Optimizing my 68060 machine

[NovaCoder]

Originally Posted by trixster

Some 060 bustest figures

6) Blizzard 1260 80mhz (ram 80mhz) = 57.6MB/s read, 38.5MB/s write
7) Matze’s yet to be released 68060-TK 100mhz (ram 100mhz) = 87.8MB/ readm, 79.8MB/s writem

Matze card looks like a monster, should be very fast

[AmigaHope]

There was an add-on board for the A3640 called XCalibur that added a local RAM bus. I wish someone would reverse-engineer that to make a clone. It would instantly turn the A3640 (and clones) into much better solutions.

A really smart person could even integrate it into the A3660 design.

[Chucky]

the XCalibur yes. but too many drawbacks. as it doesn't work with DMA (so forget A3000 and SCSI) etc.
I had plans for a similiar thing. but dropped it due to lack of DMA. would be too much nagging about that.

[pipper]

Does it HAVE to be DMA? The mainboard ram would still be dma-ble, right? And the scsi controller is almost the only thing that can use DMA.

[Chucky]

nah it does not HAVE to. but. that was one big thing and. then why even bother. the 3640 design is so crappy anyway..

[RDP]

WarpEngine 040@40

readm: 50.1
writem: 28.1


Is your Kickstart rom mapped to your fastmem?


RDP

[AmigaHope]

Originally Posted by Chucky

the XCalibur yes. but too many drawbacks. as it doesn't work with DMA (so forget A3000 and SCSI) etc.
I had plans for a similiar thing. but dropped it due to lack of DMA. would be too much nagging about that.

As long as you can set your DMA to be allocated from mainboard RAM (you can do this with SCSI drives using the MASK mountlist setting, masking out any address bits higher than mainboard -- accelerator memory sits above mainboard, starting at the upper 5 bits) you should still be okay. Don't know how Deneb allocates, and I think that's the only other Z3 DMA device.

You could even silkscreen this info on the PCB so there's no way people could not notice. You could include a jumper or softboot mode that either disabled accelerator RAM or set it to lower priority so that you could boot into a safe mode to adjust MASK settings.

A3640/A3660 is crappy but it's also the only board we have that doesn't cost a fortune. Nobody's making new cards except the A3660 and the A3640 is the only widely available one that doesn't command a huge premium on ebay.

[grelbfarlk]

Originally Posted by AmigaHope

As long as you can set your DMA to be allocated from mainboard RAM (you can do this with SCSI drives using the MASK mountlist setting, masking out any address bits higher than mainboard -- accelerator memory sits above mainboard, starting at the upper 5 bits) you should still be okay. Don't know how Deneb allocates, and I think that's the only other Z3 DMA device.

You could even silkscreen this info on the PCB so there's no way people could not notice.

Zorro 3 DMA devices:
Deneb
Fastlane
A4091
Zorram

Non Zorro 3 DMA devices:
SCSI controllers onboard the Accelerator- GVP Trexx-II, CSMK 2, 3, PPC, WarpEngine, etc
We really liked DMA SCSI controllers because we could stream video from the HDD without wasting CPU time on it, but it was still possible to output to a non DMA RTG card.

The question is what do you really need DMA for?
Stuff like the Elbox FastATA is a pretty fast HDD controller but it doesn't do DMA. PIO is fine-ish as long as it's fast. If the CPU has fast access to the RAM do we really care whether it's DMA? But if the CPU isn't going to have fast access to the RAM because of no DMA, then yeah it's pretty pointless.

I'm having a hard time picturing the situations where having fast RAM that isn't DMA is going to actually make things worse.
Will a Z3 RTG card fall on its face if it can't access CPU DMA RAM? Mostly all of the traffic is the other way from CPU RAM to RTG GFX RAM anyway, and RTG GFX RAM is almost never DMA. Exceptions being like ... CSPPC+CVPPC or BPPC+BVPPC(I have doubts about these), GVP 030 with EGS 110/24 (maybe?), CSPPC+GREX, uh that's about it.

[DarrenHD]

Originally Posted by grelbfarlk

Zorro 3 DMA devices:
Deneb
Fastlane
A4091
Zorram

I am fairly certain ZorRAM is not using Zorro 3 DMA.

In my experience any Zorro-3 added ram is dog slow, and that includes gfx card memory added (eg. with Elbox Mediator or ZorRAM). We tested this extensively on OS4-Classic to try to use it as "main" ram, and it made the system too slow to be useable. So we disabled that feature and ended up using it only as SWAP memory. On OS3 it's not as bad...but OS4 is rather "heavy" so it made a huge impact.

Darren

[grelbfarlk]

Originally Posted by DarrenHD

I am fairly certain ZorRAM is not using Zorro 3 DMA.

In my experience any Zorro-3 added ram is dog slow, and that includes gfx card memory added (eg. with Elbox Mediator or ZorRAM). We tested this extensively on OS4-Classic to try to use it as "main" ram, and it made the system too slow to be useable. So we disabled that feature and ended up using it only as SWAP memory. On OS3 it's not as bad...but OS4 is rather "heavy" so it made a huge impact.

Darren

Elbox Mediator is not 68k DMA, only possibly when communicating between PCI devices.

Regardless of what you have on the Zorro-3 bus, you're looking at something like 12MB/s on a good day, compared to >50MB/s on CSPPC RAM. With OS4, I thought the distinction between DMA and non-DMA RAM was that you could actually use DMA RAM as system RAM, while non-DMA could only be used as swap space.

I thought between the Zorram and the Bigramplus one was DMA and one was not, I thought it was the E3B board as they were the only company which produced a DMA Zorro card since the 90s. Amiga.resource.cx says both of them are.

[alexh]

Zorro III DMA devices are all DMA masters. They initiate the DMA transfer. Zorro III RAM cards are all slaves, they don't initiate DMA transfers. So all Zorro III RAM cards are NOT classified as "Zorro III DMA" just "Zorro III"

I think Amiga resource is wrong when it lists ZoRAM and BigRAM plus as "supports DMA (if Buster 11 is installed)". How does a RAM card initiate a DMA transfer?

And why wouldn't Zorro III DMA devices work with Buster 09? (They support DMA)

[grelbfarlk]

Originally Posted by alexh

Zorro III DMA devices are all DMA masters. They initiate the DMA transfer. Zorro III RAM cards are all slaves, they don't initiate DMA transfers. So all Zorro III RAM cards are NOT classified as "Zorro III DMA" just "Zorro III"

I think Amiga resource is wrong when it lists ZoRAM and BigRAM plus as "supports DMA (if Buster 11 is installed)". How does a RAM card initiate a DMA transfer?

And why wouldn't Zorro III DMA devices work with Buster 09? (They support DMA)

So is that the reason that the X-Calibur doesn't work with Z3 DMA devices, because it acts as a Z3 DMA Master?

Buster-11 has fixed DMA master, which allows a single DMA master. So you could have a total of one Z3 DMA device.

Buster-9 allowed multiple DMA masters, but was buggy. So the advice was if you used Z3 cards, upgrade to Buster 11. Some cards had enough workarounds to deal with the Buster-9 like the Fastlane and the Deneb.

Seems like there's a missing part to this. X-calibur didn't work with a system that had a Z3 DMA master like A4000T onboard SCSI or an A4091. But I don't see where A3000 owners said that onboard SCSI stopped working, probably because you could not fit an X-Calibur on an A3640 and install the drive cage.

[Chucky]

the reason the X-Calibur doesn't handle DMA is simple: the 3640 logics does not handle DMA "inside" the 040 part.

a 040 card is basically a 040->030 adapter.. if you have DMA you must have a 030->040 adapter aswell.
(very simplified) this logic simply does not exist.

[Chucky]

and the ram on the x-calibur is "inside" the 040 part.

[grelbfarlk]

Originally Posted by Chucky

the reason the X-Calibur doesn't handle DMA is simple: the 3640 logics does not handle DMA "inside" the 040 part.

a 040 card is basically a 040->030 adapter.. if you have DMA you must have a 030->040 adapter aswell.
(very simplified) this logic simply does not exist.

Right I didn't think think of it from the view of a DMA device, makes sense.
So for the three useful boards that support DMA, A4091, FastLane and Deneb, two of which would still work as they support PIO mode (Fastlane and Deneb.) Still seems like it would be worth doing.

                                 EmuTest Sieve Dhrystone  Sort    Matrix  IMath MemTest InstTest
X-Calibur (040/25, 3.0)          10.29    5.20    9.31    8.08    6.45    4.90    4.85    9.01
X-Calibur (040/25, 3.0 in RAM)   10.29    5.20    9.31    8.08    6.45    4.90    4.84    8.98
A3640 (040/25, 3.0)              8.66     2.70    9.34    7.31    4.40    4.86    1.27    5.73
A3640 (040/25, 3.0)              9.02     2.76    9.40    7.38    4.46    4.95    1.29    5.99
G-Force (040/40, 3.0)            12.67    7.76    14.90   12.74   9.81    7.84    5.85    8.04
WarpEngine (040/28, 3.0)         11.71    5.57    10.67   9.22    7.10    5.62    4.06    9.87
WarpEngine (040/28, 3.0 in RAM)  11.71    5.56    10.67   9.21    7.10    5.62    4.19    9.88

[trixster]

Losing dma on a Deneb is no great loss

[trixster]

some more results added to bustest in post #11

[grelbfarlk]

Trixster you could add the Quikpak 060 results:

BusSpeedTest 0.19 (mlelstv) Buffer: 262144 Bytes, Alignment: 32768 (bustest fast rom)
Memtype 	Cycle   Bandwidth
fast 	readw 	44.8 ns 	44.6 meg/sec
fast 	readl 	71.1ns 	56.6 meg/sec
fast 	readm 	73.1ns 	54.7 meg/sec
fast 	writew 	44.8ns 	44.7 meg/sec
fast 	writel 	89.8ns 	44.6 meg/sec
fast 	writem 	89.0ns 	45.0 meg/sec
rom 	readw 	44.8ns 	44.7 meg/sec
rom 	readl 	69.9ns 	57.2 meg/sec
rom 	readm 	73.0ns 	54.8 meg/sec

Memspeed Benchmarks (Memspeed Copyright 1996 Frank Wille)

Ram type: 60ns Fast Page Mode (FPM) vs. 60ns EDO Ram (EDO)
Type 	FPM Read 	EDO Read 	% Change
Chip Ram 	4617 kB/s 	4617 kB/s 	0.0%
Fast Ram 	46492 kB/s 	54545 kB/s 	+17.3%
Rom 	        46586 kB/s 	54545 kB/s 	+17.1%
Type 	        FPM Write 	EDO Write 	% Change
Chip Ram 	6969 kB/s 	6968 kB/s 	-0.01%
Fast Ram 	38927 kB/s 	44055 kB/s 	+13.2%
Type 	        FPM             Copy EDO Copy 	% Change
Chip Ram 	2777 kB/s 	2777 kB/s 	0.0%
Fast Ram 	22470 kB/s 	26716 kB/s 	+18.9%

From:http://www.hd-zone.com/2016/08/04/qu...0-information/

[trixster]

I will, Sir. That looks to be a rapid card.

I’ll probably make a new thread, and add a fancy table like you have.

[AmigaHope]

Originally Posted by Chucky

the reason the X-Calibur doesn't handle DMA is simple: the 3640 logics does not handle DMA "inside" the 040 part.

a 040 card is basically a 040->030 adapter.. if you have DMA you must have a 030->040 adapter aswell.
(very simplified) this logic simply does not exist.

That doesn't mean you can't use DMA at all though, just that you can't DMA to the X-Calibur's RAM. DMA to motherboard RAM should still work fine. You just need to have a way to tell the card's driver to not try to use X-Calibur RAM.

For SCSI DMA devices like the A4091, this is accomplished with the MASK setting in the mount entries. As long as the driver honors the settings, this should be okay.

For the Deneb you may have to tweak the driver a bit to make sure it only tries to use motherboard RAM. I'm not sure how difficult this would be. A cheesy way out could be to patch in a different AllocMem library function that acts as a wrapper for the "real" AllocMem that deliberately avoids memory pools using the upper 5 bits, and then patch the Deneb driver to use that AllocMem function.