I've noticed on several crunchers statistics that AMD systems do unbelieveable well, and INTEL systems are way under par.

Why the difference? What specifically about each processor produces different results? I'm only curious because I feel like I'm wasting my time using an INTEL system to run this program when a similarly equivelant AMD system will produce 10 times the results.

Baffled...

---

Visit us at Christianboards.org

AMD processors have better floating point units, so are well suited to number crunching applications. Intel chips cater better for multimedia type applications. Where the difference can be lower is with small applications like Seti, this is because they can get better use from the Intel chips larger cache sizes.

---

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

adrianxw got the point right.
As this is a know feature of AMD-Intel differences, I put some offort to complete a table that also compares how AMD and Intel is doing among BOINC projects.
Note that it is not to be a benchmark results, nor 100% accurate (which is not possible due to nature of the projects; e.g. different WU size/time-to-complete). It was intended to offer an idea what project should users choose with their hardware so that it is 1) capable of doing the job (memory usage, network traffic, HD space) and 2) make their CPUs working effeciently. For sure, there are other aspects why people choose project: their interests, beliefs, support from dev team (Rosetta is doing great so far!) etc.

---

Thanks for the explanations. This helps on understanding.

---

Visit us at Christianboards.org

Honza, great table - I feel compelled to point out that the Rosetta WU/day limit is 200, not 100, though :-)

---

Team CFVault.com
http://www.cfvault.com

You can quickly scan the following link (link), it's stats for CPU type for Rosetta sorted by RAC. You'll see that the majority of the top CPU are AMD.

PS: Interestingly enough, AMD Sempron chips (which are about $100 Canadian) hold their own. Pretty cheap option if you're buying a cruncher. :P

---

Something else which sprung to mind is in the earlier versions of BOINC, Hyper Threding machines were not claiming correctly, which of course, hits the apparent performance of Intel. I don't know if this is still an issue.

---

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

Honza, great table - I feel compelled to point out that the Rosetta WU/day limit is 200, not 100, though :-)

That's not a fixed number. I have one machine that is allowed 500, and during the problems with BOINC 4.19 generating eroors, I think one machine was down to 5 a day.

/Ib

---

Honza, great table - I feel compelled to point out that the Rosetta WU/day limit is 200, not 100, though :-)

That's not a fixed number. I have one machine that is allowed 500, and during the problems with BOINC 4.19 generating eroors, I think one machine was down to 5 a day.

/Ib

Thanks for clarifying that.

---

Team CFVault.com
http://www.cfvault.com

Is it just me? I do not see a link to Honza's table ...

---

its not a link, its the table included as image further down this thread

---

--
Florian
www.domplatz1.de

The credits range for Rosetta is much larger that 10-30.
My smallest WU gave 2,7 Credits and my largest about 37 credits and I guess that with the new WUs we might even see 40 or maybe 50.

---

Thanks for the feeback.
Yes, daily quota has a dynamic concept on BOINC - it goes down when particular host has many WU errored out which is smart: it is preventing WU pool from draining by unstable or otherwise faultly host (or even application error).

When i look at my two boxes, one has 100 WUs per CPU/day, another has 500 WUs per CPU/day.
Both are enough to prevent your machine from being idle even when running 24/7 Rosetta exclusive.

@ Foxfire. Well, only CPDN provides fixed credit per WU (CPDN model) without lottery [I believe PrimeGrid has a fixed value per WU type as well]. This makes credit award a bit difficult to range. I'm not into credit so my main aim was (and still is) to delineate project requirement in term of RAM, internet traffic a time-to-complete of each WU on typical machine (mostly what I posses).

When you want to make it more accurate, you run into troubles: WU on many project are not one like another and even is unknown before it completes [not constant time to complete, typically BURP], there may be a change in application or even some optimalized one. And there credit...benchmark giving large range of numbers involving OC of CPU and/or RAM, optimalized clients, differences among OSes (mainly Win and Linux) etc.

So, I redrawed from comparing AMD and Intel based on credit award even some credit hunters might be interested. I'm more into science anyway...

Well, it is just to 'get the idea' of what each project may be like and if you machine is capable of doing it.

---

Why are the intel machines rewarded double the numbers with same time crunching
Getting sponsored by intel or what ?

---

Why are the intel machines rewarded double the numbers with same time crunching
Getting sponsored by intel or what ?

That's quite ironic- the thread was about AMD systems getting more credit, but that was before the C2D came out. It's only Core2 CPUs that get high scores - the other Intel CPUs don't, and AMD chips with a decent amount of cache still do pretty well.

It's nothing to do with any optimisations in favour of either type of CPU.

---

I got the impression from other threads that it was the cache of the cpu chip that helped or hindered the points spread, based on what kind of a WU was being crunched. Some do better with AMD cpu's and their architecture and others don't get affected by the AMD cpu and yet other WU's really dig a hole in your credit if your a AMD user like me. I have a older 2800+, it works well for some WU's and others it makes my credit drop like a rock.

---

Well indeed i get no grip on how we are rewarded
And i must confess its seems only those new c2d and quads intels get double points for same work.
But also my own credits are totally random for some units with lots of cpu time i get half the credits while others seems to get 110% credits.

---

most of the time recently my credits are +/- .5 to 1 point from the claimed credit. some WU's way back took me for a real dive. lost about 10-15 points in the average graph. never recovered.

---

Well indeed i get no grip on how we are rewarded
And i must confess its seems only those new c2d and quads intels get double points for same work.

For the same work they get the same points. The only reasonably accurate way to tell how much work has been done is by the number of decoys produced on a given work unit.

---