As the graph shows, the GeForce3 card has higher memory bandwidth that widens the gap at higher resolutions.
Here's his post which notes other details and tests (Cinebench, Fractal Carbon demo, etc.)

" got back from the apple store (glendale) with my new dp1g machine about three hours ago. this is what i have to report so far:

osx 10.1.2 (5p63)
dvd-rw dvr-104
ram was 512 cl3 (i took it out)
seagate barricuda hd

smp bench 6 threads 531%
testbench x 245%
a.fractal 7.625 gflops
dnetc 20.8Mkeys
cbench 8.7, 10.6, 12.4, 21.8

here's the good stuff. quake with same config (fresh + s_mixahead .1). rez, gf3, gf4mx (in fps):

0640 182 179
0800 174 159
1024 152 122
1280 101 077
1600 070 049

i'm putting pix/shots/etc here: http://duxbury.la/dp1g/dp1g_1.html
brett. "

RC5 Benchmark Comparison - 1GHz Dual G4 vs Other Multi-Processors: (from a reader's email today)

" Hi Mike,
Somebody listed a G4/1000 DP on the RC5 multi-processor speeds website. I've compiled a summary of some of the fastest chips and converted it into an attached html table. Of course, the G4 wins handily. Your readers might find this interesting/inspiring.
-=s=-"

Here's the table Steve sent:

Memory Bandwidth Test Results: (see farther down for GaugePro and Throughput results, etc.)
Brett later posted in the forums:

" about memory. yesterday a man named basil who said he was studying cross-platform memory performance asked me to run a program called "memperf": http://www.cs.inf.ethz.ch/CoPs/ECT/

i did and asked him how the dp1g stacked up. this was his reply (proviso: these were his initial impressions):

Not too bad for a Mac. Compared to an Athlon:

• L1 cache: not very good compared to an Athlon, but that's a matter of clock rates, definitely an improvement. It's still only 32KB, AthlonXP: 128KB
• L2 cache: (small improvement)
• sequential read: good
• sequential write: Athlon is much better
• random access read: better than Athlon
• random access write: same as Athlon's L2
• L3 cache: Most Athlon systems don't have L3 caches. Instead they use DDR memory. Since the Mac's new L3 cache also is DDR, we have Apple's DDR Ram competing with the Athlon's big main memory DDR here. Apple's L3 is MUCH faster (1.2x sequential read, 2.5x sequential write, 4-6x random read/write) (P4/Rambus, random: even worse), doesn't surprise me much since it's running at higher clock speed.
• Main memory: PowerMac vs. Athlon: SDRAM vs. DDR:
  sequential read: Athlon wins
  random read/write: Mac clearly wins
  read/write combined, sequential: Mac clearly wins
  read/write combined, random: Mac clearly wins

It's also faster than the memory of the old 733 and 867 PowerMacs (couldn't run the tests on a dual 800, so if you have one... ;-), which was pretty slow.
fwiw. and cpu temp is still a constant -1C.
brett. "

Brett's GaugePro [v1.1 available here] results are shown below (higher than 7450 G4s, but still lower than 7400/7410 CPUs, even my PB G4/400 results were over 230MB/sec with its 7410 CPU. However the 7450 CPU generally had appx 150MB/sec or lower rates). As noted last year, GuagePro does not accurately report 7450 or later G4s as far as CPU type or speeds. (See differences in the two reports - GaugePro results vary a lot so it's not a good benchmark really.)

Brett's results

(three CL2 256MB dimms)

Brett later wrote "I used v1.1 of gaugepro. i was just on a different tab. also, virtual memory on/off had no effect on the #s. the only differnce between my machine and jon's is that i switched out the apple 512mb cl3 ram module for three crucial 256mb cl2 modules prior to testing."

Jon's Dual 1GHz results with v1.1 I sent him (run after a clean reboot since GaugePro results vary a lot if other things are running). Jon's results were with 1GB of RAM (CL3 I assume). With 1GB of RAM, so VM is disabled in OS 9. [Brett noted no difference with VM on or Off] GaugePro memory rate results can vary, so it's not a good benchmark really. (See below for Stream benchmark results)

Jon's results

(verifed v1.1 used, 1GB RAM, VM off)

Stream Results: STREAM is a cross-platform memory benchmark. There are no "G4 optimized" Stream versions, but the 601 optimized version seems to deliver the best results with a G4. Here's Jon's results with his dual g4 1Ghz with 1GB of RAM.

Jon's results

(1GB RAM, VM off)

Brett sent results of the same Stream test with his Dual 1GHz:

Brett's results

(3 CL2 256MB dimms)

Throughput 1.5 Results: [info/download available here.]

Brett's Results:

(Using GeForce3 card)

Another Dual G4 1GHz owner sent Throughput results showing much higher scores than Brett's. ( Brett later noted using a Geforce3 vs the OEM GeForce4MX used by Jon below.)

Jon's Results:

(GeForce4MX card w/shipped drivers)

Unreal Tournament Tests: UT is more CPU bound than graphics card and the "wicked400" demo is a good workout for the system (lots of action/affects/firing). A reader posted his Dual 1GHz G4 UT/Wicked400 demo results:

" I ran a few quick Wicked 400 tests

Dual G4/1000 512 cl3 (apple stock)
GeForce 4MX
80GB (Seagate ST380021A)
[High/Low refer to detail settings]

640 x 480
HIGH avg 30.54 min 20.33 max 55.34
LOW avg 50.49 min 35.82 max 72.62

800 x 600
HIGH avg 29.74 min 19.59 max 45.29
LOW avg 48.85 min 34.37 max 69.53

1024 x 768
HIGH avg 29.05 min 19.13 max 45.80
LOW avg 46.71 min 33.08 max 66.86

Virtual Memory and Apple Talk off. 32 bit. (0 min frame rate set in prefs) 210mb allocated to UT. Pretty much the same fps throught all resolutions. I'll try with a Cube/Radeon 32mb Friday afternoon and see how it compares. It's noiser than the G4/867. In the house I can hear it down 2 flights of stairs.
bigguy "

iMovie2 Export Tests (OS X): T. Synder sent results of my iMovie2 Export test used in many reviews in the last year or so. (Stacking all the clips from the Tutorial project end-to-end with no transitions and then timing with a stopwatch how long it takes to export to "CDROM Medium" settings.) Here's his results with the Dual 1GHz, my prev. OS X 10.1 tests with a Dual G4/500 and a reader's Dual G4/800 results:

• Dual G4 1GHz: 60 Seconds
• Dual G4 800Mhz: 77 Seconds
• Dual G4 500MHz: 114 Seconds

(My DP500 results and the dual 800mhz results are from this previous OS X 10.1 vs 10.0.4 article.)

GeForce4MX Resolutions: (from a reader's email)

"Attached is a screen shot of what resolutions work on my Viewsonic P815 21" monitor. It is attached to my G42G nVidia MX4 twinview next to the Apple 17" display.
Jon"

GF4MX modes w/Viewsonic P815 21" CRT

For previous System tests, see the Systems page. Other Graphics card comparisons/reviews/articles are on the Video articles page, graphics card section.