AmigaONE benchmark - Important

[Hammer]

==========================================
Athlon XP 1800+, ASUS nForce 2, WinXP-SP1

[Feb 18 22:25:23 UTC] Automatic processor type detection found
                      an AMD K7-6 (Athlon XP/MP/-4) processor.
...
[Feb 18 22:25:44 UTC] OGR: using core #1 (GARSP 5.13-B).
[Feb 18 22:26:03 UTC] OGR: Benchmark for core #1 (GARSP 5.13-B)
                      0.00:00:16.31 [11,800,528 nodes/sec]
==========================================

I can probably squeeze little more performance IF I turn off the network services (e.g. bridges, internet), Norton Anti-Virus 2002, MSN messenger, Asus probe, WinXP’s GUI, OBDC services, etc and increase priority for dnetc task.

For some reason, my Athlon XP 1800+ box is faster than ksk's Athlon XP 1800+ native box.

[Hammer]

You have it the opposite way round. x86 needs modern DDR-RAM for speed because of legacy CPU design,
 

What do you mean by "legacy CPU design"?

For AMD Athlon XP’s case, the DDR architecture was inherited from DEC's Alpha AXP architecture.  What do you expect from same engineers who designed the Alpha CPU?

Athlons XPs works fine enough with KT133A (that's with 133Mhz SDRAMS) chipsets.  

PS; L1/L2/L3 Cache setup works fine(i.e. a speed boost for the given core) for K6 Core( for K6-III product line)....

while PPC is the opposite and is not sped up much by extra external RAM but is sped up greatly by extra cache.

One should be focusing on bandwidth throughput instead of the buzzwords i.e. what is the bandwidth throughput of PPC G4 while it talks to the outside world i.e. 64bit x 133mhz perhaps.

[Hammer]

AmigaPro wrote:
I think its interesting how poorly all of these chips do, both AmigaOne and x86, compared to the venerable old CS-PPC...

http://amigapro.com/Images/280.gif

The old CS-PPC (mine is slightly oc'd there, to 66/280MHz) is just under half as fast in OGR as the AmigaOne G4/800, and it completely buries the Athlon per MHz.  The 604e has a better FPU than both chips as well.

IPC(instruction per cycle) alone doesn’t rescue a CPU’s fortunes…  

One should be focuing on IPC x Mhz = Performance.
PS; Athlon XP with the Barton Core should improve the IPC side of things.  

[Hammer]

>Does any of this x86 vs PPC matter?

Well, somebody made a comparison with Pentium 4 2.4Ghz and X86 in general.

[Hammer]

Focal_Loco wrote:
Hello!  I am not much of a power user, nor do I consider myself a knowledgeable computer user, but I found this intersting bit of information on Digita's Organizer for PC, on an article entitled "Evolution of the CPU"  'The Pentium ...  'interestingly,  the awkward and old architecture is such a barrier to improvement, that most of the Pentium-compatible CPU's (NextGen Nx586, AMD K5, Cyrix M1, and even the Pentium's replacement, a 2-chip-14-stage CPU/Cache module named 'P6', do not even clone the Pentium, but emulate it with specialized hardware decoders which convert Pentium instructions to RISC-like instructions which are executed on specially-designed superscalar RISC cores, actually faster than the Pentium itself.  One rumour has it that IBM is developing hardware to translate Pentium instructions to run on a PowerPC CPU  called the 615.'  So maybe all the "modern" enhancements to x86 are just workarounds to the CPU's limitations.

The K7 and K8 families does that trick (i.e. breaking down the X86 instructions into RISC like instructions before feeding into the main pipeline stages).

IBM's PPC 970 also has decoding/crush stage  for PPC instructions.

[Hammer]

But, with a G4 @ 800Mhz getting 11,000,000+ nodes is pretty impressive indeed. That surely ends the argument of "More Mhz means faster/better"

AMD has stated instructions per cycle x clock speed = performance.

"More Mhz means faster/better" is dependant on the targeted market.  

For example, Pentium 4’s target market is well suited for the entertainment market i.e. play games with very high FPS such as in Quake III, Unreal Tournament and 'etc'.  AthlonXP/NForce2 combo should be competitive in that regard.

One should be focusing on end product’s performance and price.  “Instructions per cycle” by itself is pretty point less. It’s just an academic exercise.

[Hammer]

ksk wrote:
Has anyone yet run SPECmark, Bytemark or any other little tougher tests?

How does the L2 and L3 cache show up in those benchmarks?

Just get some late model Power Macintosh’s benchmarks.

[Hammer]

ksk wrote:

mips_proc wrote:

Really AmigaONE G4-800Mhz wins over P4 2,6Ghz, this without level3 cache?

if it where only true...
...

But it IS true. On those tiny little tings, like OGR/RC5.  ;-)

On the other tiny little tings e.g. OpenSSL, the outcome is quite different for PPC.

The ideal benchmark should be a multi-combo benchmarks.

For OpenSSL benchmarks refer to
http://www.macosxhints.com/article.php?story=20020113045343563

[Hammer]

After bbrv's recent hints about A1G4XE's pathetic performance,

Any numbers?

 I want to see benchmarks run on the A1. (but also benchmarks run on MOL should be ok)

What for? Does Pegy run "AmigaOS 4.0"?
I have enough Linux boxes inside my house..

“AmigaOS 3.9” compatibility can be obtain via the cheaper X86 based solutions (still waiting for Berniethlon).

The JAVA version: http://www.netlib.org/benchmark/linpackjava/

Dependant on the Java virtual machine.  Good for benchmarking between Java VMs btw.

[Hammer]

ksk wrote:

Hammer wrote:
...
For OpenSSL benchmarks refer to
http://www.macosxhints.com/article.php?story=20020113045343563

 :-o  :-o  :-o  :-o  :-?
 

Its just an illustration on how some benchmarks can change a particular outcome. OpenSSL benchmark doesn’t tell the real picture except for OpenSSL based tasks.

Ideally, one should use a combo benchmarks.

Oracle or MySQL benchmarks would be nice for my case.

[Hammer]

Just that Peg1G3@600 "performs better than" A1G4XE.

Need more than that i.e. refer AMD's benchmark disclosures as an example.

I woud like to get rid of my noisy x86 kludge HW.

Would that change the fortune of AmigaOS significantly?

But when using identical JVM, it's also good for benchmarking system performance, right?

Some of benchmarks doesn't reveal Java VM type.

Refer to http://www17.tomshardware.com/graphic/20021218/vgacharts-05.html
Athlon XP/nForce VS Pentium 4 in regards to entertainment market segment.

[Hammer]

ksk wrote:

BTW: I think this benchmark digs out some possible memory bus bottlenecs: http://www.netlib.org/linpack/

The JAVA version: http://www.netlib.org/benchmark/linpackjava/

Using IE 6.0 SP1(via MS Java VM), WinXP Pro SP1,  Athlon XP 1800+(1.5Ghz), ASUS A7N8XD-D. This test box delivers 127.795 Mflops.

(Point scoring can be increased by switching  to ram's bandwidth to PC2700 (with suitable 333Mhz FSB CPU), I’m currently using PC2700 DDR RAMs @ PC2100 speeds).  

PC3200 could be use for 400Mhz FSB Athlon XP (o/c)for the maximum potential of EV6 bus).

This test box has beaten Ron Buss's "124 Mflop/s;  AMD XP 2700+; 2/10/03" results.

I don't know what has happen to Ron Buss's Athlon XP 2700+ box... It should be faster compared to 266Mhz FSB class 1.5Ghz Athlon XP CPUs .

I may try Sun's Java VM later.

PS; I'm well aware of IBM's PPC 970 (release @ 2003 H2). I don't know why you have posted this? This product has yet to materialized in the market place.

Secondly, the IBM PPC 970 has competitions from Intel’s IA-64, Intel’s Pentium 4 (Prescott core) and AMD’s X86-64.

Noted that IBM PPC 970's has an estimated of 55 million transistors (similar to Pentium 4 and Athlon XP (Barton core). So much for peace and quiet.

[Hammer]

Adobe's Great Render Race (PIV vs Dual-G4).
=======================================================================
Dell's Pentium IV 3.06Ghz VS Apple's 1.25Ghz Dual-Proprocessor Power Macintosh G4.

Adobe's applications used in benchmarking are;
+ After Effects®
+ Illustrator®,
+ Photoshop®

Click here for more info.

Just a side note; Intel seems to have embedded X86-64 features into the Prescott/Nocona core.

Click here for more information.