Dual versus Single CPU in Mac Os 9 Real test

[Tron]

Dual 1.42 versus 1.42 Single (1.25 overclocked to 1.42) in the same machine (G4 MDD).

I have done several speed test -primary video compression- swapping two processors cards in my G4 MDD.

Firsts of all: nothing take benefits of the dual processor, excep some task in some applications designed for dual processors.

Some apps claimed to be dual aware, but I couldn't see or measure some improvement as Virtual PC 6 or Apple DVD player.

Toast Video CD -> 67% faster on Dual

Apple MPEG2 ->45% faster on Dual

3ivx -> 10% faster on Dual (1)

mp4 -> 8% faster on Dual

Cleaner 6 ->20% faster on Dual (2)

(1) In others test show 30% faster on Dual

(2) In others test show 0% faster on Dual

Qt Player 3ivx (divx/mp4 playback) 12-24% faster on Dual

Sound Jam MP -> 38% faster on Dual

Photoshop 5.5 is also a 15-20% faster on dual but I couldn't measure well.

Sorry for my Englis

 

[trag]

That's some great information. Thank you for sharing the results of your real world tests. That's the best kind of information to see and it is rare.

 

[ppuskari]

Great info!

It really highlights which apps are dual or even multi-thread aware as well as which apps/codecs are memory and/or io bound instead of cpu bound.

This is the kind of stuff that is so hard to explain to people at work. We just added 9 procs to the Z10 mainframe and only got 15% more improvement in speed. What did we do wrong?... sigh.

:scrambled:

 

[Bunsen]

Are the CPUs exactly the same? ie, the same model/series of G4, the same amount of cache at L1, L2 and L3, the same cache:cpu speed ratio?

 

[Tron]

Both are G4 revision 3.3 (7455B) but diferent production number.

Both G4 have 32kInstr/32kData L1 cache at CPU speed.

Both G4 have 256k L2 cache at CPU speed.

Both G4 have L3 cache at 1/6 CPU speed.

Dual have 2MB L3 cache per processor.

Single have 1MB L3 cahe per processor.

I couldn't see or measure any diference in speed due to the extra caché in Dual G4. Even in the monoprocessor tasks tests done in both CPU cards. Despite the video test will benefit of it, at least in theory.

Furthermore, testing the 166Mh MDX FSB in my other MDD (Single G4 overclocked to 1.667 Mhz, where I writing this) I couldn't saturate it easily. To do it I have to playback a full HD (1900 x 1080) uncompressed video!

Sorry for my English.

 

[Bunsen]

Your English is fine. Thankyou for the extra information.

Dual have 2MB L3 cache per processor.Single have 1MB L3 cahe per processor.

I would expect that to make some difference, but I have no idea how much.

 

[Quadraman]

I saw an old review of a prototype G4 upgrade from Daystar that had 2m cache vs 1m cache in the production model and the results weren't impressive. Running about 20 apps and benchmarks the 2m cache model only scored a combined 3% faster overall and showed no benefit at all in about half the tests so Daystar dumped the idea because the extra cost wasn't worth it for the minimal benefit on only a handful of apps that were able to utilize the additional cache. The difference in the cache in this test probably had little effect on the outcome. Still, the classic Mac OS is not multi processor aware, so your apps have to be if you want to see any real benefit from the second CPU. It's like running non-Altivec aware apps on a G3. You won't see any real speed difference from a G4.

 

[Cory5412]

You say you didn't notice a difference in Virtual PC. Did VPC claim to be dual-aware? If so, probably what you'd need to do then is test out running two virtual PCs on each the dual and the single. Hypothetically if VPC is dual-aware, it would run one virtual PC on processor "A" and the other on processor "B" and you could get a lot more done, or whatever you were doing with VPC.

Otherwise, pretty informative test.

It's a fairly telling indicator that either most apps are just not multi threaded by the time of OS9, or that OS9 has the ability to shove big tasks onto the second CPU, causing that marginal increase in performance, as seem in 3ivx, cleaner and mp4.

DVD player, being mainly of a single task, probably gets moer "dual benefits" when you run it on really slow duals, like 400 or 450MHz dual G4s, and probably then mainly because the app can claim a whole processor to itself while letting other things happen on the "main" processor.