For those interested in benchmarks, these are the results from
an AMCC 440EP 533 Mhz evaluation board running the Stream
benchmark under Linux 2.6.18

/usr/checkbench/stream # ./stream_d
-------------------------------------------------------------
This system uses 8 bytes per DOUBLE PRECISION word.
-------------------------------------------------------------
Array size = 2000000, Offset = 0
Total memory required = 45.8 MB.
Each test is run 10 times, but only
the *best* time for each is used.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 112887 microseconds.
(= 112887 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Copy: 262.6503 0.1221 0.1218 0.1223
Scale: 245.8928 0.1302 0.1301 0.1303
Add: 253.5979 0.1893 0.1893 0.1893
Triad: 251.2734 0.1911 0.1910 0.1912

For comparison, these are the results from an AOne 750Fx 800 Mhz
and an Amiga 4000T 604e 180 Mhz both running AmigaOS4:


-------------------------------------------------------------
Function Rate (MB/s) Avg time Min time Max time
Copy: 153.4213 0.2094 0.2086 0.2106
Scale: 140.3010 0.2289 0.2281 0.2301
Add: 143.2011 0.3360 0.3352 0.3381
Triad: 145.8209 0.3299 0.3292 0.3322
(* taken from the stream_os4.lha archive on OS4Depot)

-------------------------------------------------------------
Function Rate (MB/s) Avg time Min time Max time
Copy: 72.0192 0.3780 0.3777 0.3790
Scale: 72.0291 0.3781 0.3776 0.3809
Add: 80.1754 0.5093 0.5089 0.5124
Triad: 80.2038 0.5092 0.5087 0.5129
(* test done with Array Size = 1700000 instead of 2000000 to
fit into 64Mb installed ram)

Reassuming:

AMCC 440 EP: 262.6503 / 245.8928 / 253.5979 / 251.2734 MB/s
A1 750Fx G3: 153.4213 / 140.3010 / 143.2011 / 145.8209 MB/s
A4000T 604e: 72.0192 / 72.0291 / 80.1754 / 80.2038 MB/s

_________________
Massimiliano Tretene, ACube Systems, Soft3

@m3x

So it's about 70% faster than a G3 800MHz AOne?

Last edited by -Sam- on 26-Sep-2006 at 01:23 PM.

_________________
Sam

@m3x

It proves what we have been known from the Samantha specs - Samantha has more efficient memory controller than Articia S. For a comparision Results for Pegasos 2 (with Marvell Discovery II controller and 7447 CPU) are 330.8 / 306.3 / 297.3 / 296.0. It has to be noted however, the Stream test tells us nothing about computing power of the CPU.

Last edited by Krashan on 26-Sep-2006 at 01:31 PM.

_________________
Reggae · MorphOS Files · DigiBooster 3

@Krashan

CPU Performance of the 440EP is 1334 Dhrystone MIPS @ 667 MHz. The 750GX delivers 1856 DMIPS @ 800 Mhz. Which means that the 440EP@667MHz has about 70% of the performance of a 750GX@800 which is found in uA1's.

@Krashan

Relatively speaking, ten years old CSPPC is still amazingly fast.

_________________
Amiga Developer
Amiga 500, Efika, Mac Mini and PowerBook

@Krashan

Quote:

It has to be noted however, the Stream test tells us nothing about computing power of the CPU.

Indeed, the figures give little or no clue as to how the system will perform doing real work. And even taking the memory sub-system in isolation, those results are pretty awful. Any system with PC133 memory that wasn't designed by a monkey (ie, everything but the A1) will give better results, so the potential gained by using DDR266 is completely wasted.

_________________
Who do you serve, and who do you trust? - Galen

@itix

Quote:

Relatively speaking, ten years old CSPPC is still amazingly fast.

That was also my impression. The memory performance of the CSPPC is quite impressive for its age.

_________________
This signature is in the middle of a much needed facelift!

@m3x
Since this is an Amiga related project, all benchmarks, pictures, etc. will only be used to attack you. You should know better.

I do find the relative speeds against the A1 interesting. Can't wait to see how OS4 performs on this board while compiling some nasty C++ code compared to my A1s (see signature). Let me know when you want to send a board over and I'll give you my address--no problem, really--my pleasure.

_________________
ExecSG Team Lead

@ssolie

If it wouldn't be so tragic, your smiley would at least interpret some comments as jokes. But sometimes I have the feeling, we have some soldiers here on a crusade. Less funny.

@Krashan. Err, helllooo ?! We all know the sam has no G4, and nooohh, it's not a desktop board, and noooohh, it has no L2 cache. What's you'r point ? And yes, it has about half the bandwith of a DDR2 implementation (32bit instead of 64) - if you want to comment the Peg2 performance based on pure memory controller performance (Marvel should be twice as fast).

oh, dm...couldn't resist. Ignore my comment.

_________________
AmigaOne XE G3/933/VIA/FM801/R200 (fixed),
G3SE/600/Voodoo3/Sil680/RTL8139/SBLive! (noiseless!)
µA1-MK2/G3/933/R200/CMI8738
XE/G4 (broken 7450/800)

If this board does end up running OS4 stable, then this board will be good enough for a while for me.

@Jorge

We do not discuss Pegasos here, I only put results as a comparision. My point is that I'm interested in a desktop computer capable of running AmigaOS 4 for moderate price. I just analyse if Samantha can be such a computer. I must admit power to price relation is not very encouraging. My point of view may be a bit skewed by my monthly salary being about 500 euros , but well... I can always sell my Pegasos, but I clearly see Samantha cannot be seen as a replacement, whatever nice benchmark results it has.

Anyway I'm waiting to see Samantha running AmigaOS 4 (or maybe MorphOS, but I doubt it will ever happen...). Test are tests and there are also real life applications.

@m3x

I've read your post carefully again and found out that these tests were performed on AMCC evaluation board. Of course as memory controller is built into CPU Samantha results will be probably the same. But does it mean Samantha is not yet able to run any code, even in laboratory conditions?

_________________
Reggae · MorphOS Files · DigiBooster 3

@COBRA

The MIPS value says not much about everyday usage. The 440 lacks L2 cache. have fun while waiting for fetching data from the ram...

_________________
My programs: via.bckrs.de
MorphOS user since V0.4 (2001)

@polka.

I think the memory controller on the CSPPC is 64-bit wide, using paired 32-bit simms.

Wish there were some Efika benchmarks as well.

_________________
Cuida tus piedras gordas.

A1200/030 32Mb
A4000D
A1-X1000.

@Ami603

I see no point in benchmarks...they usually don't tell the real world performance of the machine...

And also what's the point of comparing it to other PPC cpus families?

As far as I am concerned if Sam runs OS4 it is an Amiga, the Peg isn't

_________________
www.loriano.pwp.blueyonder.co.uk

@TheDaddy

well, no more excuses for you when this samantha thing appears, you have been sitting on the fence for too long allready.

get on with it

_________________
See my blog and collection website! . https://www.blog.amigaguru.com

@Zylesea
Quote:

have fun while waiting for fetching data from the ram...

I'll be more than glad to test that theory myself.

Don't worry, we will be having plenty of fun trying OS4 on all sorts of new hardware (negotiations permitting of course).

_________________
ExecSG Team Lead

@Zylesea

Quote:

The MIPS value says not much about everyday usage. The 440 lacks L2 cache. have fun while waiting for fetching data from the ram...

You're right, the DMIPS gives you only a rough idea, there are a lot of other factors to consider. The lack of L2 cache sounds worrying, however the CSPPC boards do not have L2 cache either, yet they give you performance very similar to A1s when you scale the clock speed. Consider an MPEG-based video decoder. Most of the decoding time is spent in the IDCT algorithm and reconstruction of the motion blocks. The IDCT algorithm is performed on 6 matrices, each containing 8x8 coefficients (32 bits each), so it is about 1500 bytes of data, which easily fits into the L1 cache of any CPU so the 440 should perform just like any other. The motion vector reconstruction involves adding the decoded matrices to a referenced area from the stored previously decoded frame. In DVD resolution movies this is a 720x576x8bit luminance plane and two 360x288x8bit chrominance planes, which is over 600kBytes of data for just one reference frame. Most frames are B frames (bidirectionally predicted) which means they are decoded using two reference frames, that is over 1.2Mbytes of data being accessed for decoding a single frame. The L2 cache of the 750GX is 1MB, which is not enough to fit all that data. The L2 cache of the 750FX is 512kB and not surprisingly it decodes DVDs at exactly the same speed as the 750GX. Even DivX movies are not decoded faster by the 750GX, where there are no B frames and in theory the reference image should fit into the L2 cache, but the currently decoded image is also using up the L2 cache, removing most of the previously decoded image from the L2 cache.

So I'd say wait for the first real speed tests before drawing any conculsions.

@m3x

Impressive for a poor 440EP (533Mhz)

_________________
ikir

@ikir

Impressive indeed, a 667MHz with 1GB of ram would do nicely too

Gimme Moana now running OS4.x and we are in heaven!

_________________
www.loriano.pwp.blueyonder.co.uk