..and anybody else interested. I have some memory/vram bandwidth results from X-Ray's CSPPC/060 + Picasso-IV
I've emphasised the interesting values:
First the old PixelTest, which runs 680x0 native:
Surface width: 640, height: 480, modulus: 0
Surface hwWidth: 640, hwHeight: 480
Test data pixel format
Bytes : 2, endian swapped
Bits : A[ 0] R[ 5] G[ 6] B[ 5]
Offsets : A[ 0] R[ 11] G[ 5] B[ 0]
Maxima : A[ 0] R[ 31] G[ 63] B[ 31]
Window pixel format
Bytes : 2, endian swapped
Bits : A[ 0] R[ 5] G[ 6] B[ 5]
Offsets : A[ 0] R[ 11] G[ 5] B[ 0]
Maxima : A[ 0] R[ 31] G[ 63] B[ 31]
Read RAM : 63000.00 K/sec
Write RAM : 62562.07 K/sec
RAM->RAM : 29702.97 K/sec
RAM->RAM (OS) : 27245.80 K/sec
RAM->RAM[16] : 35294.12 K/sec
Read VRAM : 3674.54 K/sec
Write VRAM : 9714.29 K/sec
RAM->VRAM : 9677.42 K/sec
RAM->VRAM[16] : 9590.41 K/sec
VRAM->RAM : 3345.72 K/sec
VRAM->RAM[16] : 3636.36 K/sec
Conversion : 9677.42 K/sec [output bandwidth]
Conversion : 4954838.71 pix/sec
Conversion attained 100.00% copy speed
It would appear that the move16 based copy to VRAM has no benefit for this card. It gives a significant boost to the CVisionPPC/BVisionPPC cards (about 30% more usually).
Next the memspeed test, which runs PPC native under WarpOS:
Memory/Bus bandwidth estimation (WarpOS/PPC) (c) Karl Churchill 2004
System Info
CPU : 604e [PVR:0x00090204]
CPU : 233.333 MHz
FSB : 66.667 MHz
Fast [cache normal] allocated : 33554432 bytes at 0x091DC020
Fast [writethrough] allocated : 33554432 bytes at 0x0B1DE020
Chip [noncacheable] allocated : 1048576 bytes at 0x0001C020
VRAM [noncacheable] allocated : 614400 bytes at 0x491D4C20
Estimating VRAM access bandwidth
Read VRAM (64-bit) : 4085.50 KB/s
Read VRAM (32-bit) : 3870.67 KB/s
Read VRAM (16-bit) : 1938.11 KB/s
Read VRAM (8-bit) : 970.04 KB/s
Write VRAM (64-bit) : 9757.70 KB/s
Write VRAM (32-bit) : 9764.06 KB/s
Write VRAM (16-bit) : 4881.67 KB/s
Write VRAM (8-bit) : 2447.77 KB/s
Estimating fast RAM access bandwidth [normal]
Read FAST (64-bit) : 113196.66 KB/s
Read FAST (32-bit) : 107050.43 KB/s
Read FAST (16-bit) : 93402.03 KB/s
Read FAST (8-bit) : 77159.73 KB/s
Write FAST (64-bit) : 72027.53 KB/s
Write FAST (32-bit) : 72082.94 KB/s
Write FAST (16-bit) : 72018.26 KB/s
Write FAST (8-bit) : 71870.02 KB/s
Estimating fast RAM access bandwidth [writethrough]
Read FAST (64-bit) : 113254.54 KB/s
Read FAST (32-bit) : 107072.29 KB/s
Read FAST (16-bit) : 93383.50 KB/s
Read FAST (8-bit) : 77159.69 KB/s
Write FAST (64-bit) : 70959.87 KB/s
Write FAST (32-bit) : 35614.40 KB/s
Write FAST (16-bit) : 17889.43 KB/s
Write FAST (8-bit) : 8988.15 KB/s
Estimating chip RAM access bandwidth
Read CHIP (64-bit) : 5438.37 KB/s
Read CHIP (32-bit) : 5442.07 KB/s
Read CHIP (16-bit) : 5435.64 KB/s
Read CHIP (8-bit) : 5431.88 KB/s
Write CHIP (64-bit) : 3223.18 KB/s
Write CHIP (32-bit) : 3222.22 KB/s
Write CHIP (16-bit) : 3223.98 KB/s
Write CHIP (8-bit) : 3223.14 KB/s
Estimating fast RAM [normal] to fast RAM [normal] bandwidth
FAST -> FAST (64-bit) : 73863.01 KB/s
FAST -> FAST (32-bit) : 73679.63 KB/s
FAST -> FAST (16-bit) : 50373.24 KB/s
FAST -> FAST (8-bit) : 26862.58 KB/s
Estimating fast RAM [normal] to fast RAM [writethrough] bandwidth
FAST -> FAST (64-bit) : 46512.67 KB/s
FAST -> FAST (32-bit) : 28161.36 KB/s
FAST -> FAST (16-bit) : 15786.53 KB/s
FAST -> FAST (8-bit) : 8428.53 KB/s
Estimating fast RAM [normal] to VRAM copy bandwidth
FAST -> VRAM (64-bit) : 9705.18 KB/s
FAST -> VRAM (32-bit) : 9706.10 KB/s
FAST -> VRAM (16-bit) : 4873.94 KB/s
FAST -> VRAM (8-bit) : 2442.94 KB/s
Estimating VRAM to RAM [normal] copy bandwidth
VRAM -> FAST (64-bit) : 3844.42 KB/s
VRAM -> FAST (32-bit) : 3545.00 KB/s
VRAM -> FAST (16-bit) : 1813.46 KB/s
VRAM -> FAST (8-bit) : 915.23 KB/s
Estimating VRAM to RAM [writethrough] copy bandwidth
VRAM -> FAST (64-bit) : 3868.92 KB/s
VRAM -> FAST (32-bit) : 3476.85 KB/s
VRAM -> FAST (16-bit) : 1741.90 KB/s
VRAM -> FAST (8-bit) : 870.36 KB/s
Estimating FAST [normal] to CHIP copy bandwidth
FAST -> CHIP (64-bit) : 3074.22 KB/s
FAST -> CHIP (32-bit) : 3125.90 KB/s
FAST -> CHIP (16-bit) : 3122.60 KB/s
FAST -> CHIP (8-bit) : 3112.73 KB/s
Estimating CHIP to FAST [normal] copy bandwidth
CHIP -> FAST (64-bit) : 5086.79 KB/s
CHIP -> FAST (32-bit) : 5168.88 KB/s
CHIP -> FAST (16-bit) : 4955.48 KB/s
CHIP -> FAST (8-bit) : 4566.40 KB/s
Estimating CHIP to FAST [writethrough] copy bandwidth
CHIP -> FAST (64-bit) : 5163.28 KB/s
CHIP -> FAST (32-bit) : 4735.89 KB/s
CHIP -> FAST (16-bit) : 4189.41 KB/s
CHIP -> FAST (8-bit) : 3394.25 KB/s
It would appear that the Picasso-IV absolutely maxes out at about 9600K/s when writing to VRAM as can be seen from the PPC and 680x0 both.
The BVisionPPC (on 040) maxes out at about 13MB/s for a move16 based copy, the PPC maxes out at about 16MB/s. The 060 versions of the card are a bit higher still (even on the PPC side).
From other results (which I will need to dig out) it is clear the Mediator/Voodoo combo is not any faster for system <-> vram than the good old P-IV :-)
