Why is my i7 system so slow in SD MPEG encodes in CS5?

[Randall Leong]

Originally Posted by Harm Millaard

Another thing to remove your doubts, look at the consistency of results. The AMD score is purely caused by the fact that AMD misses sorely needed instructions for MPEG encoding. The other low score with CS5 is caused by a slow CPU, very limited memory and an inadequate disk setup.

Well, the slowpoke Intel CS5 system not only has a slow CPU (albeit one that's overclocked), it has much less L2 cache than a Q9650 (which also ranked lower than all of the other MPE systems though it had better MPEG scores than the Q8300 system). Plus, no 150GB 7200 RPM hard drive has as fast of a raw transfer speed as most of the newer 500GB hard drives; in fact, two 150GB hard drives even in RAID 0 might actually still be slower than a single 500GB hard drive.

In addition, the GTX 470, which did so well with MPE enabled, proved to be a slowpoke in software-only mode (as far as the timeline render test results are concerned) even compared to the ATi-equipped systems on the list. The relatively immature drivers for the GTX 4xx series may be partly to blame.

[Randall Leong]

Originally Posted by Randall Leong

Well, the slowpoke Intel CS5 system not only has a slow CPU (albeit one that's overclocked), it has much less L2 cache than a Q9650 (which also ranked lower than all of the other MPE systems though it had better MPEG scores than the Q8300 system). Plus, no 150GB 7200 RPM hard drive has as fast of a raw transfer speed as most of the newer 500GB hard drives; in fact, two 150GB hard drives even in RAID 0 might actually still be slower than a single 500GB hard drive.

In addition, the GTX 470, which did so well with MPE enabled, proved to be a slowpoke in software-only mode (as far as the timeline render test results are concerned) even compared to the ATi-equipped systems on the list. The relatively immature drivers for the GTX 4xx series may be partly to blame.

I was wrong on the last paragraph there. It turned out that the slowpoke in question was running with a less-than-optimal memory configuration - 16GB of RAM via four 4GB DIMMs running in a "Flex" memory controller mode. In that mode, 12GB had been running in triple-channel mode while the remaining 4GB had been running in single-channel mode. This mixture of memory controller modes significantly increases the overall latency of the i7's memory controller, resulting in these slowdowns.

And now I think I know why my results that I had submitted still have not shown up on the PPBM4 list on the site: I failed to specify that the MPE was turned on (GPU accelerated mode) or off (software-only mode). So I had to rerun my tests (this time, the two 1TB Seagate drives which used to be in a RAID 0 array were then configured as separate drives for this test), and I now got an overall score of 33.6 sec., with the AVI performance at 6.7 sec, the MPEG time of 7.9 sec. and the rendering time of 19 sec. (and this is with no RAID). I am currently rerunning the same system with a two-disk RAID 0 array, and report back to you.

An update with the RAID: As it turned out, putting the two drives in the RAID 0 shaved off only 1.8 sec. off of the AVI encode times. The MPEG encode and the render times remained about the same. Hence, I can concur with Harm that a two-drive RAID 0 array is not worth the trouble unless one or two additional drives (configured as separate volumes) are added to the system. However, this is applicable only for SD video editing. With HD video editing, I will have to wait for the release of PPBM5 before I can jump to any conclusions. (I resubmitted only the results with three separate drives and no RAID.)

[Randall Leong]

...I discovered that I have too little RAM in my machine, so my MPEG-2 DVD performance is still below par for an i7-920 overclocked to 3.7 GHz. With a score of 185 seconds on that test, it is slower than even an AMD system in that test. Otherwise, the 83-second score for the disk I/O is as expected for only a two-disk RAID 0, while the scores for H.264 Blu-ray and CPU/GPU MPE Off are also below par. And even if I upgraded the RAM to 12GB, the GPU and disk scores would have still limited my system's performance.

In other words, an astronomically expensive system (priced at well over $5,000 total with more than four HDDs and a dedicated hardware RAID controller) is required in order to achieve anywhere near top scores in the PPBM5 list.

Hence, I will be submitting several sets of scores for PPBM5 as soon as I re-run my tests with more RAM and/or an NVIDIA GPU.

[Randall Leong]

It is too little RAM in my system. Today I upgraded from the 6GB that my system had been running for several months to 12GB. Although my system did not overclock as well with six sticks of RAM as it did with only three sticks of RAM, my MPEG-2 encodes took about half the time that it did with only 6GB. (And this is with the CPU and memory at the default stock speeds.)

I will gradually overclock the CPU with this new configuration, and submit to Bill all of the sets of results that I kept.

Another update:

I finally got the 12GB system stable at an overclock to 3.5GHz. My test scores in PPBM CS5 is even better than at stock although its 427-second total time is primarily due to the lack of MPE GPU acceleration. I currently have an HD 5770, but would love to get a good NVIDIA card.

In short, although PPro CS5 can run with as little as 4GB of RAM, I strongly feel that it needs at least 12GB of RAM to run at its best. (I confirmed this after running the PPBM4 test after I upgraded the system to 12GB of RAM. No wonder why I had to cheat in order for my system to achieve a good score with only 6GB of RAM.)

[Randall Leong]

I have recently "upgraded" the CPU in my main rig to an i7-950. This meant that I decided to rebuild the original i7-920 rig with the Intel-brand DX58SO motherboard and 6GB of RAM. And since my current GTX 470 was my only CUDA card with sufficient RAM to run MPE's GPU mode, I went out and picked up a 1GB GT 240 with GDDR5 memory on clearance.

I ran the PPBM5 test with the CPU locked at its stock 2.67GHz (Premiere Pro CS5 5.0.2), and to my surprise I got an MPEG-2 DVD score of around 145 seconds. Not as fast as I would have gotten with 12GB, but quite respectable for a stock-speed i7 with only 6GB. The MPE timeline rendering score was also very respectable (at 14 seconds), particularly for a stock-speed i7 with such a cheap graphics card. If there is a weak point to that system, it was the AVI (disk) performance: This auxiliary system does not currently have a RAID array for the work drive, resulting in slower-than-expected disk performance.

Overall, I am quite happy with this auxiliary rebuild. An overclock to what I had the CPU at when the 920 was in my main rig, a faster disk setup and double the RAM would have put this system too close to the CS5 performance of my main rig for comfort, so I decided to leave it at stock speed with only 6GB. The only upgrade that I have planned for that system for the foreseeable future is a faster disk setup (two or three identical fast 7200 rpm disks in RAID 0).

[Randall Leong]

Originally Posted by Randall Leong

I have recently "upgraded" the CPU in my main rig to an i7-950. This meant that I decided to rebuild the original i7-920 rig with the Intel-brand DX58SO motherboard and 6GB of RAM. And since my current GTX 470 was my only CUDA card with sufficient RAM to run MPE's GPU mode, I went out and picked up a 1GB GT 240 with GDDR5 memory on clearance.

I ran the PPBM5 test with the CPU locked at its stock 2.67GHz (Premiere Pro CS5 5.0.2), and to my surprise I got an MPEG-2 DVD score of around 145 seconds. Not as fast as I would have gotten with 12GB, but quite respectable for a stock-speed i7 with only 6GB. The MPE timeline rendering score was also very respectable (at 14 seconds), particularly for a stock-speed i7 with such a cheap graphics card. If there is a weak point to that system, it was the AVI (disk) performance: This auxiliary system does not currently have a RAID array for the work drive, resulting in slower-than-expected disk performance.

Overall, I am quite happy with this auxiliary rebuild. An overclock to what I had the CPU at when the 920 was in my main rig, a faster disk setup and double the RAM would have put this system too close to the CS5 performance of my main rig for comfort, so I decided to leave it at stock speed with only 6GB. The only upgrade that I have planned for that system for the foreseeable future is a faster disk setup (two or three identical fast 7200 rpm disks in RAID 0).

By the way, 5.0.2 appears to work much better than 5.0.0 or 5.0.1 with less than 12GB of RAM. In my testing, I got an MPEG-2 DVD encoding time of about 120 seconds at 3.4GHz with 6GB of RAM running 5.0.2 compared to the over 180-second result at 3.7GHz with the same 6GB of RAM running 5.0.1.