Promise IDE/Intel IDE comparison - PATA - P4C800E-Deluxe

[Noozer]

I ran some bencmarks earlier with Sandra 2004 against my hard drive. I was
pretty shocked with the difference. Since I'm waiting for the download of
the Suse 9 Live Eval and MandrakeMove 9.2 Linux CD's I thought I'd post the
results here.

System:
- P4 2.6Ghz running 3.2Ghz
- 512meg DDR533 memory @ 1:1
- Maxtor 6Y080L0 PATA drive (80gig 2meg cache)
- ATI 9600XT video card

I ran the benchmark once using the Windows cache, then twice without the
cache. Then I shut down, moved the IDE cable from the PROMISE PATA connector
to the standard Primary IDE connector and rand the same tests. Here are the
results...

*** PROMISE CONTROLLER ***
Windows Disk Cache Used : Yes | No | No
Benchmark Breakdown
Buffered Read : 733 MB/s | 1273 MB/s | 1200 MB/s
Sequential Read : 26 MB/s | 33 MB/s | 33 MB/s
Random Read : 8 MB/s | 6MB/s | 6MB/s
Buffered Write : 529 MB/s | 1141 MB/s | 1200MB/s
Sequential Write : 18 MB/s | 34 MB/s | 34MB/s
Random Write : 7 MB/s | 9 MB/s | 9MB/s
Average Access Time : 6 ms | 9 ms | 9ms (estimated)


*** INTEL CONTROLLER ***
Windows Disk Cache Used : Yes | No | No
Benchmark Breakdown
Buffered Read : 67 MB/s | 70MB/s | 83MB/s
Sequential Read : 19 MB/s | 30MB/s | 36MB/s
Random Read : 8 MB/s | 5MB/s | 6MB/s
Buffered Write : 73 MB/s | 74 MB/s | 74MB/s
Sequential Write : 30 MB/s | 35MB/s | 37MB/s
Random Write : 3284 kB/s | 7MB/s | 8MB/s
Average Access Time : 5 ms | 10ms | 10ms (estimated)

Notice the HUGE differences in buffered read/write times!!!! Guess the
drive is going to stay on the Promise controller.

 

[JJLKJ]

Do you have any idea what you are talking about? Did you even look at those
benchmark results? First of all I have run benchmarks on the same board, on
both controllers, and performance is nearly identical, both with and without
raid, definitely nothing that you would probably notice in every day use.
Second, the benchmarks you have for the Promise controller are not even
remotely possible. 1200MB/s reads and writes? You do realize that your
drive has a "theoretical" max of about 100-133MB/s depending on the model,
and that you will never even achieve close to that in a non-raid setup.
Usually you will get around 80% of the theoretical, or at least in that
ballpark. Considering you got about 1000% of the theoretical max, I think
you must be doing something very wrong!!!

 

[kony]

Do you have any idea what you are talking about? Did you even look at those
benchmark results? First of all I have run benchmarks on the same board, on
both controllers, and performance is nearly identical, both with and without
raid, definitely nothing that you would probably notice in every day use.
Second, the benchmarks you have for the Promise controller are not even
remotely possible. 1200MB/s reads and writes? You do realize that your
drive has a "theoretical" max of about 100-133MB/s depending on the model,
and that you will never even achieve close to that in a non-raid setup.
Usually you will get around 80% of the theoretical, or at least in that
ballpark. Considering you got about 1000% of the theoretical max, I think
you must be doing something very wrong!!!

The buffered read/write scores are basically determined by the speed
of the memory, amount of cache, not the HDD's sustained throughput.

Even so, there was something wrong with the Intel controller, it
should actually score slightly higher than the promise card given same
ATA rate and integrated into southbridge.

 

[Noozer]

Do you have any idea what you are talking about?

I don't know... Do you know what you're talking about?

Did you even look at those
benchmark results? First of all I have run benchmarks on the same board, on
both controllers, and performance is nearly identical, both with and without
raid, definitely nothing that you would probably notice in every day use.
Second, the benchmarks you have for the Promise controller are not even
remotely possible. 1200MB/s reads and writes? You do realize that your
drive has a "theoretical" max of about 100-133MB/s depending on the model,
and that you will never even achieve close to that in a non-raid setup.
Usually you will get around 80% of the theoretical, or at least in that
ballpark. Considering you got about 1000% of the theoretical max, I think
you must be doing something very wrong!!!

I booted the PC, ran the benchmark and pasted the results to notepad and
saved the file. Then I shut down, moved the IDE cable to a different
connector on the mainboard, booted and did the exact same thing. I just
tried again and still see similar numbers.

Kinds hard to screw up, don'tcha think?

The whole reason I posted is that I *DON'T* understand why there would be
such a difference. I'd be glad to know why the numbers are so different.

 

[stacey]

I don't know... Do you know what you're talking about?


I booted the PC, ran the benchmark and pasted the results to notepad and
saved the file. Then I shut down, moved the IDE cable to a different
connector on the mainboard, booted and did the exact same thing. I just
tried again and still see similar numbers.

Kinds hard to screw up, don'tcha think?

The whole reason I posted is that I *DON'T* understand why there would be
such a difference. I'd be glad to know why the numbers are so different.

Benchmark software is screwy. BTW looks like as far as real throughput, the
intel controller is faster.

 

[JJLKJ]

The buffered read/writes are still limited by disk speed. You won't exceed
the capability of the drive period. I have seen mention of the benchmark
giving unusually high results on Windows 2000, without SP2, although I
haven't seen how unusually high they were talking. Maybe there's something
else going on, but I can guarantee the 1200MB/s buffered read/writes are not
"real" results. It should be obvious that the 1200MB/s benchmark is a bad
reading for whatever reason. Believe me, you wouldn't have been the first
to discover that the Promise controller performed more than 10 times better
than the Intel. Like I said, I ran the benchmarks and the readings come out
just like they should, well under the theoretical max of the drive speed.
I've never run a Sisoft Sandra Benchmark, on ver. 2002-2004, where the
buffered read/writes exceeded what the disk could do. My point is not that
you made a mistake, but that for some reason the benchmark obviously is
giving a false reading on your system. Just didn't want you to committ to
the promise controller based on that benchmark, or spread the word to other
readers that the promise controller is much better, when in fact it is very
similar in performance to the Intel.

I actually stick with the Intel controller because I know I will always be
able to get updated drivers from Intel, and Intel chipset support is always
good in Linux too. When you have an onboard controller from a third party,
it often becomes a problem in the future. You often end up not being able
to use a driver or controller bios from the chip manufacturer, but having to
wait for an update from the motherboard manufacturer. In almost every case
I have ever seen, the motherboard manufacturer stops providing updates, long
before the chip manufacturer does.

 

[Mark]

pasted the results to notepad and saved the file.

Just a suggestion. Usenet posts are traditionally and still most often
done with a non-proportional type font. Outleak, among its many other
issues, defaults to the wrong type. So your pasted results are not
aligned very well.

As to the drive differences, I looked at the random reads/writes. In
that case, they are almost identical. But you really should look for
better HD benchmark software. The name of one escapes me at the moment,
but it is used specifically for testing drives. I believe I read about
it over at AnandTech.