17 replies to this topic

[Careless]

Hello everyone.

A while back, a fellow member had made a thread about Altivec Fractal Carbon, which was a PPC only binary, and was able to test the number crunching speed of a processor, with a single result at the end. that was version 1.3

on that version, i scored a best of 5000 MegaFlops, while utilizing a "maximum count" of 65536 in the program settings with nothing else running in the background.

so i reinstalled, using JaS DVD 10.4.6, and using myzars SecUpd003 + AMD Enabler. and then i went to download the program again, and realized it is now called Altivec Power Fractal, and it is also at version 1.4, which is the new Universal Intel/PPC Binary.

so i thought, hmmm maybe i'll see like a 20% increase in performance.
but what i got was totally unexpected.

8673.2 MegaFlops.

now i will link you, and you will all test this program =)
and this will become a sticky, as it is more reliable than xbench in terms of computing power, i find. (hint hint)

For the sake of this thread, please use the following parameters:


Color Speed: 10
Maximum Count: 262144
Zoom Factor: 2


this will ensure that everyone has scores based on the same test.
upon opening the program, you will see that it starts to test your computer already. just let it do its thing, and then change your Maximum Count to the aboven number, as every other parameter is default in this test.

here is a link for you testers.

Altivec Power Fractal

good luck!

I'll start with my result at the designated settings....

Elapsed Time:79.5 Seconds
Achieving 8683.0 MegaFlops


your turn.  

Attached Files

•  Careless_Carbon.png   385.24K   89 downloads

[bofors]

Careless, on May 31 2006, 03:15 PM, said:

so i thought, hmmm maybe i'll see like a 20% increase in performance.
but what i got was totally unexpected.

Trying to emulate Altivec with current SSE3 implementations (via Rosetta) is backward.  AltiVec is way beyond present SSE3 and it will take Merom, Conroe or Woodcrest to even begin to compete.  Your results do not surprise me.

[Careless]

bofors, on May 31 2006, 03:29 PM, said:

Trying to emulate Altivec with current SSE3 implementations (via Rosetta) is backward.  AltiVec is way beyond present SSE3 and it will take Merom, Conroe or Woodcrest to even begin to compete.  Your results do not surprise me.

hmmm. can you explain this more?!
i used to get like a 5000 score when i was running under rosetta, but now im running native, so it's about 35-40% higher.

also, did you end up getting a dual-socket motherboard  you asked me about?!
my inbox got erased the night before i took a trip out of the province, so i totally forgot about it


i was to inform you about the New MCP5 that nvidia is releasing with socket am2.
and i was going to tell you to wait until the end of the summer to see if it's worth getting instead of a conroe

sorry bofies!

[phi]

So ok,

on my laptop (MBP 2,16 but with Single Channel DDR2!) I get

115,9 sec  7141.5 MFlops

and on the Mactel (2x3.82GHz PentiumD930)

95.2 sec    7245.9 MFlops

On the PentiumD i remember to have a score about 4700 under rosetta and the old version.

I don't think that this test is more comparable than xbench or so, but i don't want to say that xbench is in any kind good. Its fast and easy, and totally inacurate. But the fractal test is only measuring the FPU performance which is not the only thing a good processor must do. In Final Cut for example the PentiumD is about 30% faster than the CoreDuo (86 to 112 seconds).

[New001]

So much to say!


Now, I just found this out, so lol. My 3GHz Pentium 4 takes 128.0 seconds and gets ~5300 MegaFlops

Using the no no SSE code option my 3GHz Pentium 4 gets ~1600 MegaFlops.


--------------------------

Sadly, my brand new single core iMac G5 2.1GHz had to go in for repair (), the one day I had it I happened to test this Application on it.

~7000 MegaFlops


'Nuff said. . .

[macprodan]

Perhaps the DDR2 helps alot in this Kind of app. Plus how Optimised is the SSE code compared to the Altivec. I know Altivec is God to PPC Mac Lovers but the A64/P4 and New Core Solo/Duo is a much Better Processor than the PPC, the intel Core architecture has some serious Power under the Hood its like a cross between PIII/P4 and the PM with More of Everything that Made the PM so Good..

[Order2Chaos]

My lowly sempron 2800+ gets 2799.4 MF with an elapsed time of 246.5 sec ,  max count= 262144. No one has posted any celeron results so I don't really have any thing to compare with at the moment.

[macprodan]

Settings As Directed. Dual Core X2 @ 2 x 2.3 Ghz / DDR 460

87.4 Seconds
7893.6 MegaFlops

Settings As Directed. Single Core X2 @ 2 x 2.3 Ghz / DDR 460

172.8 Seconds
3994.4 MegaFlops

[bofors]

Careless, on May 31 2006, 06:53 PM, said:

hmmm. can you explain this more?!
i used to get like a 5000 score when i was running under rosetta, but now im running native, so it's about 35-40% higher.

Here is the basic deal with SSE and Altivec.  SSE only uses 64-bit registers as of now (they go to 128 with Core Duo 2), while Altivec uses 128-bits.  This means that Altivec is twice as fast from the get-go.  It takes SSE at least two cycles to complete any vector operation, whereas Altivec can do it one.  The Altivec instruction set also contains operations that SEE does not have.  The result is that emulating Altivec with SSE results in bad performance.  Furthermore, I do not believe the Rosetta that "officially" emulates Altivec and in anycase Apple told developers that it did not originally.  So, Altivec emulation in Rosetta probably not optimized either.

I will post some results from my G5 (which should help make my point).

Quote

also, did you end up getting a dual-socket motherboard  you asked me about?!
my inbox got erased the night before i took a trip out of the province, so i totally forgot about it
i was to inform you about the New MCP5 that nvidia is releasing with socket am2.
and i was going to tell you to wait until the end of the summer to see if it's worth getting instead of a conroe

No, I have not needed anymore computing horsepower yet.  But I am interested in AMD's "4 x 4" platform, as it should be much cheaper than a dual Woodcrest and more overclockable.

[New001]

Hehehe. With your settings, Careless, I get 8023 MegaFlops on my single core 2.1GHz iMac G5, funny how with default I get around 7200 MegaFlops.

[Khan_Man]

Hi,

Using PowerFractal 1.4.1

On my Conroe E6600 (2.4gig), D945GCL, 2 gig ram, 10.4.6  I got:
17186.5 MegaFlops


MBP C2D 2.33, 2 gig ram, 10.4.7:
16485.3 MegaFlops

MBP CD 2.0, 1.5 gig ram, 10.4.6
7104.1 MegaFlops

P4 3.0gig, D915GUX, 2 gig ram, 10.4.6:
5436.8 MegaFlops

MacPro 4 core, 2.66 gig, 2 gig ram, 10.4.8:
37584.4 MegaFlops

Ginghus

[Zulu.Walker]

Interesting app, I get nearly half of what a MacPro Quad-Core gets, which is a good indication since I only have a dual-core C2Duo. Another benchmark test to alleviate my "hack" performance concerns.

Specs in sig!
60.8seconds
21274.9 MegaFlops

I'm glad everything looks like it's running at a comparatively competitive level compared to official Apple Hardware.

Quote

This code has achieved:

1.5 GigaFlops on a G4/450MHz,
13.5 GigaFlops on a DPG5/2GHz,
22 GigaFlops on 16 G4/400's,
217 GigaFlops on 33 XServe DPG4/1GHz's,
233 GigaFlops on 56 DPG4/533's + 20 DPG4/450's
and
1.21 TeraFlops on 128 Xserve DPG5/2GHz's.

I compute like 16 G4/400's and much faster than the DPG5/2GHz. Pretty nice IMHO Loving it. I even had OmniWeb and Azureus 3 running in the BG, could've reached at least 22GFlops easy on a clean reboot

[New001]

http://image.mkzip.c...96Picture 1.png

15,083 MegaFlops using both cores. Not bad, used to be around 8,000. Using the scalar mode results in roughly 1,400 MegaFlops per core.

[Dainas]

Heres another ancient bump, but since not too much goes on in this subforum no shame in that.

2001 Digital Audio Powermac with sonnet Dual 1.8ghz; 16522.6 mflops in 47secs using requested settings.



Being as since I keep blowing money on my PC this old powermac will be busy for a bit longer. Still runs well enough that I'm not tempted to join the osx86 world just yet.

[u1m2]

Here is mine. 27.1 seconds and 25409.8 Mflops.

[Paranoid Marvin]

On my Powerbook G4 - 1.5ghz:

206.7 seconds 3755.4 mflops

[ced1610]

 Image_2.png   1.07MB   6 downloads

My p5k  q6600 hackintosh :
2 gb ddr2 800 Mhz

very fast !!!!!    


44776,2 Mflops...    

AI cs3 + itunes + xbench + safari on background...    

[ARMAGEDDON]

Power Mac G5 Quad 2.5GHz, 4GB RAM, OS X 10.4.11.

38.2 sec, 48129.7 MFlops.

Power Mac G4 Quicksilver 2001 Dual 7448 2.0 GHz OC, 1,5GB RAM, OS X 10.4.11.

44.4 sec, 22053.1 MFlops (OS X)

39.0 sec, 24355.0 MFlops (OS 9.2.2)

P.S.: Very well running...