ragemem benchmark results thread !

@shadowsun

please try RamSpeed from OS4depot and report your results.

http://os4depot.net/share/utility/benchmark/ramspeed.lha

@m3x

Cyberstorm PPC 366MHz Amiga 4000, OS 4

5.Workbench:> ramspeed -b2 -g1 -m1
RAMspeed (UNIX) v2.3.0 by Rhett M. Hollander (Alasir Enterprises), 2002-04

1Gb per pass mode

INTEGER & READING 1 Kb block: 1312.82 Mb/s
INTEGER & READING 2 Kb block: 1280.00 Mb/s
INTEGER & READING 4 Kb block: 1312.82 Mb/s
INTEGER & READING 8 Kb block: 1312.82 Mb/s
INTEGER & READING 16 Kb block: 1383.78 Mb/s
INTEGER & READING 32 Kb block: 1219.05 Mb/s
INTEGER & READING 64 Kb block: 95.34 Mb/s
INTEGER & READING 128 Kb block: 95.17 Mb/s
INTEGER & READING 256 Kb block: 95.34 Mb/s
INTEGER & READING 512 Kb block: 95.34 Mb/s
INTEGER & READING 1024 Kb block: 94.46 Mb/s
5.Workbench:> ramspeed -b3 -g1 -m1
RAMspeed (UNIX) v2.3.0 by Rhett M. Hollander (Alasir Enterprises), 2002-04

1Gb per pass mode

INTEGER Copy: 70.96 Mb/s
INTEGER Scale: 70.52 Mb/s
INTEGER Add: 79.34 Mb/s
INTEGER Triad: 76.30 Mb/s
---
INTEGER AVERAGE: 74.28 Mb/s

5.Workbench:> ramspeed -b1 -g1 -m1
RAMspeed (UNIX) v2.3.0 by Rhett M. Hollander (Alasir Enterprises), 2002-04

1Gb per pass mode

INTEGER & WRITING 1 Kb block: 1280.00 Mb/s
INTEGER & WRITING 2 Kb block: 1248.78 Mb/s
INTEGER & WRITING 4 Kb block: 1219.05 Mb/s
INTEGER & WRITING 8 Kb block: 1248.78 Mb/s
INTEGER & WRITING 16 Kb block: 1248.78 Mb/s
INTEGER & WRITING 32 Kb block: 1137.78 Mb/s
INTEGER & WRITING 64 Kb block: 54.82 Mb/s
INTEGER & WRITING 128 Kb block: 54.64 Mb/s
INTEGER & WRITING 256 Kb block: 54.70 Mb/s
INTEGER & WRITING 512 Kb block: 54.70 Mb/s
INTEGER & WRITING 1024 Kb block: 54.41 Mb/s
5.Workbench:>

@HammerD

---> L1 <---
INTEGER & READING 32 Kb block: 1219.05 Mb/s
READ32: 1386 MB/Sec

INTEGER & WRITING 32 Kb block: 1137.78 Mb/s
WRITE32: 1163 MB/Sec

---> RAM <---
INTEGER & READING 1024 Kb block: 94.46 Mb/s
READ32: 98 MB/Sec

INTEGER & WRITING 1024 Kb block: 54.41 Mb/s
WRITE32: 56 MB/Sec

Quite the same results Normal, anyway.

Waiting for a Sam440 result, to see where the cache problem is. Sam, or my code.

5.RAM Disk:> ramspeed -b 1
RAMspeed (UNIX) v2.3.0 by Rhett M. Hollander (Alasir Enterprises), 2002-04

4Gb per pass mode

INTEGER & WRITING 1 Kb block: 165.29 Mb/s
INTEGER & WRITING 2 Kb block: 165.29 Mb/s
INTEGER & WRITING 4 Kb block: 165.16 Mb/s
INTEGER & WRITING 8 Kb block: 165.03 Mb/s
INTEGER & WRITING 16 Kb block: 165.16 Mb/s
INTEGER & WRITING 32 Kb block: 165.29 Mb/s
INTEGER & WRITING 64 Kb block: 165.03 Mb/s
INTEGER & WRITING 128 Kb block: 164.76 Mb/s
INTEGER & WRITING 256 Kb block: 164.23 Mb/s
INTEGER & WRITING 512 Kb block: 164.23 Mb/s
INTEGER & WRITING 1024 Kb block: 163.84 Mb/s
INTEGER & WRITING 2048 Kb block: 162.93 Mb/s
INTEGER & WRITING 4096 Kb block: 163.06 Mb/s
INTEGER & WRITING 8192 Kb block: 163.06 Mb/s
INTEGER & WRITING 16384 Kb block: 162.93 Mb/s

5.RAM Disk:> ramspeed -b 2
RAMspeed (UNIX) v2.3.0 by Rhett M. Hollander (Alasir Enterprises), 2002-04

4Gb per pass mode

INTEGER & READING 1 Kb block: 2844.44 Mb/s
INTEGER & READING 2 Kb block: 2925.71 Mb/s
INTEGER & READING 4 Kb block: 2925.71 Mb/s
INTEGER & READING 8 Kb block: 2925.71 Mb/s
INTEGER & READING 16 Kb block: 2968.12 Mb/s
INTEGER & READING 32 Kb block: 2659.74 Mb/s
INTEGER & READING 64 Kb block: 276.38 Mb/s
INTEGER & READING 128 Kb block: 275.27 Mb/s
INTEGER & READING 256 Kb block: 270.18 Mb/s
INTEGER & READING 512 Kb block: 270.18 Mb/s
INTEGER & READING 1024 Kb block: 270.18 Mb/s
INTEGER & READING 2048 Kb block: 270.54 Mb/s
INTEGER & READING 4096 Kb block: 270.54 Mb/s
INTEGER & READING 8192 Kb block: 270.54 Mb/s
INTEGER & READING 16384 Kb block: 270.18 Mb/s

5.RAM Disk:> ramspeed -b 3
RAMspeed (UNIX) v2.3.0 by Rhett M. Hollander (Alasir Enterprises), 2002-04

4Gb per pass mode

INTEGER Copy: 191.76 Mb/s
INTEGER Scale: 189.89 Mb/s
INTEGER Add: 198.00 Mb/s
INTEGER Triad: 180.07 Mb/s
---
INTEGER AVERAGE: 189.93 Mb/s

5.RAM Disk:> ramspeed -b 4
RAMspeed (UNIX) v2.3.0 by Rhett M. Hollander (Alasir Enterprises), 2002-04

4Gb per pass mode

FL-POINT & WRITING 1 Kb block: 330.32 Mb/s
FL-POINT & WRITING 2 Kb block: 330.86 Mb/s
FL-POINT & WRITING 4 Kb block: 330.32 Mb/s
FL-POINT & WRITING 8 Kb block: 330.32 Mb/s
FL-POINT & WRITING 16 Kb block: 330.86 Mb/s
FL-POINT & WRITING 32 Kb block: 330.86 Mb/s
FL-POINT & WRITING 64 Kb block: 330.32 Mb/s
FL-POINT & WRITING 128 Kb block: 329.79 Mb/s
FL-POINT & WRITING 256 Kb block: 327.16 Mb/s
FL-POINT & WRITING 512 Kb block: 327.16 Mb/s
FL-POINT & WRITING 1024 Kb block: 326.11 Mb/s
FL-POINT & WRITING 2048 Kb block: 322.52 Mb/s
FL-POINT & WRITING 4096 Kb block: 323.03 Mb/s
FL-POINT & WRITING 8192 Kb block: 323.03 Mb/s
FL-POINT & WRITING 16384 Kb block: 322.52 Mb/s

5.RAM Disk:> ramspeed -b 5
RAMspeed (UNIX) v2.3.0 by Rhett M. Hollander (Alasir Enterprises), 2002-04

4Gb per pass mode

FL-POINT & READING 1 Kb block: 1969.23 Mb/s
FL-POINT & READING 2 Kb block: 1988.35 Mb/s
FL-POINT & READING 4 Kb block: 1988.35 Mb/s
FL-POINT & READING 8 Kb block: 1988.35 Mb/s
FL-POINT & READING 16 Kb block: 1988.35 Mb/s
FL-POINT & READING 32 Kb block: 1780.87 Mb/s
FL-POINT & READING 64 Kb block: 228.32 Mb/s
FL-POINT & READING 128 Kb block: 226.80 Mb/s
FL-POINT & READING 256 Kb block: 224.07 Mb/s
FL-POINT & READING 512 Kb block: 224.32 Mb/s
FL-POINT & READING 1024 Kb block: 224.56 Mb/s
FL-POINT & READING 2048 Kb block: 224.56 Mb/s
FL-POINT & READING 4096 Kb block: 224.81 Mb/s
FL-POINT & READING 8192 Kb block: 224.81 Mb/s
FL-POINT & READING 16384 Kb block: 224.32 Mb/s

5.RAM Disk:> ramspeed -b 6
RAMspeed (UNIX) v2.3.0 by Rhett M. Hollander (Alasir Enterprises), 2002-04

4Gb per pass mode

FL-POINT Copy: 353.10 Mb/s
FL-POINT Scale: 324.05 Mb/s
FL-POINT Add: 276.88 Mb/s
FL-POINT Triad: 277.01 Mb/s
---
FL-POINT AVERAGE: 307.76 Mb/s

@shadowsun

Quote:
Thanks for the test, indeed it seems there is a misconfiguration on your machine.
Could you please try running the little executable from here:

http://www.sam4x0.com/files/mmucr.lha

and try Crisot test again?

So the problem does not come from my code.

I had many reports of slow L1 write speed from Sam users. If it's just a software configuration problem, maybe it's interesting to write a mini tuto explaining how to configure this.

Losing half the interest of L1 cache is a big bottleneck for the whole system.

@m3x
here's my results after running mmucr before running ragemem...my results appear to be roughly the same so it looks like it's a non flex Sam issue perhaps.

RAGEMEM v0.37 - compiled 11/06/2010

CPU: AMCC PPC440EP 1.3 @ 799 Mhz
Caches Sizes: L1: 32 KB - L2: none - L3: none
Cache Line: 128

---> CPU <---
MAX MIPS: 1598

---> L1 <---
READ32: 2852 MB/Sec
READ64: 5666 MB/Sec
WRITE32: 2862 MB/Sec
WRITE64: 5686 MB/Sec

---> RAM <---
READ32: 298 MB/Sec
READ64: 298 MB/Sec
WRITE32: 165 MB/Sec
WRITE64: 165 MB/Sec
WRITE: 894 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 14 MB/Sec
WRITE: 50 MB/Sec

@328gts

You previous tests did not show any mistake.

This patch in olny for thoses experiencing slow L1 cache speeds on some Sam440 (which is not your case).

@K-L

ah ok i thought m3x mentioned something that would help speed up ram speeds ..

@328gts

The video bus on the original sam really is slow...not much better than classic ;)

My Peg2:

CPU: Motorola MPC 7447/7457 Apollo 1.1 @ 999 Mhz
Caches Sizes: L1: 32 KB - L2: 512 KB - L3: none
Cache Line: 32

---> CPU <---
MAX MIPS: 2998

---> L1 <---
READ32: 3810 MB/Sec
READ64: 7619 MB/Sec
WRITE32: 3265 MB/Sec
WRITE64: 2541 MB/Sec

---> L2 <---
READ32: 1722 MB/Sec
READ64: 2249 MB/Sec
WRITE32: 1551 MB/Sec
WRITE64: 1955 MB/Sec

---> RAM <---
READ32: 146 MB/Sec
READ64: 146 MB/Sec
WRITE32: 215 MB/Sec
WRITE64: 438 MB/Sec
WRITE: 811 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 45 MB/Sec
WRITE: 221 MB/Sec

@m3x

Ragemem before patch :

---> CPU <---
MAX MIPS: 1598

---> L1 <---
READ32: 2839 MB/Sec
READ64: 5641 MB/Sec
WRITE32: 167 MB/Sec
WRITE64: 335 MB/Sec

---> RAM <---
READ32: 277 MB/Sec
READ64: 277 MB/Sec
WRITE32: 166 MB/Sec
WRITE64: 328 MB/Sec
WRITE: 898 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 14 MB/Sec
WRITE: 50 MB/Sec


MMUCR :
OLD SPR[0x03b2]: 0x01490000
NEW SPR[0x03b2]: 0x00010000


Ragemem after patch :

---> CPU <---
MAX MIPS: 1598

---> L1 <---
READ32: 2841 MB/Sec
READ64: 5642 MB/Sec
WRITE32: 2859 MB/Sec
WRITE64: 5679 MB/Sec

---> RAM <---
READ32: 277 MB/Sec
READ64: 277 MB/Sec
WRITE32: 164 MB/Sec
WRITE64: 164 MB/Sec
WRITE: 890 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 14 MB/Sec
WRITE: 50 MB/Sec



So, it seems to be OK for

---> L1 <---
WRITE32: 2859 MB/Sec
WRITE64: 5679 MB/Sec

but note the
---> RAM <---
WRITE64: 164 MB/Sec (before : WRITE64: 328 MB/Sec)

When I switch off, the computer, it resets the problem : is it normal and must I put MMUCR in wbstartup ?

(I've got another problem with my SAM, what's the best way to contact you directly :
info at soft3.net
or on www.acube-systems.biz (contacts) ?)

@shadowsun

Good to see the L1 problem solved, the RAM speed now is in line with other Flex 800 machines (before it was wrong)

You need to re run the patch every time you reset the machine.

A bit more info about the "patch": the problem was seen the first time on the new Sam460, slow L1 access speed. After a lot of research, it turned out the problem was a misconfigured register (MMUCR) which sometimes is set incorrectly.

Then we looked at the 440, to see if the problem was there as well, but we never incountered the problem in our lab, until this test results came out from other users, which showed than some 440 may show the problem as well.

A new 440 kernel (currently in betat testing) set the register correctly, but I don't know when it will be released.

In the meantime, my intention, is to put a little executable on OS4Depot next days.

PS. you can contact me here: mtretene _at_ acube-systems.com

@m3x

Happy to see this small benchmark enlightened a MMUCR problem on 440.

But about the 460, maybe you can explain me the L2 write speed?

---> L2 <---
READ32: 1061 MB/Sec
READ64: 1061 MB/Sec
WRITE32: 521 MB/Sec
WRITE64: 520 MB/Sec

---> RAM <---
READ32: 311 MB/Sec
READ64: 310 MB/Sec
WRITE32: 521 MB/Sec
WRITE64: 521 MB/Sec
WRITE: 1251 MB/Sec (Tricky)

Read speed is acceptable, but write speed is strange. Once again a problem, or just the PPC460 design?

@Crisot

Quote:
Right now, it's a "safe" setting which preserve data coherency when DMA is involved (the L2 controller do buses snooping)
We'll do more tests to see if we can "relax" the setting a bit, or if it's better to keep it.

And here are the results of the austrian jury:

CSPPC, CVPPC, OS4.0

RAGEMEM v0.37 - compiled 11/06/2010

CPU: 604e 9.514 @ 150 Mhz
Caches Sizes: L1: 32 KB - L2: none - L3: none
Cache Line: 32

---> CPU <---
MAX MIPS: 290

---> L1 <---
READ32: 511 MB/Sec
READ64: 1022 MB/Sec
WRITE32: 419 MB/Sec
WRITE64: 792 MB/Sec

---> RAM <---
READ32: 71 MB/Sec
READ64: 76 MB/Sec
WRITE32: 45 MB/Sec
WRITE64: 46 MB/Sec
WRITE: 73 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 5 MB/Sec
WRITE: 17 MB/Sec

Regards
Ruediger

@Crisot
#36
Paul Zeger's A1 and mine are both from the same cradle --CompuQuick-- in Ohio. Mine has no fixes to the buses or USB; ISTR his has them. Anyway, I'm not posting mine. There approximately the same as Paul's at post #43. I did the tests with SetA1 and without SetA1 and didn't see any differences.

@m3x
CPU: AMCC PPC460EX 1.2 @ 1166 Mhz
---> VIDEO BUS <---
READ: 72 MB/Sec
WRITE: 261 MB/Sec
Wow, to those figures! I noticed the PegIIs were 'besting' the A1s, now the SAM460 is looking good & fast!

@Snuffy

Quote:
The built-in video is on a 66Mhz, 32 bit bus. That works out to a maximum of 264MB/second. Sounds like the write speed is hitting the bus limit. Can't do much better than that!

***** Before SAM440ep_setup:

RAGEMEM v0.37 - compiled 11/06/2010

CPU: AMCC PPC440EP 1.3 @ 666 Mhz
Caches Sizes: L1: 32 KB - L2: none - L3: none
Cache Line: 128

---> CPU <---
MAX MIPS: 1327

---> L1 <---
READ32: 2447 MB/Sec
READ64: 4883 MB/Sec
WRITE32: 2456 MB/Sec
WRITE64: 4888 MB/Sec

---> RAM <---
READ32: 276 MB/Sec
READ64: 297 MB/Sec
WRITE32: 164 MB/Sec
WRITE64: 164 MB/Sec
WRITE: 887 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 14 MB/Sec
WRITE: 49 MB/Sec

***** After SAM440ep_setup:

RAGEMEM v0.37 - compiled 11/06/2010

CPU: AMCC PPC440EP 1.3 @ 666 Mhz
Caches Sizes: L1: 32 KB - L2: none - L3: none
Cache Line: 128

---> CPU <---
MAX MIPS: 1327

---> L1 <---
READ32: 2446 MB/Sec
READ64: 4885 MB/Sec
WRITE32: 2455 MB/Sec
WRITE64: 4890 MB/Sec

---> RAM <---
READ32: 276 MB/Sec
READ64: 297 MB/Sec
WRITE32: 164 MB/Sec
WRITE64: 164 MB/Sec
WRITE: 887 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 31 MB/Sec
WRITE: 49 MB/Sec

Indeed, doubled write performance on the V-Bus - thanks, Max!

Ok a before and after comparison

Before installing Sam440 Patch
Quote:

After installing Sam440 Patch

Quote: