Thoughts about the SuperPI benchmark program

So, we've all been testing our laptops (and desktops too) with it. So have I, and there are still some things unclear to me.

(1) Put aside the AC versus battery performance, I've noticed that even when I run the test over and over again (always plugged in), under the exact same conditions, I sometimes get noticeable fluctuations in the results. But that's only on my laptops - it seems pretty consistent on my desktop.

For instance, on my P-M, the 2M test took between 1:50 and 2:00. Sometimes a fast result followed by a slow one, sometimes vice versa. On my P4-M I got results between 3:10 and 3:21.

(2) I assume that the speed of the RAM would make some difference here. For example, I have DDR333 in my Thinkpad, and I think that if I had DDR400, it'd be faster. Am I wrong?

(3) Does the amount of RAM make a difference here? I wouldn't think so, because SuperPI shows it allocates very little memory for the benchmark.

(4) For those who ran tests other than 2M, it's easy to see that the times increase nonlinearly with the number of digits, i.e. 2M usually takes more than twice the time that 1M takes, and that in itself isn't too strange. However, I did notice that a PC that may be faster for short tests will be slower for long tests.

For instance, I compared my 3.0 Northwood to my 1.8 Dothan.

Here are the results for 512K to 8M:

Code:

Digits    Northwood 3.0    Dothan 1.8
512K          0:21            0:16
  1M          0:47            0:44
  2M          1:49            1:50
  4M          4:02            4:21
  8M          8:32            9:26

So as you see, the Dothan is faster in the beginning, but the Northwood pulls far ahead in the longer tests. Is this due to the simple fact that the Northwood has twice the RAM (1GB over 512MB) which affects longer tests more (if that's even true, I'm not sure), or is it due to the architecture differences between the processors?

I wonder.

 

One thing that you might not have considered. Since this is a CPU intensive application, your notebook might be reaching critical temps on long tests and hence the CPU is clocking down to protect itself. And maybe the RAMs do make a difference for longer tests.

I'll give it a shot on mine in a while.

 

I'm resurrecting this thread by right of thread starter.

Things got even weirder. I installed a second 512MB stick in my laptop, and at first it seemed that it actually ran slower. I started seeing times over 2 minutes (up to 2:05 sometimes).

Make no mistake, the second stick has the same specs as the first one (speed, latencies). It is a NoName stick rather than a brand one, but I'm not sure whether it should affect performance. I ran MemTest for a while and it found no errors, so the stick performs fine (apparently), but it just seemed that if I take it out, SuperPi runs a bit faster on average.

But what's weirder? Usually I get times around 1:55-2:00 now (with both sticks in). Once I ran a test and it gave me 1:48, which is an all-time low for the laptop. And it was with lots of processes in the background and a couple of Firefox windows open (idle, though).

Now, my desktop - more weirdness. Remember how I said that it showed consistent times test after test? Well, I upgraded to SP2, disabled some background services and ran the test again. At first it gave me 1:47, compared to the usual 1:49. I played with the PC a bit, ran the test again, and it got down to 1:42 without any explanation. It seems to run a bit faster overall, here are the updated times:

Code:

Digits    Northwood 3.0    Dothan 1.8
512K          0:20
  1M          0:46
  2M          1:42            1:48
  4M          3:50
  8M          8:29

Why doesn't this thing behave consistently?

 

my tupenny....
I actually go with Quester a little bit then veer a little....I think heat can be an issue....like quester says, cpu may be under clocking abit...but i also suspect ram coolant's(the north bridge coolant, as it's called) role in this...do laptops have as good coolants as desktops? Also, the noname ram, while being good at latencies etc, can be producing more heat, there by reducing the speed of both the sticks....The IC design of this ram can be one of the reasons for heat....i don't point to bad design, but to outdatedness...
well, i said a lot without being too convinced myself...

 

Yeah, if I was sure that it is due to the other stick, I'd just take it back and demand them to replace it with something else. The problem is that because of all the inconsistensies I cannot be sure that it has anything to do with it. After all, the laptop's best time was shown with both sticks in there. I may run some more tests with just one of the sticks later. Can't be bothered right now...

And how would heat directly affect the speed? I mean, not as a result of underclocking. Is there like "unvoluntary underclocking" that takes place with hardware when it gets hot?

 

They become inconsistent when they get hot. in the case of ram, once it gets to the critical temperature, it may drop a couple of 'bursts' and since it thus gets inconsistent, the cpu automatically clocks to the lowest speed that the ram is functioning at(under the inconsistency). CPUs are like that...it adjusts itself to the max safe speed that the ram can provide....It cannot demand from ram...it's the one that has to adjust...
we may call this 'involuntary underclocking'....
that is where the ram coolars(ram fans) come into picture...to maintain certain temp on the northbrige...but say if one IC is made in a way that it gets hot faster than the other one, cpu voluntarily takes in less from ram(both sticks), though the cooler stick may still be in fine fettle...
having said all this, I am not sure if this is the reason for the actual question, i can't also imagine rams getting too hot easily...
why don't you try not running any programs, and disabling background services(dunno if it's safe---check---) and disabling indexing etc?

 

I used the "How to optimize a notebook computer" guide the other day, that really is excellent. I saw a 20% speed boost.

Chaz

 

I actually came back to these forums after a relatively long break to report my latest findings regarding this thing.

Today I tried to replace the noname stick with another 512MB Samsung stick (almost identical to the primary stick of my Thinkpad). Well, it showed the exact same results. So it wasn't due to the noname RAM being problematic.

What I can conclude is this: when you run SuperPI immediately after boot, the results with two 512MB sticks are about 2-4% slower than with one 512MB stick. And that's pretty consistent. Now, if you run it at various conditions, it seems to vary a lot. I actually got 1:46 with two sticks once...

Now the reason to this phenomenon is unclear to me. It could be due to Windows working a bit slower because the addressing space is twice as large. It could be due to purely physical factors (2 sticks slower than 1). It could be a glitch of my specific installation on my specific laptop.

Definitely not anything worth raising a fuss about, though.

 

things will get more interesting next year with the new Intel processors, one thing they're claiming is that after a program has been run a few times the processor will become "better" at processing those instructions, so theoretically Super Pi speeds should get better after 2-3 runs. We'll have to see.

 

I.e., a processor that learns to adjust its branching predictions and various optimizations, based on some short-term memory. Sounds cool. From some reason it made me think of Schwarzenegger and his "Neural Network Processor" in The Terminator series. Hehe...

 

Sure, RAM plays the role. You can see it by running this test together with some RAM control software, as I did. Somewehere above 5M you need more than 512MB RAM to run the test without using the HDD.

 

Possibly... I'll rerun some of the longer tests at some point and see.

 

Load Bionic CPU Peeker v2.7 and run it with superpi and see it yourself!