PDA

View Full Version : Dual 800 Benchmark Requests/Results


MacRumors
Aug 18, 2001, 05:03 PM
With the Dual 800's trickling out, a lot of people are looking for benchmarks. I've got mine sitting here, so if there are any requests for Benchmarks, let me know -- with specific links to the requested software in question.

I will add my results to this thread, and encourage others to do the same.

arn
Aug 18, 2001, 05:04 PM
http://24.6.46.233/testbench/smpbench-mosx.tar.gz

SMPBench under OS X after a fresh boot.
Dual 800/G4. 512mb CL2 Ram. GeForce 3


Ars Technica SMP Benchmark [Ver: 1.6] Clock Precision: < 1 us

System Info:
PowerMac3,5
PowerPC 7450
2 Processors
CPU Speed: 799 mhz
Bus Speed: 133 mhz
RAM: 512 megs


1 thread:
- time elapsed -----> 28.90s
- charged time -----> 28.88s
----- score --------> 221.42%

2 threads:
---- time elapsed ------> 16.41s
---- charged time ------> 30.12s
-------- score --------> 390.08%

4 threads:
---- time elapsed ------> 15.76s
---- charged time ------> 31.09s
-------- score --------> 406.16%

6 threads:
---- time elapsed ------> 15.17s
---- charged time ------> 29.97s
-------- score --------> 421.91%

8 threads:
---- time elapsed ------> 15.41s
---- charged time ------> 30.42s
-------- score --------> 415.31%

[Edited by arn on 08-18-2001 at 06:13 PM]

arn
Aug 18, 2001, 05:12 PM
Cinebench 2000 V1.0 Performance
****************************************************
Shading (CINEMA 4D) : 6.87 CB
Shading (OpenGL) : 9.07 CB
Raytracing (Single CPU): 9.78 CB
Raytracing (Multiple CPU): 17.18 CB

OpenGL Shading is 1.32 times faster than CINEMA 4D Shading!
2 CPUs are 1.76 times faster than 1 CPU !

****************************************************

Dual 800/G4. 512mb CL2 Ram. GeForce 3. Under OS 9.2 - standard extensions.

http://www.maxon.net/pages/download/benchmarks.html

arn
Aug 18, 2001, 05:19 PM
Ars Technica CPU Benchmark [Ver: 1.6] Clock Precision: < 1 us

1: 119% [FP ] 4.42s NTSC YIQ -> PAL RGB
2: 146% [FP ] 3.28s NTSC YIQ -> PAL RGB
3: 179% [FP ] 4.12s FP Divide
4: 132% [MEM] 1.52s Fast Fourier Transform 64k
5: 231% [FP ] 0.65s Fast Fourier Transform 16k
6: 175% [INT] 3.34s LZ77 (Lempel-Ziv) Data Compression
7: 148% [INT] 1.19s LZ77 (Lempel-Ziv) Data Decompression
8: 184% [INT] 1.87s RGB to CMYK with Color Correction
9: 217% [INT] 1.44s Blowfish Block Cipher Encryption (448 bit key)
10: 194% [INT] 1.38s File Allocation Table Manipulation
11: 206% [INT] 1.48s Encode G.723-40 Audio 100k
12: 195% [INT] 1.23s Complex Bitfield Operations
13: 139% [FP ] 1.98s Spherical Harmonics Legendre Polynomial
14: 163% [INT] 3.63s Dhrystone 2.1 (500k Iterations)
15: 134% [FP ] 2.18s Whetstone 1.2 (1000k Operations)
16: 158% [MEM] 1.78s LinPack (500x500)
17: 158% [INT] 9.54s TSCP 1.71 (Chess Benchmark)
18: 165% [MEM] 6.23s Sieve of Eratosthenes (Prime Generator)
19: 201% [INT] 2.75s Towers of Hanoi


---------------------------------------------------------------------
FINAL SCORE: INT: 184 FP: 158 MEM: 152 SIMD: 0
PEAK SCORE: INT: 186 FP: 160 MEM: 153 SIMD: 0
---------------------------------------------------------------------

http://132.239.148.52/tb16.sit

Dual 800/G4. 512mb CL2 Ram. GeForce 3. Under OS 9.2 - standard extensions.

arn
Aug 18, 2001, 05:27 PM
94.9 FPS

Dual 800/G4. 512mb CL2 RAM. GeForce 3

Amazing Quality, millions of colors, 1024x768

http://www.altorsys.com/HTMLAltor/ProjectBSoftware.html

pbg4user
Aug 18, 2001, 09:02 PM
how much time it takes the Mac os x version of seti@home to complete one unit
note:the mac os x version of seti@home is MP enabled

the download site for the software is

ftp://ftp.cdrom.com/pub/setiathome/setiathome_macOSX_3_05.hqx

arn
Aug 18, 2001, 11:06 PM
hmm....

Does it show you that stat anywhere? or do you just have to calculate it yourself.

arn

SolidBrass
Aug 19, 2001, 12:28 AM
How about RC5-64 running native in OS X?

(Must be run at the command line, "./dnetc')

http://http.distributed.net/pub/dcti/current-client/dnetc-macosx-ppc.tar.gz

My Dual450 does a bit over 9 milion keys a second.

blakespot
Aug 19, 2001, 07:19 AM
My God. Can you imagine a Beowulf cluster of these things?




blakespot

anonymous
Aug 19, 2001, 09:03 AM
As I understand, apple ships PC133/3-3-3 with all their Macs. You however have CL2 (2-?-? i think? or ?-2-? maybe). Is this aftermarket, did you remove the existing RAM? Or did Apple just ship your box with CL2?

arn
Aug 19, 2001, 09:50 AM
It came with CL3 ram. I bought 2 256mb sticks from OWC that are CL2.

arn

pbg4user
Aug 19, 2001, 11:20 AM
Originally posted by arn
hmm....

Does it show you that stat anywhere? or do you just have to calculate it yourself.

arn

i think you have to calculate it yourself but i don't know.
I'd like to see how fast a dual 800 is.

arn
Aug 19, 2001, 07:26 PM
Distributed Net's most recent public release under OS 9:

13.85 MKeys/sec

Under OS 10, I get much slower results... like 4-5MKeys/sec.

As for SETI@Home

I just tried the OS 10 version and I calcuated it out to roughly 6-something hours per unit.

arn

arn
Aug 19, 2001, 09:33 PM
From the internal SuperDrive:

4-5x encoding of American Music by Violent Femmes

From internal HD:

15-16x encoding of the same

From External Firewire Yamaha 16x10x40 CD-RW

15-16x encoding of the same

arn

A
Aug 19, 2001, 10:39 PM
http://bbs.xlr8yourmac.com/ubb/Forum3/HTML/001131.html

Thanks.

blakespot
Aug 20, 2001, 08:08 AM
Arn, is that iTunes under OS 9 or OS X?

It's eerie--you are seemingly the first person on the planet to have received a dual 800.



blakespot

T'hain Esh Kelch
Aug 20, 2001, 08:13 AM
Quake III or UT benchmarks ?

Yeah...I'm a gamer.. ;)

Classic
Aug 20, 2001, 09:36 AM
Arn,

Did you dual ship with OS 9.2 or 9.2.1?

According to think secret:

http://www.thinksecret.com/features/limelightchannel.html

There are reportedly Quicksilvers shipping with OS 9.2.1

johnnyvn
Aug 20, 2001, 10:14 AM
Here goes:

A Dual-800 with 768megs of RAM. This unit does not have a Superdrive. CD-RW only. Apple "TwinView" Graphics card. Aside from those build-to-order requests, it is a stock unit.


Cinebench 2000 test:

Shading (4D): 7.08 CB
Shading (OpenGl): 9.06 CB
Raytrace (Single): 9.93 CB
Raytrace (Multi): 17.22 CB


Throughput V1.5 test:

CPU 213.3
FPU 213.6
Altivec 220.5
Copypits 252.7


Altivec Fractal test:

(Stock settings except Maximum Count set to 65536)
6.2 Gigaflops


Processor query:

7450
Unknown Rev.
PVR 0x80000201


That's all I've got so far. I love this wonderful beast!

John

arn
Aug 20, 2001, 11:20 AM
Unfortunately, I don't own Quake 3... I downloaded the demo and the newest client... but didn't get it working after a quick try. I may pick up Quake 3... but someone else might beat me to the benches... or if anyone has detailed instructs on how to set it up with the demo/client.

The Quicksilver dual800 did come with OS 9.2.1

arn

passwordisdigital
Aug 20, 2001, 01:19 PM
SETI@Home is not multi-processor (or Altivec) aware. If you want to test it's true SETI speed and I'd love to see the results if you do, you would need to run two copies of the software simultaneously. Supposedly the UNIX version running from the command line in OS X is the faster version available on the Mac and the OS would basically give each program a seperate processor to work with.

Just for your info, my 733 PIII at work takes around 9-10 hours to complete a work unit. If the 800 MHZ finished in around 6 hours, that's a pretty big difference in speed, considering it's only a 67 MHZ jump. Running two copies could finish a work unit in an average of 3 hours....

Classic
Aug 20, 2001, 02:14 PM
Contrary to popular belief, and to what you are lead to believe by the benchmarks posted here, it appears that apple is now "officially" shipping the dual 800s.....

http://www.apple.com/pr/library/2001/aug/20pmg4.html

Whatever... I guess the postings of benchmarks must be fabrications...

I wonder why they didn't come out last week and say they were shipping them?

As blakespot said to Arn, "It's eerie--you are seemingly the first person on the planet to have received a dual 800."

Jake R
Aug 20, 2001, 03:21 PM
My Sawtooth G4/450 completes a SETI unit in about 5.25 hours. What gives?

The GUI version of SETI is MP enabled, in that the GUI code is a separate thread, allowing it to be dispatched to a second processor. If you want to run command line only, you'll have to run two at a time in order to push both processors.

-Jake

arn
Aug 20, 2001, 03:59 PM
The SETI@Home bench was on OS X...

With RC5 the OS X version seemed much slower under OS X than the OS 9 version...

I'm not sure which version of SETI you're using... but I'll give the OS 9 version a try when I get home (tommorrow evening).

arn

sauria
Aug 20, 2001, 08:56 PM
Gauge Pro 1.1 Memory Bandwidth?

How's the speaker hiss?

sauria
Aug 20, 2001, 09:03 PM
Originally posted by Jake R
My Sawtooth G4/450 completes a SETI unit in about 5.25 hours. What gives?

-Jake

Odd, that must be a verison prior to 3.03? Should be around 13 hours max.

ddtlm
Aug 21, 2001, 12:35 AM
No way a 450 can do the current seti packets that fast... back in the days of 2.x yes, but not 3.x. Back then there was a huge benefit to large processor caches, which has been elliminated. If your G4 really can do 6 hours with the new client, its doing pretty well (but lagging the top PC's). My 700mhz 1mb xeon (linux) is somewhere around 8.5 hours per packet, per CPU. Higher clock speeds do make a difference too, so I wouldn't go bragging about the G4's times until you get some fast Athlons benched.

Also, its inexplicable that RC5 shows such a speed difference between OS's. Some I can see, but that much... in general, I would expect ALL raw computing to be a little slower in X than 9.

ddtlm
Aug 21, 2001, 12:49 AM
From here: http://www.teamlambchop.com/bench/303results.htm, we see that with a "standard packet" top PC's go as low as 3:22, for a 2gz P4. *One* 2gz P4, note not a dual Xeon or anything... so bringing in two G4's would be unfair and meaningless. So anyway, unless you get some really good numbers here, seti is one place that Macs are not gona lead.

If 6 hours hold true, that'll put you somewhere around 1gz in P3 and Athlon speed (going by the above URL).

But anyway, seti is just one benchmark.

Megaquad
Aug 21, 2001, 04:49 AM
hey,can you try running a 640*480 fullscreen divx movie?

PBG4user
Aug 22, 2001, 09:22 AM
Originally posted by passwordisdigital
SETI@Home is not multi-processor (or Altivec) aware. If you want to test it's true SETI speed and I'd love to see the results if you do, you would need to run two copies of the software simultaneously. Supposedly the UNIX version running from the command line in OS X is the faster version available on the Mac and the OS would basically give each program a seperate processor to work with.

Just for your info, my 733 PIII at work takes around 9-10 hours to complete a work unit. If the 800 MHZ finished in around 6 hours, that's a pretty big difference in speed, considering it's only a 67 MHZ jump. Running two copies could finish a work unit in an average of 3 hours....

i do have to say that u r wrong. You have to read the OSX_Installation_FAQ it says it is Multiple Processor enabled. Oh Yeah, and one more thing i think that the actual time it takes it to complete a work unit is 3 hours
because the total time shown is for both processors.

Jake R
Aug 22, 2001, 09:24 AM
Originally posted by sauria
Originally posted by Jake R
My Sawtooth G4/450 completes a SETI unit in about 5.25 hours. What gives?

-Jake

Odd, that must be a verison prior to 3.03? Should be around 13 hours max.

Why, did they change it? I haven't run it in a long time. That's an OS9 speed from about six months ago.

-Jake

Jake R
Aug 22, 2001, 09:28 AM
Originally posted by ddtlm
From here: http://www.teamlambchop.com/bench/303results.htm, we see that with a "standard packet" top PC's go as low as 3:22, for a 2gz P4. *One* 2gz P4, note not a dual Xeon or anything... so bringing in two G4's would be unfair and meaningless.

That fact that a DP G4 is readily available doesn't make in unfair and meaningless. You can get a DP PC system to use if you want to compete. They end up costing about the same...

I can't wait until the G5 comes out... should be six months or so. It is reportedly 2-3 times faster than the Itanium, per clock, and will hit higher clock rates than Itanium will anyway. Reportedly.

-Jake

Makosuke
Aug 22, 2001, 08:08 PM
There seems to be a surprising amount of confusion about how SETI@home comes up with its numbers, and although I'm not an expert at all, here's what I noticed during some experimentation a few weeks ago (might be wrong now, of course):

The OS9 version of SETI gives you a number of hours taken to complete a work unit. This is true on single or dual processor machines. It does (apparently) take advantage of DP machines, however, and seemed to be able to finish a unit in around 5-6 hours on a DP 533... working out to around 10-12 hours on a single G4 533.

On X, it also uses both processors (without launching two instances of the app, at least for the GUI version), but the timer it shows is for processor time, not actual work time. That is, on DP machines the timer goes up by two minutes for every minute it's running, because each processor contributes one minute of work. At least for me, it seemed to be taking around the same amount of time (though a bit slower, probably due to overhead)--12-14 hours on the counter, which only took 6 or 7 hours of real time to complete. (And it was definitely using both processors--the CPU meter gague was maxed out on both.)

Basically, the Classic or X versions are about the same, but they count the time on the screen differently, making X look slower, or 9 faster.

As far as P4/Athlon comparisons, who knows. SETI@home isn't AltiVec optimized, though--it'd be interesting to see how fast an optimized version would run.

And by the way--the reason the work units take longer now is that they now include a lot more calculations, since (on average) computers are much faster than when the project started. More calculations equals more interesting results from data analysis.

Vsx1
Aug 24, 2001, 08:27 PM
As a comparison to an older machine my iMac 350 MHz w/ 576 MB RAM under 9.1/9.2.1 takes about 20 hours per work unit. Also using SETI@home 3.03.

vsx1@mac.com

Gulliver64
Aug 31, 2001, 06:20 AM
For comparision:
QS 867/384MB RAM, 60GB, NVidia 2MX, DVD-ROM/CD-RW

Cinebench:
Shading: 7,45 CB
Shading (OpenGL): 9,77 CB
Raytracing: 10,70 CB
GL-Factor: 1,31x

Throughput 1.5:
CPU: 213,4
FPU: 213,8
AltiVec: 221,0
CopyBits: 253,0

spikey
Aug 31, 2001, 10:54 AM
very true jake, i have a feeling it could be a top chip too.
Infact from what i know the dp800 G4 delivers nearly as much gigaflops as the itanium does.

Megaquad
Aug 31, 2001, 12:08 PM
how much gigaflops delivers itanium?

booby
Sep 8, 2001, 04:09 PM
Originally posted by arn
hmm....

Does it show you that stat anywhere? or do you just have to calculate it yourself.

arn

When you have completed a package in SETI it tells you the length of time it took..on a G3 400mhz it took me(if my memory serves me) approximately 13 hours to complete one package
SETI is a good Benchmark tester..i am curious too.:c)

Jake R
Sep 9, 2001, 02:26 AM
Originally posted by spikey
very true jake, i have a feeling it could be a top chip too.
Infact from what i know the dp800 G4 delivers nearly as much gigaflops as the itanium does.


Uh, the DP800 delivers a potential 11.8 gigaflops. I think the Itanium delivers like six.

-Jake

TigerKR
Dec 14, 2001, 03:58 AM
Everyone with a 7450 G4 (or 2) should download the 2.8016.470 client. There is about a 25% improvement in keys/sec.

Classic:
http://http.distributed.net/pub/dcti/current-client/dnetc-macos-ppc.sit

Mac OS X:
http://http.distributed.net/pub/dcti/current-client/dnetc-macosx-ppc.tar.gz

sauria
Dec 14, 2001, 09:07 AM
Originally posted by TigerKR
Everyone with a 7450 G4 (or 2) should download the 2.8016.470 client. There is about a 25% improvement in keys/sec.

Classic:
http://http.distributed.net/pub/dcti/current-client/dnetc-macos-ppc.sit

Mac OS X:
http://http.distributed.net/pub/dcti/current-client/dnetc-macosx-ppc.tar.gz

25% better in OS9? Do you have the data?

Thanks