First time mac user

Tern
Tern
Joined: 27 Jul 05
Posts: 309
Credit: 71,180,862
RAC: 29

RE: There is another

Message 19936 in response to message 19934

Quote:

There is another powermac11,2 I am comparing my results to. Checkout knightrider in the list of top computers, he is currently number 1 (except for the two bogus entries).

Where would I find performance vs power setting?

Ok, you got me very curious, so I did some digging...

His average time is 3 hours and 53 minutes! That _IS_ twice the speed of a Mini!

He has 2GB of RAM, which would account for some of the difference; 512MB is fine for one CPU, but with four processes running, you're probably slamming the drive with virtual memory swapping. He is also running _some_ kind of optimized client, as he's claiming 172 credits per result - which frankly is ridiculous when the project average is somewhere around 70, but it's getting him a granted average of about 80. Actually, his results say he's running 4.44, which had several optimized versions available, some of which were "invisible", so I can't tell exactly what he's running. Regardless, by claiming that much, he is going to always be the "top claim" on a result, and be thrown out, along with the lowest claim. So the "second highest" will always set the credits awarded. His benchmarks of 4600/16500 are lower than yours were at 6800/23500 when you had the 300-credit claims... There is no "rule" against running optimized clients, but even his 172-credit claims are going to raise eyebrows. Claiming 300 is just asking to be accused of cheating - running one to _equalize_ the field is one thing, running one to _dominate_ the field is another. As your results are taking almost twice as long, you're going to be asking for almost twice the credit you should; if you find and fix the 7-hour issue and get down under 4, then the "superbench" _might_ be fine, but until then, I'd definitely avoid it.

As for performance settings, I don't remember exactly, it's been at least two OS versions ago... I just now looked in System Preferences/Energy Saver, and on the Mini, there's nothing like what I remember seeing on the Dual. The iBook is closer, with settings for "better performance" or "better energy savings" - this may be what you're looking for. The control panels are different for different CPUs, giving access to different features and settings.

If you want to be up there with Knightrider, I suspect you'll need to go to 'best performance' on all the settings, _and_ get more RAM. And be careful on the _type_ of RAM, as well; you're into overclocked-PC type area with the Quad, and need _really_ stable RAM, not the bargain-bin stuff I can get away with using. Einstein wants about 60MB per process; the OS kernal wants over 700MB. So any machine with less than a GB _will_ be running in VM a lot. If you want to find out if the RAM is really the problem, it's simple to do - just set your preferences to allow BOINC to only use 1 CPU, and see how long the results take.

Keep us informed - there are a lot of us who firmly believe that the Quad may just be _the_ best machine for BOINC. It'd be nice to know exactly what it takes to get the best performance out of one.

MarkF
MarkF
Joined: 12 Apr 05
Posts: 393
Credit: 1,516,715
RAC: 0

Bill Thanks, I cranked up the

Bill
Thanks, I cranked up the performance option under preferences->power saving. I don't think memory it is a problem I am currently using less than half of it according to the activity monitor.
(I am sllllowly begging to learn macies)
I agree the claimed credit is out of line. I have removed the optimized boinc client.
I will let you know what effect the change has.

ps I would welcome any information on other performance settings.

MarkF
MarkF
Joined: 12 Apr 05
Posts: 393
Credit: 1,516,715
RAC: 0

Thanks to all for the

Thanks to all for the help.
The option is system perferece/energy saver/optins/processor performance/highest
While I have never been a Mac fan, this time thier choice of features matches my goals. I am truly happy about the return on my investment.

Tern
Tern
Joined: 27 Jul 05
Posts: 309
Credit: 71,180,862
RAC: 29

RE: Thanks, I cranked up

Message 19939 in response to message 19937

Quote:
Thanks, I cranked up the performance option under preferences->power saving.

Well, something is working - even though I suspect the results you just returned were started well before you made this change, they ran in about 4.2 hours instead of 7... :-)

As for RAM - in Activity Monitor, make sure you have both Real Memory and Virtual Memory visible (from the View menu). You ideally would want the Einstein processes to have the highest possible Real value with the lowest possible Virtual, but even with massive RAM, that just won't happen. You have to actually select the Einstein processes and "inspect" them, looking at the Statistics tab. The key is "Page Ins" - when I look at this on my iBook G3 (384MB and running many apps) the number is 387. When I look at it on the Mini (512MB and running only BOINC) the number is 0. Zero is definitely what you want. Faults and context switches are also bad things - the fewer the better. I don't have a Mac at the moment with what I would call "enough" RAM (my Dual had 2GB, and I wished it had more...) so I can't compare to see what these should be, as even the Mini shows them being pretty high (millions, 3 hrs in).

When you say you're using only half the memory per Activity Monitor - you're looking at "free" on the System Memory tab? I don't have any idea why, but it seems the Mac will always have at least 10-20% free, sometimes much more, yet the much-more-important "page ins/outs" number below that will be way too high. I suspect that because the BOINC processes are at such a low priority, they get shuffled out _too_ easily. If page ins/outs isn't climbing, then you're okay. If it _is_ climbing, then I'd say you need more RAM. But hey, RAM for the Quad isn't cheap (if it's like my Dual, you want _4_ matching sticks at a time, not just 2), and if you can get 4 results every 4 hours with 512MB, I wouldn't complain a bit!

MarkF
MarkF
Joined: 12 Apr 05
Posts: 393
Credit: 1,516,715
RAC: 0

Bill You are correct about

Bill
You are correct about changing the option after the WUs had started. They where slightly more than an hour into their crunch time when I suspended then changed the option. Based on that and extprolation I feel confident that I will be getting the performance I expected. I may gain some points with more physical memory but for the moment I will forgo the expense. I will update the thread as events evolve.
For the moment PowerMac11,2 rules.
Again thank you and Michael for your assistence (+ anyone I forgot).

Paul D. Buck
Paul D. Buck
Joined: 17 Jan 05
Posts: 754
Credit: 5,385,205
RAC: 0

RE: But hey, RAM for the

Message 19941 in response to message 19939

Quote:
But hey, RAM for the Quad isn't cheap (if it's like my Dual, you want _4_ matching sticks at a time, not just 2), and if you can get 4 results every 4 hours with 512MB, I wouldn't complain a bit!


Interesting comments.

How do I know if I need 4 sticks instead of two? On my G5 the book says that I only have to have pairs.

On my G5, only a dual, 2.0 GHz; I am getting about 4 hours 30 Min per work unit. So, a 4 hour timing is reasonable ...

Tern
Tern
Joined: 27 Jul 05
Posts: 309
Credit: 71,180,862
RAC: 29

RE: How do I know if I need

Message 19942 in response to message 19941

Quote:
How do I know if I need 4 sticks instead of two? On my G5 the book says that I only have to have pairs.

If your manual says pairs, then you're fine with pairs... the Dual I used was bought through the developer purchase program and was one of the very first ones - either they changed something since then, or (also entirely possible) I'm remembering the recommendations wrong.

I note that a couple of Quads have already made their way to the top-20 on SETI too. :-) Quite a machine. Imagine what they'd be doing if IBM had actually delivered that promised 3GHz version!

Michael Roycraft
Michael Roycraft
Joined: 10 Mar 05
Posts: 846
Credit: 157,718
RAC: 0

RE: RE: How do I know if

Message 19943 in response to message 19942

Quote:
Quote:
How do I know if I need 4 sticks instead of two? On my G5 the book says that I only have to have pairs.

If your manual says pairs, then you're fine with pairs... the Dual I used was bought through the developer purchase program and was one of the very first ones - either they changed something since then, or (also entirely possible) I'm remembering the recommendations wrong.

I note that a couple of Quads have already made their way to the top-20 on SETI too. :-) Quite a machine. Imagine what they'd be doing if IBM had actually delivered that promised 3GHz version!

Bill,

Just out of curiosity, does Mac have no provision for overclocking?

microcraft
"The arc of history is long, but it bends toward justice" - MLK

Tern
Tern
Joined: 27 Jul 05
Posts: 309
Credit: 71,180,862
RAC: 29

RE: Just out of curiosity,

Message 19944 in response to message 19943

Quote:
Just out of curiosity, does Mac have no provision for overclocking?

Not... exactly... The "official" answer is "No, you can't overclock a Mac". There isn't a "BIOS" on the motherboard for example where you could just set the clock speed. This is needed on PCs because the same motherboard could be used for a wide range of different CPUs - Macs tend to have a custom motherboard designed for each model.

However - _sometimes_, the same motherboard will be used for different speeds within the same model. For example, the Mini ships with either a 1.25GHz or 1.42GHz option. Changing the motherboard from one to the other is "hardware", not a software setting somewhere, but it's fairly easy to do. The risk is that the CPU used in your 1.25GHz model was one that didn't "pass" at 1.42GHz, so changing the setting could cause errors. (And for whatever reason, _some_ Minis were actually shipped set at 1.5GHz, with the box still marked 1.42, so obviously the motherboard can handle more than 2 speeds.)

There are obviously other things that people do to get more speed, but in general, it's hardware changes and hacks, and not just a software setting. And usually the gain is fairly small, because Apple balances everything so carefully to get the maximum performance to begin with. The biggest exceptions are usually when a model was "supposed" to have, say, a 2.5GHz CPU, but the vendor couldn't supply enough, so Apple only released that model at 2GHz. Then if you can find a 2.5GHz chip, you can replace it and tweak the MB and have a "Model XYZ" Mac running 25% faster than any others of that model.

As far as overclocking in general on a G4/G5, unlike say an AMD 3700 "San Diego", there just isn't much room to do so even if you could. I get the impression that a _lot_ of AMD and Intel CPUs are "labeled" based on demand; that they're internally "able" to run faster but are sold based on market demand rather than actual reliability at a certain speed. In the G4/G5 market, if a chip could pass the tests (in quantity, not individual specimens) to be sold as a 1.5GHz rather than a 1.42GHz, believe me, they'd do it.

Michael Roycraft
Michael Roycraft
Joined: 10 Mar 05
Posts: 846
Credit: 157,718
RAC: 0

Bill, Once again, thank

Bill,

Once again, thank you for the education. :-) Doggone, there goes that 3GHz G5.. thought that would've been attainable, only an 11% kick. You're correct about AMD and Intel - the practice is known as "speed-binning", and it starts almost immediately when a new core debuts. Enough of that though, I don't want to further disrupt this Mac thread.

microcraft
"The arc of history is long, but it bends toward justice" - MLK

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.