This is an interesting benchmark of pure FPU grunt. Is there any site that
collects these results? If not, I think it'd be a great idea.

Summary of systems (best viewed with a fixed width font :):

| CPU         | RAM | Time | PPSPMHZ* |
|-------------|-----|------|----------|
| Athlon 950  | ??? |   56 |   14.78  |
| Athlon 1200 | 128 |   49 |   13.37  |
| Athlon 800  | 128 |   77 |   12.77  |
| P3 800      | 256 |   77 |   12.77  |
| G3 350      | 128 |  188 |   11.95  |
| Celeron 400 | 192 |  168 |   11.70  |
| P 200 MMX   | ??? |  540 |    7.28  |
| K6-2 400    | 192 |  400 |    4.92  |
| ???         | ??? |  699 |     ???  |
|-------------|-----|------|----------|
* Pixels per second per megahertz

Very interesting results, especially for the Athlons, though quite
explainable. The Athlon 950 is faster in PPSPMHZ (I've got to think of a
better acronym!) because at this speed, the limit would be how fast you can
move stuff too and from RAM. Therefore, the score in PPSPMHZ woudl drop as
you got faster speeds (more PPS but a greater increase in MHZ). The Athlon
800 probably has a lower bus speed.

Another good indicator would be PPSP$ (pixels per second per $).

| CPU         | RAM | Time | Price | PPSP$ |
|-------------|-----|------|-------|-------|
| Celeron 400 | 192 |  168 |   40  | 117.0 |
| Athlon 800  | 128 |   77 |   98  | 104.2 |
| Athlon 950  | ??? |   56 |  153  |  91.8 |
| Athlon 1200 | 128 |   49 |  267  |  60.1 |
| K6-2 400    | 192 |  400 |   33  |  59.6 |
| P3 800      | 256 |   77 |  180  |  56.7 |
| P 200 MMX   | ??? |  540 |  ???  |   ??? |
| G3 350      | 128 |  188 |  ???  |   ??? |
| ???         | ??? |  699 |  ???  |   ??? |
|-------------|-----|------|-------|-------|


Shame on us all! The Celery wins! Of course, this is probably not a good
test as a single Athlon 800 ($163) would be far cheaper than buying 2
Celeron 400's and networking them together. However, a BP6 with two Celeron
400 would be nice (and cheap).

Okay, that's the end of my ramblings. No flame on the rankings please :)

Michael

Post a reply to this message

On 2001-01-26 23:49, Michael Brown <emb### [at] i4freeDELETETHISconz> wrote:
>This is an interesting benchmark of pure FPU grunt. Is there any site that
>collects these results? If not, I think it'd be a great idea.

http://www.haveland.com/povbench/


>Summary of systems (best viewed with a fixed width font :):
>
>| CPU         | RAM | Time | PPSPMHZ* |
>|-------------|-----|------|----------|
>| Athlon 950  | ??? |   56 |   14.78  |
>| Athlon 1200 | 128 |   49 |   13.37  |
>| Athlon 800  | 128 |   77 |   12.77  |
>| P3 800      | 256 |   77 |   12.77  |

I was surprised that the P3 800 is exactly as fast as the Athlon 800 so
I tried it on our P3's:

 | P3 800      |1024 |   94 |   10.46  |
 | P3 600      | 256 |  134 |    9.78  |
 | P3 500      | 256 |  172 |    9.14  |

Quite a bit lower. 

Is your P3 result correct, or did you mistakenly put in the Athlon
results a second time? 

Of course that could also be a compiler issue: All our three boxes are
running Linux. I used the official build on the 800 and 600 and compiled
it myself on the 500.

Also, all three boxes have an ASUS board. 


>PPSPMHZ (I've got to think of a better acronym!)

Since BPB (bang per buck) seems to be a used for the performance/cost
ratio, how about BPC (bang per clock)? :-)

	hp

-- 
   _  | Peter J. Holzer    | All Linux applications run on Solaris,
|_|_) | Sysadmin WSR       | which is our implementation of Linux.
| |   | hjp### [at] wsracat      | 
__/   | http://www.hjp.at/ |	-- Scott McNealy, Dec. 2000

Post a reply to this message

First things first:

> I was surprised that the P3 800 is exactly as fast as the Athlon 800 so
> I tried it on our P3's:
These aren't my results - apart from the K6-2 at the bottom :( - merely just
what was reported in this thread

> Since BPB (bang per buck) seems to be a used for the performance/cost
> ratio, how about BPC (bang per clock)? :-)
Definately better!

Then on to the fun part. I crunched the numbers from the site you supplied
and came up with these results (plus more for Sun SPARCs, G3s etc):

Processor           |  Min  | Average | Max   | Notes
--------------------|-------|---------|-------|---------------------------
DEC Alpha 21264     | 32.75 |  50.89  | 60.49 |
AMD Thunderbird     | 22.93 |  30.39  | 36.41 |
AMD Athlon          |  3.55 |  29.45  | 41.39 | Includes some Thunderbirds
AMD Duron           | 26.75 |  28.95  | 35.59 |
Intel Pentium2 Xeon | 20.81 |  23.87  | 26.21 | V. small sample
DEC Alpha           | 17.28 |  23.64  | 44.94 | Unspecified chips
Intel Pentium3      | 17.28 |  23.64  | 44.94 | Coppermine + classic
Intel Celeron       |  4.44 |  22.73  | 37.66 | All celerons
Intel Pentium2      |  9.13 |  22.11  | 44.06 |
AMD K6-3            |  9.20 |  21.49  | 29.13 |
DEC Alpha 21164     |  4.90 |  21.27  | 34.04 |
Intel Pentium Pro   |  7.53 |  20.28  | 34.09 |
Cyrix M2            | 14.40 |  18.58  | 27.31 | V. small sample
AMD K6-2            | 12.52 |  18.28  | 40.64 |
AMD K6              | 10.83 |  17.96  | 29.13 |
Intel Pentium4      |       |  17.02  |       | One only
Intel Pentium       |  5.63 |  16.75  | 27.65 |
Intel Pentium MMX   |  6.68 |  15.86  | 21.63 |
Cyrix 6x86MX        |  8.90 |  14.82  | 24.55 | V. small sample
Cyrix 6x86          |  7.61 |  12.50  | 26.75 |
DEC Alpha 21064     |  4.53 |  12.07  | 17.99 |
AMD K5              |  3.33 |  10.59  | 18.83 |
486DX               |  4.40 |  7.82   | 10.12 |
Cyrix MediaGX       |  5.40 |  5.63   |  5.83 | V. small sample
--------------------|-------|---------|-------|---------------------------

These numbers are in BPC (bang ber clock) and are not really useful except
for comparing the relative FPU grunt of each processor. Even then it's
sometimes not useful as things like memory bandwidth limitations kick in and
push the results down for higher speed processors. What I'm saying is that
these numbers have almost no practical use, and the better one would be the
BPB (bang per buck) rating of each which you can get from the site mentioned
in the last post and which I forgot to include, and can't include because
I've already posted this :)

The only use is to see that the FPU in the Alpha (the chip of my dreams :)
should really be looked at by Intel and AMD! Hehe, imagine that. Skyvase.pov
run on a TBird 1000 at a BPC rating of 50. Hmm, that's be 15.7 seconds.
Oooooohhhh yeeeaah.

Moving right along, this is how I see the results. There are several
unexpected things and several expected things (as there are with most things
:).

Expected:
1) Alpha tops the list by a large margin.
2) TBird/Athon comes second (I'm an AMD fan :)
3) Pentium 4 rates not-so-good with its high clock speed limited by memory
bandwidth.
3) Xeon beats the P2
4) K6-2's lose out to the P2 (nooooooooooooo!)

Unexpected:
1) Celeron beats the P2 (by a small margin, but lumping all the celerons
together, includeing the handicapped ones should drag it down).
2) Xeon doesn't totally cream the P2. With all the cache it should leave the
P2 in the dust!
3) Pentium MMX loses to the plain vanilla pentium. Again, a memory bandwidth
thing again.
4) The MediaGX (MMX equivilent I thought) gets smoked by even a 486!

And maybe some more but I can't be bothered analysing it any more. It's
11:30 at night and I'm falling asleep.

If anyone wants the raw results in XL form (it was a real pain trying to
extract it from the HTML, and during the process I screwed up all except the
speed, time, and processor) just e-mail me after removing the obvious from
the address (or post here if I don't respond within 24 hrs). Also, if you
want the results for antything else I've probably got it (or you can visit
the site :)

Cheers,
Michael

PS: Seeing a 7mhz 68000 take 2:12:14 to render it makes me feel better :)

Post a reply to this message

I have only 48Mb of RAM in my P200MMX.. dunno how much of a detrimental
affect that would have had on the results.

--
Andy Cocker

---------------------------------------------------------------
listen to my music at:
www.mp3.com/lunarland
---------------------------------------------------------------

'I spilled spot remover on my dog. He's gone now. '
'I went to a restaurant that serves "breakfast at any time."
So I ordered french toast during the Renaissance. '

 - Steven Wright.
---------------------------------------------------------------

Post a reply to this message

The kindly Rev. overheard "Michael Brown"
<emb### [at] i4freeDELETETHISconz> saying on 26 Jan 2001: 

>This is an interesting benchmark of pure FPU grunt. Is there any
>site that collects these results? If not, I think it'd be a great
>idea. 
>
>Summary of systems (best viewed with a fixed width font :):
>
>| CPU         | RAM | Time | PPSPMHZ* |
>|-------------|-----|------|----------|
>| Athlon 950  | ??? |   56 |   14.78  |
>| Athlon 1200 | 128 |   49 |   13.37  |
>| Athlon 800  | 128 |   77 |   12.77  |
>| P3 800      | 256 |   77 |   12.77  |
>| G3 350      | 128 |  188 |   11.95  |
>| Celeron 400 | 192 |  168 |   11.70  |
>| P 200 MMX   | ??? |  540 |    7.28  |
>| K6-2 400    | 192 |  400 |    4.92  |
>| ???         | ??? |  699 |     ???  |
>|-------------|-----|------|----------|

Hmmmmm.  Better tweak my motherboard, I think.  10.2613 PPSPMHZ on my 
1.2 GHz Athlon that just arrived. 11.15 with the VC6 engine.

Still, 23 (VC6 engine) or 25 seconds (original engine) on POVBench with 
my antivirus, ZoneAlarm, WebWasher, Winamp Agent, etc., etc., all 
running makes a somewhat nicer environment for raytracing than my 5-
year-old P133 which runs the skyvase POVBench in... hang on, it's still 
running... almost done... zzzzzz... huh?... oh yeah, where was I?...  
458 seconds!

OMG!  It runs the POVBench 18.32 times faster!  Maybe I'll leave that 
mobo alone.

Anybody wondering about the paucity of my output in p.b.i had better 
hang on to their hats.

FYI, the new box has got 512MB SDRAM, two 45GB 7200RPM IBM disks 
configured as RAID0 running through a Promise Ultra ATA100 onto an Asus 
AK74-EC motherboard, stuffed (along with a VisionTek GeForce 2 Ultra 
board, a Soundblaster Live Platinum, a Pioneer DVD-ROM, a TDK CD-ROM 
burner, and a Linksys 10/100 network card) into a SuperMicro SC750a 
case with 4 case fans.  Alas, using an external BitSurfr Pro to 
connect, since the ^*#*$&s at the phone company and the cable company 
haven't got broadband to my neighborhood yet.  Runs Windows ME, which I 
semi-hate.  Bose speakers, Sony Trinitron monitor, Logitech wireless 
keyboard and mouse.

Sorry for all the bragging.  But just once before I die I wanted to 
have the best computer in my neighborhood.  Er, best until April, when 
they start shipping the new Microns.  Fame is fleeting.  But I can live 
with that.

Haven't overclocked the CPU or video card.  Yet.  The next upgrade, 
when memory prices have fallen from their present absurd level (and my 
wallet has recovered from this absurd expenditure), is another 
motherboard and DDR SDRAM.  So my chances of having the best computer 
on my street are sunk if any of my neighbors has way too much money.  
US$350 for 128MB?????

My only excuse for my current exercise in wretched excess is that I 
haven't bought a new PC since November of 1995.

However, I believe I'm about to become a raytracer!

Other renders:

fish13      48
drums      145
ionic5      33
teapot2     82
teapot3     87
teapot4    123
ntreal      68
piece2     226
piece3      36
teapot      13
chess2     613
desk        21
diffract    41
float5      19
lamppost    89
mist        28
mtmand      19
quilt1     100
skyvase     35
sombrero    28
sunsethf    64
swirlbox    29
wg6        110
whiltile   136
woodbox     51

All done 800x600 0.3AA, .bmp output, original povengine.exe, watching 
the render, Opera and Xnews running.
-- 
Rev. Bob "Bob" Crispen
crispen at hiwaay dot net

If you get to thinkin' you're a person of some influence, try
orderin' somebody else's dog around.

Post a reply to this message

Rev. Bob 'Bob' Crispen wrote in message ...

>US$350 for 128MB?????


Consider this: six years ago, that 128MB would have run you US$12,800.

--
Mark

Post a reply to this message

> I have only 48Mb of RAM in my P200MMX.. dunno how much of a detrimental
> affect that would have had on the results.
Not much, I would say (unless it swapfiles). All these results are done for
640x480, aa 0.3, and a whole lot of other stuff. And a older skyvase.pov.
Check out http://www.haveland.com/povbench/povform.htm for details.

Post a reply to this message

"Michael Brown" <emb### [at] i4freeDELETETHISconz> writes:

> Shame on us all! The Celery wins! Of course, this is probably not a good

My moderately overclocked Duron gets 171 PPSP$ 
Stats: Duron 700@900 MHz, 256 MB, 63 s on skyvase.pov 1024*768.

-- 
ICQ 74734588

Post a reply to this message