English Amiga Board Amiga Lore


Go Back   English Amiga Board > Support > support.Hardware > Hardware mods

 
 
Thread Tools
Old 29 September 2017, 19:12   #1981
plasmab
Registered User

plasmab's Avatar
 
Join Date: Sep 2016
Location: Glasgow
Posts: 1,412
Quote:
Originally Posted by nogginthenog View Post
Gayle IDE is quite CPU intensive. Most SCSI cards should support DMA.


DMA isn’t really an advantage. It hogs the bus just like CPU PIO does. Unless you have dual port RAM that is.

EDIT: I suppose i should qualify this a bit. Its an advantage if your CPU speed sucks. If your CPU goes like poop off a stick its not massively advantageous unless its into uncontended memory.


Sent from my iPhone using Tapatalk

Last edited by plasmab; 29 September 2017 at 20:12.
plasmab is offline  
AdSense AdSense  
Old 08 October 2017, 02:06   #1982
patrik
Registered User
patrik's Avatar
 
Join Date: Jan 2005
Location: Umeň
Age: 37
Posts: 316
Dma doesn't really help with transfer rates, unless it is faster at transferring the data than the cpu is copying the data. Which I guess is a possibility for a 68000.

It will however help with the overall system performance for cases where the cpu is busy with other things during the data transfer:
- Like general multitasking performance, when there are more than one process running, so the even if one process is blocking on an I/O request taking place, the cpu is still busy.
- Another example would be if you have a single process doing asynchronous I/O and is working on data, while a request for more data is executing in the background.

Of course this depends on the dma transfer not starving the cpu of memory cycles completely, so bigger cpu caches helps. As long as it starves the cpu less time than time the cpu would have to spend doing the transfer, it is a win. Also another factor to consider is that the dma controller in theory will waste less cycles than the cpu to do a transfer as it just has to read or write the data on the bus, while the cpu has to copy the data, so it has to both read and write.

Do not forget that the dma controllers have to be able to access all of the systems memory to be able to work effectively. For example a Zorro2 dma controller in a system with fastram outside the 24-bit Zorro2-space does not help, it will rather slow things down as it will only be able to dma to/from chipmem and then the cpu has to copy from chipmem to wherever the process requesting the transfer allocated its buffer. Even if the system had placed some memory in the 24-bit Zorro2-space to allow transfers, there is nothing saying that the buffer allocated by a process resides in that memory, so in most cases, the cpu will still end up copying the data.

Examples of usable dma setups would be the classic A2091 + A2630 (all 32-bit mem in 24-bit space), A3000-scsi, Zorro3-Amiga with A4091 or Phase5 Fastlane, all Phase5 accelerators with scsi controller, WarpEngine etc (all system ram accessable by the scsi dma controller).

Last edited by patrik; 08 October 2017 at 19:38. Reason: Forgot the cpu has to copy the data
patrik is offline  
Old 10 October 2017, 17:56   #1983
Black68
Registered User

 
Join Date: Jul 2017
Location: Spain
Posts: 32
Quote:
Originally Posted by dalek View Post
@8 Bit Dreams: have you tried adding the pullup resistors to the lower address lines of the Rev 6A board? It solves issues with eproms and kickstart 3.1.

They are:
RP106 and RP107
9 x 4k7 in SIP-10

More details here: http://eab.abime.net/showpost.php?p=...&postcount=322
Adding this resistors fixed the problems I had with the PSU!

I was testing with different 3 supplies: one A600 PSU, one A500 PSU and one ATX PSU. The TF530 on both my rev.6a boards would only boot with the A600 PSU (3.0A).

Now, with RP106 and RP107 in place I can use any of the power supplies without problem!
Black68 is offline  
Old 10 October 2017, 19:54   #1984
8 Bit Dreams
Retro maniac

8 Bit Dreams's Avatar
 
Join Date: Feb 2017
Location: near Munich / Germany
Posts: 84
[QUOTE=Black68;1191173]Adding this resistors fixed the problems I had with the PSU!

I was testing with different 3 supplies: one A600 PSU, one A500 PSU and one ATX PSU. The TF530 on both my rev.6a boards would only boot with the A600 PSU (3.0A).

That's good news!
8 Bit Dreams is offline  
Old 10 October 2017, 19:55   #1985
8 Bit Dreams
Retro maniac

8 Bit Dreams's Avatar
 
Join Date: Feb 2017
Location: near Munich / Germany
Posts: 84
[QUOTE=8 Bit Dreams;1191195]
Quote:
Originally Posted by Black68 View Post
Adding this resistors fixed the problems I had with the PSU!

I was testing with different 3 supplies: one A600 PSU, one A500 PSU and one ATX PSU. The TF530 on both my rev.6a boards would only boot with the A600 PSU (3.0A).
That's good news!
8 Bit Dreams is offline  
Old 11 October 2017, 11:32   #1986
Black68
Registered User

 
Join Date: Jul 2017
Location: Spain
Posts: 32
[QUOTE=8 Bit Dreams;1191196]
Quote:
Originally Posted by 8 Bit Dreams View Post
That's good news!
Are you running your TF530 at 50mhz?

I got mine running at 48mhz very stable but without FPU. The 68882 I have are rated for 40mhz and it makes the system unstable...
Black68 is offline  
Old 11 October 2017, 14:13   #1987
plasmab
Registered User

plasmab's Avatar
 
Join Date: Sep 2016
Location: Glasgow
Posts: 1,412
I had mine running at 48mhz with FPU nicely


Sent from my iPhone using Tapatalk
plasmab is offline  
Old 11 October 2017, 16:25   #1988
Black68
Registered User

 
Join Date: Jul 2017
Location: Spain
Posts: 32
Quote:
Originally Posted by plasmab View Post
I had mine running at 48mhz with FPU nicely
I have a couple of MC68882FN40A

What FPUs are you using?
Black68 is offline  
Old 11 October 2017, 19:00   #1989
plasmab
Registered User

plasmab's Avatar
 
Join Date: Sep 2016
Location: Glasgow
Posts: 1,412
Quote:
Originally Posted by Black68 View Post
I have a couple of MC68882FN40A



What FPUs are you using?


I think it was 33s will check the mask


Sent from my iPhone using Tapatalk
plasmab is offline  
Old 11 October 2017, 20:22   #1990
8 Bit Dreams
Retro maniac

8 Bit Dreams's Avatar
 
Join Date: Feb 2017
Location: near Munich / Germany
Posts: 84
Quote:
Originally Posted by Black68 View Post

Are you running your TF530 at 50mhz?

I got mine running at 48mhz very stable but without FPU. The 68882 I have are rated for 40mhz and it makes the system unstable...
Indeed, all my TF cards running at 50Mhz (with both CPU types, plastic and ceramic) with FPU installed (MC68882FN40A) no crashes, no issues, BUT on beta 50Mhz firmware only, WHDLoad doesn't work on the newest firmware for me. It may be R3 on TF card, that causes error, value on github (BOM) shows 47k on R3, eagle schematics shows 10k, someone told me correct value should be 1k, so i am puzzled a bit, my experience is: if You install so high resistance - You will possibly end up with yellow screen even without FPU installed. have swapped R3 to 1k and now it works.
have asked Steve about correct value, but got no answer on this question..
Your problem can be also CPU itself, that refuses to run at this speed, what kind of CPU do You use?
8 Bit Dreams is offline  
Old 11 October 2017, 20:33   #1991
alenppc
Registered User
 
Join Date: Apr 2012
Location: Canada
Age: 37
Posts: 627
At 25 MHz it's most stable with R2/R3 at 47k and R1 at 1k. (Both rev 2 and rev 3).

At higher frequencies these values don't work however... I am still trying to figure out which ones to use. The only certainty is R1 always at 1K.
alenppc is offline  
Old 11 October 2017, 21:25   #1992
8 Bit Dreams
Retro maniac

8 Bit Dreams's Avatar
 
Join Date: Feb 2017
Location: near Munich / Germany
Posts: 84
Quote:
Originally Posted by alenppc View Post
At 25 MHz it's most stable with R2/R3 at 47k and R1 at 1k. (Both rev 2 and rev 3).

At higher frequencies these values don't work however... I am still trying to figure out which ones to use. The only certainty is R1 always at 1K.
i can speak from my experience:
have bought Xilinx chips from four different sources
One source are 7ns - these cpld's doesen't work correctly
two batches are working with 47K @R3
last batch had a yellow screen with 47K @R3, swapping out to 1K has solved this issue, also floppy drive is working (read & format)

will try to swap R3 to 10K and program newest firmware, maybe it will work then..
8 Bit Dreams is offline  
Old 12 October 2017, 04:07   #1993
Black68
Registered User

 
Join Date: Jul 2017
Location: Spain
Posts: 32
Quote:
Originally Posted by 8 Bit Dreams View Post
Your problem can be also CPU itself, that refuses to run at this speed, what kind of CPU do You use?
Im using the latest firmware, TF530 Rev 2 and Amiga Rev 6a boards. I have a few 33mhz plastic cpu and a 50mhz ceramic (the 50 MHz speed is exclusive to the ceramic PGA package, the plastic '030 stopped at 40 MHz).

After adding the pullup resistors to the Amiga main board, the TF530 at 33mhz runs flawless (with FPU).

It refuses to run at 50mhz but it will run perfect at 48mhz (without FPU).
Black68 is offline  
Old 12 October 2017, 13:36   #1994
alenppc
Registered User
 
Join Date: Apr 2012
Location: Canada
Age: 37
Posts: 627
The best I could do was 36 Mhz but I haven't played around with resistors.
alenppc is offline  
Old 13 October 2017, 00:55   #1995
patrik
Registered User
patrik's Avatar
 
Join Date: Jan 2005
Location: Umeň
Age: 37
Posts: 316
Sorry for being quite OT, but on the DMA topic I thought I should present an example where DMA makes a difference even on a computer with a rather hefty cpu, instead of just babbling.

A case which would be beneficial for a DMA controller is to do something which causes a high load on the cpu and does disk I/O with at the same time. Cpu time wasted on disk I/O will then be noticed.

I chose to do something quite common which can do that - download a file over the local network to disk. The data transfer through the TCP/IP-stack takes care of causing the high cpu load and for doing the disk I/O at the same time I used aget, which downloads files over http and does asynchronous writes (lets the file system handle writes in the background, while downloading the next batch of data in the foreground).

The idea was to do it on both a PIO controller and a DMA controller and compare the results. I used an A1200 which has both the standard IDE PIO controller and a Blizzard SCSI DMA controller (a working DMA setup where the controller can transfer to/from all the memory of the machine).

To make the IDE and SCSI as even as possible in transfer rate, I used UnitControl to set the SCSI to operate in 2MHz synchronous mode, which resulted in ~2.5MB/sec transfer rate, compared to the IDE transfer rate of ~2.4MB/sec. This does not matter though as the writes are done asynchronously, so even at 10MB/sec transfer rate, the SCSI would not gain anything, but more on that later.

The relevant hardware used is:
A1200
Blizzard 1260@50Mhz
Blizzard SCSI-kit IV
mSATA SSD connected to the internal IDE with mSATA->IDE adapter.
mSATA SSD connected to the SCSI-kit with mSATA->IDE adapter and Acard IDE->SCSI adapter.
Netgear FA411 100Mbit PCMCIA ethernet card

The relevant software used is:
AmigaOS 3.1
BlizKick 1.24 for kickstart in fast and with the usual modules like LocalFast and MoveVBR to improve multitasking performance
1230scsi.device 8.5
UnitControl 2.17
PFS AIO 2.3
Roadshow 1.13 (020 version)
cnet16.device 1.9
aget 0.29

Results:
Code:
9.Ram Disk:> aget 192.168.0.1/test10M.bin RAM:
Buffer size: 128kB
Sent HTTP GET request!
Done parsing HTTP header
Server response: 200 OK
Receiving file: 0% - [oooooooooooooooooooooooooooooooo] - 100% (10MB)
10MB received in 9.43s - 1.06MB/s

9.Ram Disk:> aget 192.168.0.1/test10M.bin IDE-PFS:
Buffer size: 128kB
Sent HTTP GET request!
Done parsing HTTP header
Server response: 200 OK
Receiving file: 0% - [oooooooooooooooooooooooooooooooo] - 100% (10MB)
10MB received in 12.77s - 801.49kB/s

9.Ram Disk:> aget 192.168.0.1/test10M.bin SCSI-PFS:
Buffer size: 128kB
Sent HTTP GET request!
Done parsing HTTP header
Server response: 200 OK
Receiving file: 0% - [oooooooooooooooooooooooooooooooo] - 100% (10MB)
10MB received in 8.65s - 1.16MB/s
Comparing the results for the IDE PIO controller (IDE-PFS: ) to the SCSI DMA controller (SCSI-PFS: ), the DMA controller scored a 48% higher download speed (1.16MB = 1187.84kB).

Also note that it is actually slower downloading to RAM: than to disk using the DMA controller. Thats because when writes are done asyncronously, there is no benefit that the RAM-handler finishes writes much faster than the disk (given the disk is faster than the network), but the fact that the RAM-handler uses quite some cpu time to copy data around steals cpu time - just like a PIO disk controller.

Last edited by patrik; 13 October 2017 at 00:59. Reason: removed redundant percent after %
patrik is offline  
Old 13 October 2017, 01:08   #1996
plasmab
Registered User

plasmab's Avatar
 
Join Date: Sep 2016
Location: Glasgow
Posts: 1,412
Quote:
Originally Posted by patrik View Post
Also note that it is actually slower downloading to RAM: than to disk using the DMA controller. Thats because when writes are done asyncronously, there is no benefit that the RAM-handler finishes writes much faster than the disk (given the disk is faster than the network), but the fact that the RAM-handler uses quite some cpu time to copy data around steals cpu time - just like a PIO disk controller.
Couple of things to try here.

Turn on CPU FASTROM and retry this.

With the ROM in cacheable FASTRAM it means the drivers run entirely inside cache..

I agree you can get better out and out disk performance but the CPU is stalled anyway if the ram is contended so you dont really get better system performance.

Actually surprised at how slow your RAM access is.
plasmab is offline  
Old 13 October 2017, 02:57   #1997
patrik
Registered User
patrik's Avatar
 
Join Date: Jan 2005
Location: Umeň
Age: 37
Posts: 316
Quote:
Originally Posted by plasmab View Post
Couple of things to try here.

Turn on CPU FASTROM and retry this.

With the ROM in cacheable FASTRAM it means the drivers run entirely inside cache..

I agree you can get better out and out disk performance but the CPU is stalled anyway if the ram is contended so you dont really get better system performance.

Actually surprised at how slow your RAM access is.
ROM is in fast, execbase is in fast, VBR is in fast via BlizKick.

The Blizzard 1260 cannot be accused for having very fast memory. However, the ram-handler is quite a cpu hog in 3.1 atleast. If you LoadModule the ram-handler from 3.9, you will get better figures for RAM:.

Blizzard1260 ram speed plus verification that rom is in fast:
Code:
9.Ram Disk:> bustest fast rom
BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $60310000  readw      66.5 ns   normal      30.1 * 10^6 byte/s
fast      $60310000  readl     112.1 ns   normal      35.7 * 10^6 byte/s
fast      $60310000  readm     111.1 ns   normal      36.0 * 10^6 byte/s
fast      $60310000  writew     81.4 ns   normal      24.6 * 10^6 byte/s
fast      $60310000  writel    163.8 ns   normal      24.4 * 10^6 byte/s
fast      $60310000  writem    163.7 ns   normal      24.4 * 10^6 byte/s
rom       $00F80000  readw      66.4 ns   normal      30.1 * 10^6 byte/s
rom       $00F80000  readl     111.0 ns   normal      36.0 * 10^6 byte/s
rom       $00F80000  readm     111.3 ns   normal      35.9 * 10^6 byte/s
And yes, the dma controller will steal memory cycles from the cpu. But every cycle used by the dma controller where the cpu is busy internally with an instruction or running from cache is a win.

Not implying that dma controllers helps with raw disk transfer speed, just wanted to show that they can be fun when the system is under load .

Last edited by patrik; 13 October 2017 at 02:59. Reason: controller, not transfer
patrik is offline  
Old 14 October 2017, 23:41   #1998
Black68
Registered User

 
Join Date: Jul 2017
Location: Spain
Posts: 32
Quote:
Originally Posted by alenppc View Post
The best I could do was 36 Mhz but I haven't played around with resistors.
On what Rev of the Amiga mainboard are you running the TF530?
Black68 is offline  
Old 15 October 2017, 01:11   #1999
alenppc
Registered User
 
Join Date: Apr 2012
Location: Canada
Age: 37
Posts: 627
Quote:
Originally Posted by Black68 View Post
On what Rev of the Amiga mainboard are you running the TF530?
NTSC 6A, that's the only one I have.
alenppc is offline  
Old 15 October 2017, 08:40   #2000
8 Bit Dreams
Retro maniac

8 Bit Dreams's Avatar
 
Join Date: Feb 2017
Location: near Munich / Germany
Posts: 84
Quote:
Originally Posted by alenppc View Post
NTSC 6A, that's the only one I have.
Do You have pullups installed on this board?
8 Bit Dreams is offline  
AdSense AdSense  
 


Currently Active Users Viewing This Thread: 2 (0 members and 2 guests)
 
Thread Tools

Similar Threads
Thread Thread Starter Forum Replies Last Post
Listening to Amiga music with too much treble makes it sound terrible! Foebane Amiga scene 4 08 October 2016 06:30
Terrible usb drive performance? lovinggames support.FS-UAE 5 11 March 2015 09:06
Broken MASPlayer - Terrible hissing. SS454 support.Hardware 8 17 November 2013 01:41
Accelerators pmc MarketPlace 9 04 November 2009 20:23
Problem with a terrible, simply effect! Camionsauro Coders. General 13 19 April 2009 08:53

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +2. The time now is 07:35.


Powered by vBulletin® Version 3.8.8 Beta 1
Copyright ©2000 - 2017, vBulletin Solutions, Inc.
Page generated in 0.26971 seconds with 15 queries