Memory speed; cache size?

ADDMP
ADDMP
Joined: 25 Feb 05
Posts: 104
Credit: 7332049
RAC: 0
Topic 188740

Does anyone have any data or info on how important memory speed is for E@H?

And CPU on-chip cache size?

Thanks.

ADDMP

scubyfan
scubyfan
Joined: 20 Feb 05
Posts: 18
Credit: 10725
RAC: 0

Memory speed; cache size?

> Does anyone have any data or info on how important memory speed is for E@H?
>
> And CPU on-chip cache size?
>
> Thanks.
>
> ADDMP
>

It seems the L2 cache isn't very important, at least that's the result from a "test" in the WU Speeds thread, which compared a 512Kb and 1 Meg L1 with similar clock speed and got the almost same time. Dunno about the memory speed. :)

ADDMP
ADDMP
Joined: 25 Feb 05
Posts: 104
Credit: 7332049
RAC: 0

> > Does anyone have any data

Message 9417 in response to message 9416

> > Does anyone have any data or info on how important memory speed is for
> E@H?
> >
> > And CPU on-chip cache size?
> >
> > Thanks.
> >
> > ADDMP
> >
>
> It seems the L2 cache isn't very important, at least that's the result from a
> "test" in the WU Speeds thread, which compared a 512Kb and 1 Meg L1 with
> similar clock speed and got the almost same time. Dunno about the memory
> speed. :)
>

Thanks, that's useful to know. I'm just interested in what sort of changes might be helpful & which might not.

ADDMP

Paul D. Buck
Paul D. Buck
Joined: 17 Jan 05
Posts: 754
Credit: 5385205
RAC: 0

> It seems the L2 cache isn't

Message 9418 in response to message 9416

> It seems the L2 cache isn't very important, at least that's the result from a
> "test" in the WU Speeds thread, which compared a 512Kb and 1 Meg L1 with
> similar clock speed and got the almost same time. Dunno about the memory
> speed. :)

Every time *I* look at that thread it seems to me that people are posting the benchmark times rather than the average time for the work units themselves ...

I *DO* know from personal experience that there can be as much as a one hour difference in processing times for SETI@Home work units and the only difference between the systems is the motherboard ... with the most likely issue being one had dual channel memory access and the other did not ...

scubyfan
scubyfan
Joined: 20 Feb 05
Posts: 18
Credit: 10725
RAC: 0

Both systems in the test I

Both systems in the test I referred to had dual channel memory, though it obviously wasn't scientific. I should correct myself that I meant to write L2 in my post, not L1. :)

Paul D. Buck
Paul D. Buck
Joined: 17 Jan 05
Posts: 754
Credit: 5385205
RAC: 0

> Both systems in the test I

Message 9420 in response to message 9419

> Both systems in the test I referred to had dual channel memory, though it
> obviously wasn't scientific. I should correct myself that I meant to write L2
> in my post, not L1. :)

I was almost going to launch an expiriment to test some of these issues with a practical expiriment. But, I just received a complaint about the electric bill, so, I guess I will not buy 4 computers ... :(

I will just get one ... :)

A dual-Xeon workstation ...

John McLeod VII
John McLeod VII
Moderator
Joined: 10 Nov 04
Posts: 547
Credit: 632255
RAC: 0

It really depends on the

It really depends on the project. For instance S@H does some better with a 1MB cache vs a 512 KB cache and a 256Kb cache really hurts. I expect that there is some threshold below which Einstein starts thrashing and performance drops dramatically.

Flashdrive
Flashdrive
Joined: 24 Feb 05
Posts: 2
Credit: 10096
RAC: 0

> It really depends on the

Message 9422 in response to message 9421

> It really depends on the project. For instance S@H does some better with a
> 1MB cache vs a 512 KB cache and a 256Kb cache really hurts. I expect that
> there is some threshold below which Einstein starts thrashing and performance
> drops dramatically.

Any increase on the L1,L2,And (if your CPU has it)L3 memory(cache)( is going to have a significant benefit because of the time the CPU takes to retreive data from the Main memory.The more on chip memory(cache) the more frequently used functions it can store before having to go to the main.Another way to increase speed is the latency of the main memory.Ussually the smaller the number the better.Say 2.5CL or 2CL over 3CL is best.Over head on the main memory is important as well.Boinc reguires about 256M to run but you have to remember other apps running in the background as well. 512M is normally good but is becoming less and less adequate especally if your running newer games.

ADDMP
ADDMP
Joined: 25 Feb 05
Posts: 104
Credit: 7332049
RAC: 0

> Boinc reguires about 256M

Message 9423 in response to message 9422

> Boinc reguires about 256M to run but you have to
> remember other apps running in the background as well. 512M is normally good
> but is becoming less and less adequate especally if your running newer games.

I have a computer running BOINC E@H with 128 MB RAM & WIN98, usually without much else running. Whether there is a significant penalty for such a small amount of memory is not clear. The CPU is an AthlonXP at 1700 MHz & does an E@H work unit in 30500 sec. I tried win2k first & it would not run.

ADDMP

Keck_Komputers
Keck_Komputers
Joined: 18 Jan 05
Posts: 376
Credit: 5744455
RAC: 0

> Boinc reguires about 256M

> Boinc reguires about 256M to run but you have to

I wanted to clarify something here, BOINC only requires about 12mb of memory. It is the science apps that require alot of memory.

BOINC WIKI

BOINCing since 2002/12/8

Flashdrive
Flashdrive
Joined: 24 Feb 05
Posts: 2
Credit: 10096
RAC: 0

Mainly what I was getting at

Mainly what I was getting at was the order that the cpu stores and retrieves data is the L1,L2,L3 (If installed),and Virtual memory on HD.Retreival from L1 being the fastest and virtual the slowest.So the more memory at the fastest retrieval the better.

Comment viewing options

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