**The LINPACK Thread**

[Burn]

Re: **The Official TF LINPACK Thread**

Ya Slay is right. We should have a set problem size and not require everyone to just set their memory use to "all."

 

[yzmxer608]

Re: **The Official TF LINPACK Thread**

I didn't notice that, what I used was what the default was.

So since everyone has different amounts of available memory, how much does that affect the scores?

 

[Slaymate]

Re: **The Official TF LINPACK Thread**

A different Problem size and Memory size effect how long the test takes. So if your testing with values below everyone else then your not going to run as hot which means you can run at a higher speed. LinX is a good Stress Test but a Poor Benchmark.

 

[Puddle Jumper]

Re: **The Official TF LINPACK Thread**

Ya Slay is right. We should have a set problem size and not require everyone to just set their memory use to "all."

If we did that we would have to go with a pretty small problem size to ensure that everyone can run it.

Edit:If we set a low default problem size I think we should increase the number of passes to keep the test stressful.

 

[yzmxer608]

Re: **The Official TF LINPACK Thread**

Hmmm, so should a standard be used for problem and memory size or is this just not destined to be a good benchmark?

Edit: Puddle posted at the same time.

 

[Hefemeister]

Re: **The Official TF LINPACK Thread**

I think if we want a good benchmark for stability we should start making superpi 32m runs.

 

[septic]

Re: **The Official TF LINPACK Thread**

I didn't have much difference in-between the 10 tests I ran with two different clocks, when changing the Problem size myself. But as slaymate, I'm alittle curious with the higher end machines. What I'm really wondering is if zmatt's sandbagging and going to come in later for the kill. .

 

[Slaymate]

Re: **The Official TF LINPACK Thread**

I think if we want a good benchmark for stability we should start making superpi 32m runs.

^I agree^

 

[Puddle Jumper]

Re: **The Official TF LINPACK Thread**

I think if we want a good benchmark for stability we should start making superpi 32m runs.

Superpi isn't multithreaded which IMO makes it much less interesting and relevant. Also in the past some people have complained that it is biased against AMD which I suppose is the reason we have two separate lists for it.

 

[Slaymate]

Re: **The Official TF LINPACK Thread**

Also in the past some people have complained that it is biased against AMD which I suppose is the reason we have two separate lists for it.

Well I think you'll find people saying the same thing about LinX once we see some AMD numbers.