mkv -vframes 2000 -vcodec rawvideo -an -pix_fmt yuv420p -f yuv4mpegpipe testsample.y4m
x264 testsample.y4m --threads 12 -o NUL (or /dev/null ;) )

Make sure it's cached into RAM. Vista and 7 should automatically do that, when ffmpeg writes it to the disk.
Try slower presets if your source doesn't have enough work for all threads.
I don't think the CPU detection has problems, but setting threads manually never hurts.

Would be a shame if they screwed up HT in their new chips.