AMD Chip Rapes Intel C2Duo discussion

[DanceswithUnix]

Just found this thread, my first thought was "are you running 64 bit Linux"

Firstly, the SSE implementation is I gather different on the 2 platforms, so if the SSE is done in small chunks then the AMD will be faster. That can of course be optimised out.
The other thing that strikes me: The AMD is only at 80% on one core, so you have a bottleneck. If you can feed the CPU faster, you will get more out of it.

[ikonia]

Just found this thread, my first thought was "are you running 64 bit Linux"

Firstly, the SSE implementation is I gather different on the 2 platforms, so if the SSE is done in small chunks then the AMD will be faster. That can of course be optimised out.
The other thing that strikes me: The AMD is only at 80% on one core, so you have a bottleneck. If you can feed the CPU faster, you will get more out of it.

I think you may have miss-understood - take a re-read of the thread.

The fact that the AMD is spreading the load out over multiple cores is a GOOD thing and making the AMD faster. The Intel C2d's are by comparision perfoming poor by bottlenecking 1 CPU core.

[ikonia]

x264 threads like a dream - ~181%

shame you want mpeg4 part 2, not mpeg4 part 10

As you know - I'm not big on the multimedia codec thing - so can you enlighten me on the differences, part 2 and part 10 ???

I'm using xvid for a.) good compression ratio for the quality I want b.) compatability.

I'm open to other options, more so if it lets me utilise hardware.

[DanceswithUnix]

I believe I have understood the thread, I guess I am not making my point clearly.

One core should be at max if the system is CPU limited. Both cores at max would be nice, but most programs don't balance that well.

You have a total of zero cores maxxed out on the AMD chip, so you are not CPU limited and have a bottleneck somewhere else (hard drive for example).

[DanceswithUnix]

Another thought. If you are blocking up work across multiple machines, then can you run your transcoding job twice on the Intel box and not bother with multi thread support?

[kalniel]

That is true, but it could also mean that the AMD CPU is not being stretched as much as an Intel core, thus something else inevitably becomes the bottleneck.

If the thread dispatch/balancing wasn't performing well in Intels case then one core could be unevenly loaded, causing the CPU to become a bottleneck sooner.

[directhex]

As you know - I'm not big on the multimedia codec thing - so can you enlighten me on the differences, part 2 and part 10 ???

I'm using xvid for a.) good compression ratio for the quality I want b.) compatability.

I'm open to other options, more so if it lets me utilise hardware.

mpeg4 part 2 covers one family of codecs - divx, xvid, 3ivx, nero digital, and others. in theory, if foodevice supports one of the above codecs, it should support all of them (since they're all implementations of the same spec)

mpeg4 part 10 is a different family of codecs - also known as avc or h264 (x264 being an open source implementation). it's newer, requires beefier hardware to encode or decode, but is significantly better at encoding. it's being pushed for HD content, and is one of the three codecs approved for use with HD discs (along with ye olde mpeg2 and microsoftian wmv9). sky hd uses it, fr'example. compatibility is nowhere near as universal as xvid/divx, so it's a question of use case as to whether it works for you

[DanceswithUnix]

That is true, but it could also mean that the AMD CPU is not being stretched as much as an Intel core, thus something else inevitably becomes the bottleneck.

Exactly.

The CPU is spending time waiting, so *something* else in the system is running 100%.

Is the work coming in over gigabit ethernet by any chance? If so, what chipsets are involved?

[directhex]

Exactly.

The CPU is spending time waiting, so *something* else in the system is running 100%.

Is the work coming in over gigabit ethernet by any chance? If so, what chipsets are involved?

You assume a trivially parallel scenario, if you expect 100% utilization of both cores

If there's an interdependent uneven workload on both, then it'll never properly go at full tilt

At any rate, testing suggests the AMD versus Intel problem is entirely in the xvid code, which solves the original question

[ikonia]

work is not being spread out across multiple machines - multiple machines are working indepentandly

[DanceswithUnix]

You assume a trivially parallel scenario, if you expect 100% utilization of both cores

If there's an interdependent uneven workload on both, then it'll never properly go at full tilt

At any rate, testing suggests the AMD versus Intel problem is entirely in the xvid code, which solves the original question

I was expecting a 100% on *one* core, not both. I got the impression that mpeg encoding was a pipeline rather than 2 threads handing stuff back and forward to each other. Not an mpeg expert, so if you say otherwise that is fine by me.

I understand the xvid core is wonky here, but was just wondering if the systems could go faster. Hence my comment of trying to use the core2 box as if it was 2 single core PCs to get around XVID's problems, and trying to look for other bottlenecks on the AMD box.

I just like things to go fast, it's a weakness I know