Amiga.org
Amiga computer related discussion => Amiga Hardware Issues and discussion => Topic started by: darksun9210 on January 06, 2020, 10:47:18 AM
-
hey all, not sure if i'm doing something wrong here, or if i've got a "known bad" config, or if it's just a "thing" but my A3640 card in my A4k....well in short, it sucks! i think its a rev3.1 card.
i've tested it in my A3k and the performance is about same to within a few seconds between the A3k and A4k in test renders at 4hrs 53mins, so i'm pretty sure it's not the machine.
i mean, obviously its a step up from the 030. but still.
for example. An A1200 fitted with a blizzard040@25 is nearly an hour and a half faster at 3hours 30mins, than my A4k with the A3640.
i think it's something ram related, as i was able to bring the render time for the A4k+A3640 combo down to sub 5 hours when i found some more sims to take the mainboard ram from 8 to 16meg - previously it was over 6 hours. which is a bit of a worry when a blizzard030@50 equipped A1200 will so the same job in 5hours 10mins.
Another hint is that sysspeed and bustest still tells me that my fastram access speed is in the low single digits. yes. in some cases it'd be quicker to write out to disk than ram. :-\
however, all the cpu bench programs just confirm that everything is performing as expected mips and mflops wise. which i suppose they would if their test code fits in the cpu cache...
anyway, config as follows if anyone can see anything obvious:-
A4000 desktop with 3.1roms, 32gig CF card and a cdrom on the IDE
A3640 rev3.1 (with heatsink and fan)
2x 8meg sims on the mainboard, + 2meg chip
Mediator 4000Di
Bigram 256meg Zorro3 ram
X-Surf100 in Zorro3 mode
Radeon 9250 128meg PCI (set up as 64meg graphics ram and 64meg ram/PCI DMA space)
some generic 100megabit PCI NIC
hope someone can spot something or have a suggestion of what to do! otherwise i'm on the hunt for a cpu card with it's own memory i guess?
cheers all! :)
-
otherwise i'm on the hunt for a cpu card with its own memory i guess?
Correct.
-
So, at the risk of poking the potentially large elephant in the room, i take it the A3660's don't suffer this ram access issue?
if anyone's got one of these, i'd be interested in seeing your sysspeed or bustest ram results...
-
Of course it has the very same issues, as it is the same design but with 060 CPU.
-
ah. bother. i did have hopeful imaginings of reprogrammed Gals or Pals or whatnot.
ok, so that's totally off the table then.
ok so lets see if an A3/4k varient of a TerribleFire card materialises later this year (high hopes) ;D
-
Team Apollo's vcontrol tool already has an id for Vampire3000/4000, so.....
-
I have a 3640 with the Speedgeek Pals and Gals which fixes the slow memory access. The memory access is as fast as a 3630 card.
-
ahaaa yes. the Vampire for the 3000/4000. well, my name is down for one of the A1200 versions.
i think there was an idea that they were going to use that version as the base, and and then an adaptor for the cpu slot of the CD32 and A3k/4k machines.
but then i was thinking, if it was possible to jam an A1200 card into a CD32, wouldn't someone have done it already? especially as i saw a DCE sx32pro go for over 2grand on ebay recently if i recall?
i think the TF cards are on the money too. keep it simple, keep it affordable. hopefully the TF360/TF1260 version will grow a big box CPU slot adaptor. :)
oh cool, i've been reading up on the speedgeek mods this afternoon - i didn't even know they existed! i think the 64meg hack is beyond me. but what do i need for the 3640 gal/pal programming? is this something i can do like a re-flash with an eprom burner or something similar?
-
I can confirm the the A3640 is a mediocre card at best. RAM access is fairly slow. When I added a 68060 to mine (with the adapter from TBTORRO) it certainly helped, but was not even in the same league as my Warp Engine card.
-
you could ask Cosmos to help you to ged the Speedgeek mod:
https://forum.amiga.org/index.php?topic=68796.0
-
nice one, cheers guys, i'll see if i can reach out to Cosmos re the ram access timing mod. i am curious to see what difference it can make, for something that could have maybe been in there to begin with?
also thanks for the info on the 060 upgrade, i wasn't sure if it was a worthwhile deal given the ability of the card/ram to support the chip itself - but i'll keep an eye out if anyone starts making adaptors again.
and that 64meg ram mod. i've seen reports of some people loosing access to Z3 cards. now there was some indication that it may have been down to a particular mainboard revision... or as with most things amiga, pure luck of the draw! so i think i'll leave that for the time being!
-
I had the state machine ram timing mod on my old a3640 and it improved ram access speed around 10-20%... so not very much at all.
My 4000 has the 64mb mod and it has caused not issues with zorro cards or mediator pci cards.
-
Ah-hah. ok that sorta ties in with what i've read. reducing the wait from 8 ticks to 6, in line with the A3630. but the A3640 doesn't support burst mode ram access apparently. whatever that is.
so while the 64meg upgrade is tempting, and the A3640 IS tweek-able, i think i'm going to leave my machine and cpu card as vanilla, as it's a lot of faff for not a lot of return. We're in a good time for new amiga bits appearing it seems. Just re-cap both, and wait for something more juicy to appear for the CPU slot later this year.
[edit] typo's
-
Yes, normally one would suggest you hunt a cyberstorm 2 or 3, but then you need a spare kid(ney) to buy it. :P
-
yeah we do seem to be in a bit of a sellers market for the last year or three! :-\
looks like i'm waiting for the TF/icomp/apollo pot of gold at the end of the rainbow ;D
but i'll keep my ear to the ground in the mean time
-
Burst mode is where you can transfer an entire cacheline (16-bytes) with only one setup cycle(s), as opposed to having to call out each 32-bit read/write. i.e. after the setup then each cycle you squirt another 32 bits down the bus for the next 3 cycles. This only works on a synchronous bus obviously, and the 68040/060 and 68020/030 buses are different in other ways as well. The A3000/A4000 fast slot is on a 68020/68030 bus. This is normally only handled by the caches but you can explicitly take advantage of this as well on 040/060 with MOVE16.
The A3640 has no local memory to burst to, and runs on a dissimilar bus. Even you run it synchronous to the motherboard bus (Speedgeek does this to minimize wait states), it has only bare-bones glue that will not let you burst across it.
-
How fast is the ZIII RAM compared with motherboard RAM? I'm guessing it's slower and the 16MB isn't enough to hold your OS boot and your 3D rendering program, hence the program code get's shoved into ZIII and it's doomed to run slower. Perhaps your OS needs streamlining a bit? Just a thought that's all, as those figures seem a bit extreme - the 040 should be rendering quicker than the 030. Didn't people buy A4000/040's for quicker rendering times?
-
Good info @AmigaHope , i've heard of "burstmode" for years, but never actually understood what it was!
Z3 ram is slower than mainboard ram. i'll try some bustests to compare if i get a chance this evening
@paul1981 good point, i'll try a minimal boot to see if i can keep everything in the 16meg. maybe even pull the z3 ram card out just to see what difference that makes if any.
To be honest, an 030@50 with local ram outgunning an 040@25 without local ram both does and doesn't surprise me :D
The Blizzard i have with an 040@25 and local ram shows what this CPU could do if supported properly i guess.
an A3630 does 10:34:31 (hrs:mins:sec) with a 25Mhz 882, and putting a 33Mhz crystal on there brings that down to 7:50:54
-
Just as a reference, this is my bustest output (10^6 byte/s):-
A4000 A3640 buster rev11
mainboard ram:-
readw 11.1
readl 11.5
readm 11.3
writew 6.0
writel 6.0
writem 6.1
Zorro3 ram (bigram256)
readw 7.5
readl 7.6
readm 7.6
writew 3.9
writel 3.9
writem 3.9
mediator graphics card ram (radeon 9250 128meg)
readw 2.5
readl 4.9
readm 4.9
writew 3.7
writel 7.3
writem 7.4
chipram
readw 2.3
readl 4.7
readm 4.6
writew 2.3
writel 4.6
wrltem 4.6
-
That is shocking.
-
as additional
Bustest output (10^6 byte/s):-
Mediator removed, standard busboard backplane re-installed, and Cybervision64/3D results.
(Glad to see both the Phase5 and DCE versions of the card need the card guide piece of plastic removed from both an A3k and A4k. anyway...)
readw 2.4
readl 4.8
readm 4.9
writew 3.1
writel 6.1
writem 6.2
now i distinctly remember getting 14meg a sec across the zorro bus when i had my cyberstorm2/040 card. so i'll keep hanging on for something similar.
-
I am building an Amiga Mediator Tower out of spare 4000D parts I have. I have am using the A3640 CPU card V3.1. Why does it take 30 seconds to start booting? If I hit a stop watch with a reset, it will not try to access the floppy or bring up the BIOS screen if you hold the two mouse buttons down until 30 seconds everytime. I remember something about this delay. How can I fix it?
-
do you have any IDE harddrive installed yet? If not, the machine waits for spinup of one just to be sure there is one present - or not.
-
Is there a way of shutting off the IDE or do I just have to install and IDE drive? I have the GVP SCSI Impact HCII right now. I don't remember it doing it when I was testing with my CyberStorm III card.
-
http://amigakit.amiga.store/product_info.php?products_id=831