[phantomac]

Hi there!

What should the benchmarks for a Dual-2.7 GHz G5 be? My benchmarks improved a lot after installing the alpha 5.2 worker, but they're still only half of what the top PowerMac7,3 entry on the seti@home Website reports.

My machine:

Measured floating point speed 4969.57 million ops/sec
Measured integer speed 18197.55 million ops/sec


Another thing: The top PowerMac7,3 in the list says it's got 2,5 GB RAM, my machine only reports 2 GB RAM, though it has 4,5 GB RAM installed.

[phantomac]

Ok, found it. ;-)
Installing the beta client helped resolve the problems mentioned above.

[TiloProbst]

those Top Macs are Quad CPU, so they should score way higher then your Dual machine

[phantomac]

Originally Posted by TiloProbst

those Top Macs are Quad CPU, so they should score way higher then your Dual machine

I was talking about the PowerMac7,3, not about the PowerMac11,2, which is the 4 CPU machine.

A dual 2.7 GHz G5 should still be the second fastet PowerMac out there, only surpassed by the quad machine.

[alexkan]

Originally Posted by phantomac

I was talking about the PowerMac7,3, not about the PowerMac11,2, which is the 4 CPU machine.

A dual 2.7 GHz G5 should still be the second fastet PowerMac out there, only surpassed by the quad machine.

Not that this is particularly relevant, but that might not actually be the case when it comes to SETI processing. You'd be surprised by how much of a difference the L2 cache in the dual-core Macs makes. I would expect the DC 2.3s to be the second fastest machines. For that matter, the DC 2.3s might well turn in faster individual WU times than the quads, depending on how saturated all the buses are when each CPU is running a SETI worker.

[Todd Madson]

I've still yet to see results from a DC2.3. Anyone have one to try it?

[E.T from tellus]

I was quite dissapointed of dual 2.7ghz performance.
http://setiathome.berkeley.edu/show_...hostid=1789945
My DC 2.3ghz's outperforms that easily.
http://setiathome.berkeley.edu/show_...hostid=1502330
http://setiathome.berkeley.edu/show_...hostid=1946839
E.T

[phantomac]

Yeah, the 2.7 seems to be suffering from a clash between the RAMs and the FSBs speeds.

[Todd Madson]

Interesting...

Your 2.7 gets about what my 2.5 does. Actually, a little worse in some cases.

It should do a bit better but....what kind of ram?

My 2.5 has PC3200-30330 ram. I could get faster but....not sure it's worth it.

And, you're right - that 2.3DC is an amazing performer there.

[alexkan]

Don't expect any of these machines to scale linearly in performance with clock speed. The determining factor here is definitely L2 cache and memory bandwidth, although the clock speed might affect this a little bit. This is the reason why G4s don't seem to be able to get work unit times below the 5500-second mark--combined with the fact that newer G4s have smaller L2s than would be ideal for SETI, they just can't read data from RAM fast enough. Non-DC G5s also suffer from having smaller-than-ideal L2 caches for SETI, hence the lackluster performance compared to the newer machines.

I have to say, working on these SETI clients has definitely skewed how I look at CPU performance, since the benchmarks I see most often are ones pertaining to SETI work units and FFTs, which DC G5s happen to be particularly good at.

[Todd Madson]

What I notice most of all is that there seems to be large variation
even between machines of the same type.

I ponder whether some G5 chips work better than others for
reasons of variation in manufacture.

Obviously, speed of hard disk (7200 rpm versus 10k rpm)
and speed of ram (CAS speeds) will factor in cumulatively
over time (perhaps running seti out of a ramdisk might be
something to try for fun) but my machine seems to run
into blocks below 2000 seconds more than other 2.5s.

One things for sure, if you had a couple of quads and a
couple of DC 2.3s on your team you would be hard to beat.