Blitter version test

[ijor]

As many already know, there are several version of the Blitter chip by several manufacturers.

I found that there seems to be a slightly different behavior, which seems to depend on the version of Blitter. But we are not yet sure, we need to test many machines to reach a more reliable conclusion.

Attached on this post it's a very small program. The program performs a simple Blitter test and outputs an hex number on the screen. Please test it and post your result. It would help a lot if you could include the exact version of Blitter you have, the type of machine (STE or MegaST). And any other relevant modifications that conceivable might make a difference, like a CMOS CPU.

NOTE: For STE and MegaST with Blitter only. Won't run correctly on the MegaSTE.

Thanks,

[Arne]

Sent you a picture of my strange 1040STE-Blitter.
I found another, very similar one in my IC drawer. It's an ST, too. Just three weeks younger (8945) and the Z-Number on the bottom right is different (Z927190). I am very sure that I removed it from a MegaST and not a STE.
I do have two MM9092V (NS?) and two C101643-31 with different printing on top and a C101643-011 (1987).
Do you expect different results of the same Blitter if put in a MST or STE?

[MisterUndo]

As many already know, there are several version of the Blitter chip by several manufacturers.

I found that there seems to be a slightly different behavior, which seems to depend on the version of Blitter. But we are not yet sure, we need to test many machines to reach a more reliable conclusion.

Attached on this post it's a very small program. The program performs a simple Blitter test and outputs an hex number on the screen. Please test it and post your result. It would help a lot if you could include the exact version of Blitter you have, the type of machine (STE or MegaST). And any other relevant modifications that conceivable might make a difference, like a CMOS CPU.

NOTE: For STE and MegaST with Blitter only. Won't run correctly on the MegaSTE.

Thanks,

[czietz]

Well, your test program gives different results on the two machines I tested:

1. 1040STfm (Blitter retrofitted), National Semiconductor Blitter, datecode 8824, test result 0040.
2. 1040STfm (Blitter retrofitted), ST Blitter, datecode 9018, test result 003F.

My 1040STE is in storage right now. If you tell me more about what this program is testing, I might be tempted to get the STE out of storage.

[TheNameOfTheGame]

Sent you a picture of my STE's blitter. My test returned number 0040.

[czietz]

Sent you a picture of my strange 1040STE-Blitter.

I am interested in that picture, too.

[Arne]

Sent you a picture of my strange 1040STE-Blitter.

I am interested in that picture, too.

Probably the same as your No.2 setting. It a ST brand one. I am sure ijor will clarify things.

[ijor]

Do you expect different results of the same Blitter if put in a MST or STE?

I don't, at least not for this specific test. Seems the result depends only, or at least mostly, on the Blitter version.

2. 1040STfm (Blitter retrofitted), ST Blitter, datecode 9018, test result 003F.

By ST you mean the Signetics/Thompsonlogo? C101643-011?
Yes, that would be the same as Arne's one that got the same result. So it seems to confirm this version of Blitter is slightly different.

I'll elaborate a small technical description in the next message ...

[czietz]

By ST you mean the Signetics/Thompsonlogo? C101643-011?

"ST" as in "SGS-Thomson" (not Signetics). But, yes, C101643-011. Photo is here: https://www.chzsoft.de/asic-web/images/header.jpg (as it's used as banner in my article about recovering the ASIC schematics).

[ijor]

The program simply counts the number of words blitted with a single non-hog run.

In theory, if non-hog mode is configured, Blitter is supposed to perform 64 bus cycles and then yield to the CPU. But as I commented already some time ago, Blitter has a small buglet. Blitter starts counting bus cycles, any bus cycles, as soon as it requests the bus. But "normally" the CPU would still run one bus cycle for its own before granting the bus to Blitter. And Blitter count that bus cycle as it was its own. So Blitters performs 63 bus cycles only, and not 64.

But with special code it is possible to alter slightly the bus acquisition timing so that the CPU might not run a bus cycle before granting the bus. Then Blitter would perform the 64 bus cycles. This is performed by writing to Blitter when setting busy with a special code sequence that keeps the bus idle for a couple of clock cycles. The exact timing also depends, or at least it seem so, on the Blitter version.

That's why we see $40 (64 in decimal) or $3F (63 decimal).

[ijor]

"ST" as in "SGS-Thomson" (not Signetics). But, yes, C101643-011. Photo is here: https://www.chzsoft.de/asic-web/images/header.jpg (as it's used as banner in my article about recovering the ASIC schematics).

I stand corrected.

We probably commented about this already, but I forgot. Do you have some info or hints about the history of the Blittter versions? Which version is the older and which one is the newer?

[Arne]

That's why we see $40 (64 in decimal) or $3F (63 decimal).

Expect to see much more strange values. No kidding!
I will put all my results in a PDF. Takes 5min. Comes by eMail.

[Cyprian]

The program simply counts the number of words blitted with a single non-hog run.

In theory, if non-hog mode is configured, Blitter is supposed to perform 64 bus cycles and then yield to the CPU. But as I commented already some time ago, Blitter has a small buglet. Blitter starts counting bus cycles, any bus cycles, as soon as it requests the bus. But "normally" the CPU would still run one bus cycle for its own before granting the bus to Blitter. And Blitter count that bus cycle as it was its own. So Blitters performs 63 bus cycles only, and not 64.

from my tests from 2007 it was 63 BLiTTER bus cycles plus 2 for bus mastering and 64 for the CPU

But with special code it is possible to alter slightly the bus acquisition timing so that the CPU might not run a bus cycle before granting the bus. Then Blitter would perform the 64 bus cycles. This is performed by writing to Blitter when setting busy with a special code sequence that keeps the bus idle for a couple of clock cycles. The exact timing also depends, or at least it seem so, on the Blitter version.

That's why we see $40 (64 in decimal) or $3F (63 decimal).

do you mean this one https://www.atari-forum.com/viewtopic.p ... 197#p96197 ?


I will check my STE and Mega ST with your test later.

[czietz]

We probably commented about this already, but I forgot. Do you have some info or hints about the history of the Blittter versions? Which version is the older and which one is the newer?

Unfortunately nothing reliable. I mean, we know that there are three versions in the schematics. But I don't know which versions made it into chips by the different manufacturers. Sorry.

[Arne]

Finished my tests on a 4MB 1040STE TOS 2.06 machine and some Blitter gave a different result than $003F/$0040. But all my loose Blitters in the IC drawer come from MegaSTs. Is it granted that these Blitter will do their job in a 1040STE as expected?
All I can say is that I tried one MST Blitter with the Atari diagnostic kit and performed both [G] and [Y] tests and first one threw a G1 error while the latter did not perform the drawing of a rectangle "ray by ray".

[ijor]

from my tests from 2007 it was 63 BLiTTER bus cycles plus 2 for bus mastering and 64 for the CPU

The exact timing depends on several factors, even on the specific version of Blitter, as you can see. But yes, usually Blitter steals 65*4 clock cycles from the CPU and performs 63 bus cycles.

do you mean this one https://www.atari-forum.com/viewtopic.p ... 197#p96197 ?

Not exactly sure what you are asking. But no, there is no direct relation with that thread, except, of course, that it is also about Blitter timing.

[ijor]

We probably commented about this already, but I forgot. Do you have some info or hints about the history of the Blittter versions? Which version is the older and which one is the newer?

Unfortunately nothing reliable. I mean, we know that there are three versions in the schematics. But I don't know which versions made it into chips by the different manufacturers. Sorry.

Sorry, I think I wasn't clear enough. I wasn't asking about the different versions in the schematics. But about the different Blitter parts and the different manufacturers (without relation to the schematics). Specifically I wanted to know if the SGS version is an earlier part (it seems so)? May be the first version of Blitter? Though you might have some idea.

[czietz]

Specifically I wanted to know if the SGS version is an earlier part (it seems so)? May be the first version of Blitter? Though you might have some idea.

A while ago, I had collected photos of Blitter chips for my list of ASIC manufacturers contracted by Atari. The oldest one (datecode 8809) is this image from Wikipedia: https://en.m.wikipedia.org/wiki/File:Atari_Blitter.jpg. It has the logo of "Innovative Silicon Technology" that according to my research was "a wholly owned subsidiary of SGS." Thus, I find it likely (but have no proof) that ST's (SGS Thomson's) Blitter is indeed an early design.

[Arne]

I do have such an IST Blitter (Datecode 8801). My father bought a new MST2 and it was delivered without Blitter. A couple of weeks later the dealer informed us that our Blitter IC arrived and it came with a crude A4 sheet of how to install it. I am very sure that it is this IST Blitter.

[ijor]

I do have such an IST Blitter (Datecode 8801). My father bought a new MST2 and it was delivered without Blitter. A couple of weeks later the dealer informed us that our Blitter IC arrived and it came with a crude A4 sheet of how to install it. I am very sure that it is this IST Blitter.

DId you run the test on that IST Blitter? And what result you got with that one, 3F or 40?
Thanks,

[Arne]

I do have such an IST Blitter (Datecode 8801). My father bought a new MST2 and it was delivered without Blitter. A couple of weeks later the dealer informed us that our Blitter IC arrived and it came with a crude A4 sheet of how to install it. I am very sure that it is this IST Blitter.

DId you run the test on that IST Blitter? And what result you got with that one, 3F or 40?
Thanks,

As you told me it could be defective I plugged it into a MST (see my eMail for further description).
Result was $0040. Seems only the ST brand result in $003F?

[czietz]

My 1040STE is in storage right now. If you tell me more about what this program is testing, I might be tempted to get the STE out of storage.

Making good on my promise: 1040STE, with ST (SGS Thomson) Blitter, datecode 8942: result 003F. Expected, I guess.

[uko]

Hi!
I have tested with my 520 STE and I get the $0040 value.
Blitter reference: S9018AE MM9092V
CPU: MC68000FN8 (so not CMOS version if I'm right)

For fun, I have tested emulators with STE configuration:
Hatari: $003F
Steem SSE: $0000 (seems strange may have made a mistake ? but I have checked blitter is well active... )

Hope this helps !

[scsilord]

My results
520STM(short) + Blitterboard Tos 2.06
520STE Tos 1.62

They were all in both machines $0040

National semiconductor: MM9092V * 8824
National semiconductor: MM9092V * 8830
National semiconductor: MM9092V * 8842
STMicroelectornics: C101643-001 * Atari 1988 * -28 8937 KOREA * MKXB75AQ-00 * Z923889
Unknown: 9029EYB * C101643 * Atari 1989 * -31 i60611-01 (back side) HONG KONG 60611-01

I wonder how many different kinds en datecodes of blitters will we get here.

[Mr Nours]

Hello

Here's the result on my STe ( 4Mb / TOS1.62Fr / Modded for selecting boot floppy ):
003f
The chip is labelled :
C101643-011
(c)Atari 1988
-28 9018 KOREA
MKXB75AQ-00
ST Z931529