It would be interesting to see how a PowerBook G4/667 with AltiVec off compares to an iBook 700.
It's nice to list numbers w/ AltiVec off, because that really shows you have the processor in a G4 compares to that of a G3, in a non-AltiVec optimized application. Comparing G3s to G4s in a test that is designed specifically to squeeze the most out of AltiVec is not really fair.
BTW, I believe the slowest Mac ever was the first generation 233 MHz Wallstreet, without the L2 cache. The second revision (like mine) had a 512k L2 cache, but I heard that the original cacheless 233 MHz Wallstreets weren't much faster than the 240 MHz 3400.
I know of three Lockheed Martin system engineers that are trying to get OS X to run on a VME PowerPC system that is rated at 70 GFLOPS. <img src="graemlins/surprised.gif" border="0" alt="[Surprised]" /> :eek:
<strong>I know of three Lockheed Martin system engineers that are trying to get OS X to run on a VME PowerPC system that is rated at 70 GFLOPS. <img src="graemlins/surprised.gif" border="0" alt="[Surprised]" /> :eek:
Single-port RACEway Interlink products provide 267 MB/s of communication bandwidth per slot, while dual-port RACEway Interlink products offer 533 MB/s per slot.
This seven-board system contains 22 PowerPC 7410 CPUs. Running at 400 MHz, each is supported by 2MB of L2 cache (266 MHz) and 256 MB of local memory. The I/O supports FPDP (copper) at 160 MB/s. The total compute power is over 70 GFLOPS, and the bisection bandwidth is a huge 1.6GBytes/s, with a power budget of just over 250 watts.</strong><hr></blockquote>
maybe the conformal heat dissipator cringes at PPC hotter than 7410@400
but they are running quads, if not happily yet
imagine ganging 22 SpeedHole DP1.25 with 2Mb L3/ea and 2Gb DDR?
better yet, in less than 6 months,
imagine 22 SpeedHole G5s with 4 Mb L3/ea and 4Gb DDR!
< fuzzy calculation of such hypothetically maxxed h/w... >
uh, maybe 200 GFlops from 22 DP1.25G4, 400GF from 22 G5
anybody want to correct this. tres fuzzy.
and lo, the wagering and whining commences anew until/if we could/should get such gear for silly prices
Usually in a cpu design, speed increases drops off after a while (just look at a PIII chip). The performance/MHz starts out pretty good, and then it gets wors as the MHz increases.
After reading this thread, it doesn't seem like the G4 does that at all! In some cases the performance/MHz even gets better! (especially on DP machines)!
I just found this very interesting.
Now. how woould a G4 compare to a G3 in say, mathematical non-vector equations ?
Anybody know of any good benchmark test to show that?
Alright, I managed to eek out another .4 megaflops to 171.8 in OS 9.2.2 by upping the maxcount a notch. Also ran it in OS X and got 172.
Settings:
In OS 9 and Mac OS X:
Color Speed: 0.1
Max Count: 1048576
Gallery: Zoomed Out.
Window: Fit window to main screen. No zooming afterwards.
In OS 9:
Screen @ 640x480, 256 greys, Finder off via AppleScript, optimized extension set, antialiasing off.
In OS X:
Screen @ 640x480, 256 colors, Finder off via yet another AppleScript, web sharing off, AltiVec Fractal Carbon reniced to -20, dock hidden, modem port disabled so it wouldn't decide to connect to the internet.
Took a freaking hour (3603.6 seconds and 3598.2 seconds, Mac OS 9 and Mac OS X respectively) to do each.
Just to say it again, 233MHz Rev B. iMac, 384MB RAM, OS 9.2.2 & 10.1.5.
Comments
Results: Average of just over 1400 MFlops
Not bad for a 4 year old Powerbook when new TiBooks are getting numbers around 1600 MFlops!!
seems low to me...
Maybe I just wish it was faster
It's nice to list numbers w/ AltiVec off, because that really shows you have the processor in a G4 compares to that of a G3, in a non-AltiVec optimized application. Comparing G3s to G4s in a test that is designed specifically to squeeze the most out of AltiVec is not really fair.
BTW, I believe the slowest Mac ever was the first generation 233 MHz Wallstreet, without the L2 cache. The second revision (like mine) had a 512k L2 cache, but I heard that the original cacheless 233 MHz Wallstreets weren't much faster than the 240 MHz 3400.
<a href="http://www.mc.com/literature/literature_files/conduction-cooled-vme-ppc-sys-ds.pdf" target="_blank">http://www.mc.com/literature/literature_files/conduction-cooled-vme-ppc-sys-ds.pdf</a>
Turning Altivec OFF (again with running a bunch of apps) its now abour 282 megaflops.
<strong>iMac/800/Superdrive/768 MB RAM
2480 megaflops.</strong><hr></blockquote>
Hmm interesting
iMac/800/Superdrive/512 MB RAM OS9.22
2865.9 (4.7 secs)
[ 08-21-2002: Message edited by: gsxrboy ]</p>
Can load the app into my PB 190 OS 7, but it won't run.... Pity, was looking forward to seeing the results.
2768.1 megaflops/4.8 seconds
will compare with running 10.2 this weekend (i hope).....g
<strong>I know of three Lockheed Martin system engineers that are trying to get OS X to run on a VME PowerPC system that is rated at 70 GFLOPS. <img src="graemlins/surprised.gif" border="0" alt="[Surprised]" /> :eek:
<a href="http://www.mc.com/literature/literature_files/conduction-cooled-vme-ppc-sys-ds.pdf" target="_blank">http://www.mc.com/literature/literature_files/conduction-cooled-vme-ppc-sys-ds.pdf</a></strong><hr></blockquote>
from that pdf...
[quote]<strong>
Single-port RACEway Interlink products provide 267 MB/s of communication bandwidth per slot, while dual-port RACEway Interlink products offer 533 MB/s per slot.
This seven-board system contains 22 PowerPC 7410 CPUs. Running at 400 MHz, each is supported by 2MB of L2 cache (266 MHz) and 256 MB of local memory. The I/O supports FPDP (copper) at 160 MB/s. The total compute power is over 70 GFLOPS, and the bisection bandwidth is a huge 1.6GBytes/s, with a power budget of just over 250 watts.</strong><hr></blockquote>
maybe the conformal heat dissipator cringes at PPC hotter than 7410@400
but they are running quads, if not happily yet
imagine ganging 22 SpeedHole DP1.25 with 2Mb L3/ea and 2Gb DDR?
better yet, in less than 6 months,
imagine 22 SpeedHole G5s with 4 Mb L3/ea and 4Gb DDR!
< fuzzy calculation of such hypothetically maxxed h/w... >
uh, maybe 200 GFlops from 22 DP1.25G4, 400GF from 22 G5
anybody want to correct this. tres fuzzy.
and lo, the wagering and whining commences anew until/if we could/should get such gear for silly prices
[ 08-21-2002: Message edited by: curiousuburb ]</p>
Usually in a cpu design, speed increases drops off after a while (just look at a PIII chip). The performance/MHz starts out pretty good, and then it gets wors as the MHz increases.
After reading this thread, it doesn't seem like the G4 does that at all! In some cases the performance/MHz even gets better! (especially on DP machines)!
I just found this very interesting.
Now. how woould a G4 compare to a G3 in say, mathematical non-vector equations ?
Anybody know of any good benchmark test to show that?
BoeManE
Settings:
In OS 9 and Mac OS X:
Color Speed: 0.1
Max Count: 1048576
Gallery: Zoomed Out.
Window: Fit window to main screen. No zooming afterwards.
In OS 9:
Screen @ 640x480, 256 greys, Finder off via AppleScript, optimized extension set, antialiasing off.
In OS X:
Screen @ 640x480, 256 colors, Finder off via yet another AppleScript, web sharing off, AltiVec Fractal Carbon reniced to -20, dock hidden, modem port disabled so it wouldn't decide to connect to the internet.
Took a freaking hour (3603.6 seconds and 3598.2 seconds, Mac OS 9 and Mac OS X respectively) to do each.
Just to say it again, 233MHz Rev B. iMac, 384MB RAM, OS 9.2.2 & 10.1.5.
Pretty please?
(wants to see even bigger numbers.)
<strong>Any PowerBooks out there?</strong><hr></blockquote>
My TiBook 800 is getting about 2.6 GFLOPS on this fractal test.
2770 megaflops
PowerBook 667, 768MB, 10.1.5 - 2.2G
iMac 17", 768MB, 10.2 - 2.77G
204 mflops
pim