CPU Mark - different for XPx86 than Win7x64?

[taiowa]

Hi guys,

last week, I was playing around with my three laptops (all W7home), and the CPU numbers I tested to were within a cat's hair of the norms published.

last night, i tested my A10-5800K on a fresh install of the 32-bit version of XPpro, and the numbers were something like 2800 instead of nearly 4700 - quite a difference.

This MSI H61M-P31(G3) motherboard has been giving me fits with regard to the keyboard (which mostly doesn't work until the USB is unplugged and plugged back-in - and some wireless kbd/mouse kits don't work at all), so my concern is that maybe the MoBo is defective. There's an open case with them on the kbd/mouse issues.

So, should a given CPU test-out at about the same numbers regardless of which version of the OS was in use?

Thanks,
Will Smith
p.s. all OS were fully up to date.

[David (PassMark)]

Yes, 64bit is faster.

See these posts for detail.
http://www.passmark.com/forum/showth...chmark-results

http://www.passmark.com/forum/showth...-maths-amp-PT8

[taiowa]

right, but ...

David,

I had already read those threads before posting. I get that PM7 and PM8 yield different scores...that's clear.

I also get that a program written to do a given task in 64-bit code will complete that task quicker than the same task coded in 32-bit code.

So, is it that when PM8 finds itself running on a 32-bit OS, then it has to presume that the underlying CPU is only 32-bit, and thus runs a 32-bit instance of the program in order to avoid throwing an exception...and the 32-bit instance will yield lower numbers? I never considered that. I thought there was a features test to determine what instructions were available. I'm under the impression that the 64-bit instructions will do just fine on a down-level OS...but I've never tried that. It's on the bucket list now, though!

I'll presume that's the core issue, and I'll load-up a copy of Win7 x64 and see what numbers I get.

Let me ask another question:
For the CPU test set, how much aliasing is caused by other non-CPU characteristics (if any) such as DRAM speed? My expectation is that for a well-cached machine, the whole suite will run in L1, so external DRAM won't make any difference. Is that the case?

Will
...remember: all machines wait at the same speed.

[David (PassMark)]

You can't run 64bit applications inside a 32bit operating system. If this was the case, there wouldn't be much need for a 64bit O/S.

But you can run 32bit applications inside a 64bit operating system (as long as they aren't low level device drivers).

Yes, the CPU L2/L3 cache and the speed of the RAM can impact on the CPU results. How much of an impact depends on the particular test.

[taiowa]

Right. I'm on your wavelength now. The purpose of these benchmarks is to provide a relative measure of the amount of work that a platform will perform in a given amount of time, not to confirm the published specs of a given part, like the CPU. The above A10 did produce the promised RPM in 64-bit mode.

The following results are puzzling, perhaps you can comment:
This little laptop had an AMD N660 cpu. It rings-out at about 1900. The hottest sled in the shed for the S1 socket (at TDP 35w) is the N970 which spins at about 2700...so I bought one.
http://www.cpubenchmark.net/socketType.html#id14

Interestingly, while the N970 does spin 2700, all the other numbers are lower:
PART MARK CPU 2D 3D MEM DISK
N660 700 1991 404 104 959 844
N970 665 2670 310 96 788 836

for those playing along at home, the N660 is a dual core at 3ghz, while the N970 is a quad at 2.2 ghz. AMD's full list of notebook procs is here:
http://products.amd.com/en-us/NotebookCPUResult.aspx

So, the results are in the direction of the clock speed ratios of 2.2/3.0 instead of the core*clock products ratio of 4x2.2 / 2x3.0.

Presuming the benchmarks are a fair representation of real-world performance (and it's clear you guys have spent years agonizing over that mission), then this was a $100+ waste of money...or am I missing something?

Thanks,
Will Smith

[David (PassMark)]

Probably should have looked at the single threading benchmark chart before the purchase,
http://www.cpubenchmark.net/singleThread.html

The N660, at the higher clock speed, does significantly better than the N970.

If the N970 is better or not depends on what you are using the PC for. If you are using a lot of software that is CPU bound and is mult-threaded, then the N970 is better. Otherwise use the N660.

[taiowa]

You guys are really on it. The single-thread numbers from one processor times the bus speed ratio is exactly the single-thread number for the other one.

What you're saying is that the dominant factor in measuring the remaining features other than the CPU is really the bus speed of the CPU...it's the dominant source of aliasing between tests.

In other words, there's a CPU bus speed (and/or CPU performance level) above which no improvement in a disk transfer test will be seen, and below which the CPU has insufficient resources to "max-out" the disk, so the results of the disk test will be "aliased" by the CPU.

I think I might have seen the disk test drop-off significantly when slower memory was installed, but it was not as bad as the "blow-a-rev" scenario of the past decades (before read-ahead).

This laptop occasionally gets to boot-up in Linux, but then, even when compiling a full driver suite, it's breathing heavily only for a minute or two...and I've not watched the system monitor to see if the CPU maxes-out, but I doubt it does. Otherwise, I'm in W7 surfing the web, watching videos, doing emails, and the occasional spreadsheet...none of which is going to tax the system's performance.

I'll play with the N970 for a week or so, and you're probably right - the N660 will get re-installed, and the N970 will find its way to Ebay - along with the super fast memory that also made no difference.

Is there a way to browse the PCMark ratings for given CPU types in their systems, like a min-avg-max page like the single-threaded page?

Thanks for the guidance, and for letting me play with your dynomometer...it's just a side-effect of testosterone poisoning. That's why we race cars, boats, and lawnmowers.

Will Smith
my other car is a Checker - "livin' large in the slow lane."

[David (PassMark)]

Not sure if 'aliased' is the right word.
But reducing in the CPU's clock speed will impact other areas, like 2D graphics and RAM speeds. The impact on the disk speed will be less noticeable.

The reverse is also true, higher CPU performance will increase performance in other areas, but only until a certain point, after which performance improvements will flatten off as you hit other bottle necks, like the I/O speed of the disk.

You can search for baseline files from within the PerformanceTest software to see the benchmarks from other machines.