Amiga.org
Amiga computer related discussion => Amiga Hardware Issues and discussion => Topic started by: Oldsmobile_Mike on August 23, 2013, 05:32:25 PM
-
Hi,
Are there any other GVP GeForce 040 board owners here that could post results of a few speed tests? As one of the pieces on my A2000 restoration project I've fitted it with what I thought would be a pretty decent board, the GeForce 040 with 16MB of memory. Overall Workbench performance is about what I'd expect and the board clocks about 25 MIPS, but disk transfers are awful - only about 1/4th the speed of my A500 according to SysSpeed at Create/Open/DirScan/etc., and SysInfo reports my A500 with SCSI card reader and x133 card hits about 2100K/sec, the best I can get out of the 040 SCSI is about 1700K/sec.
I've tried the controller with a variety of drives including several different SCSI hard drives and a Mechware card reader with CF, SD, etc. It doesn't seem to matter the card or drive I use, performance never goes much above 1700K/sec. I'm on the latest 4.15 GVP firmware, same as my A500 uses, and I've also tried different cables, checked and re-checked all the jumpers, tried formatting with different block sizes, number of buffers, MaxTransfer, etc.
A clue on Ralph Babel's site actually says (quote) "Note: Yes, SCSI on GVP accelerators typically is slower than on-board transfers on GVP Zorro-II hard cards", but really? This is awful. Just look at this screenshot, it takes Workbench almost 3 minutes to load off an x600 8GB CF card. Ugh! :(
Seems like the only thing I can do is lower the number of colors of my Workbench to get a little better performance, but even at 4 colors my A500 still crushes the A2000 with GeForce, and it's running an identical 16 color NTSC high res laced screen.
Any speed tips or comparisons would be helpful, thanks!!!
-
It's a long shot but is your 68040.library installed correctly?
-
Hi,
Are there any other GVP GeForce 040 board owners here that could post results of a few speed tests? As one of the pieces on my A2000 restoration project I've fitted it with what I thought would be a pretty decent board, the GeForce 040 with 16MB of memory. Overall Workbench performance is about what I'd expect and the board clocks about 25 MIPS, but disk transfers are awful - only about 1/4th the speed of my A500 according to SysSpeed at Create/Open/DirScan/etc., and SysInfo reports my A500 with SCSI card reader and x133 card hits about 2100K/sec, the best I can get out of the 040 SCSI is about 1700K/sec.
I've tried the controller with a variety of drives including several different SCSI hard drives and a Mechware card reader with CF, SD, etc. It doesn't seem to matter the card or drive I use, performance never goes much above 1700K/sec. I'm on the latest 4.15 GVP firmware, same as my A500 uses, and I've also tried different cables, checked and re-checked all the jumpers, tried formatting with different block sizes, number of buffers, MaxTransfer, etc.
A clue on Ralph Babel's site actually says (quote) "Note: Yes, SCSI on GVP accelerators typically is slower than on-board transfers on GVP Zorro-II hard cards", but really? This is awful. Just look at this screenshot, it takes Workbench almost 3 minutes to load off an x600 8GB CF card. Ugh! :(
Seems like the only thing I can do is lower the number of colors of my Workbench to get a little better performance, but even at 4 colors my A500 still crushes the A2000 with GeForce, and it's running an identical 16 color NTSC high res laced screen.
Any speed tips or comparisons would be helpful, thanks!!!
Perhaps it's the SCSI to CF reader and CF card? I had this exact accelerator back in the day with a 4GB SCSI-II HD and it screamed. Was getting like 6MB/s.
-
Asynchron or synchron mode?
-
Hi Mike,
I have a GVP 030 Combo Board, the SCSI Isn't that good. With the correct Mask setting to DMA Transfers into 32bit fastram, I get 1700kb/s as you mention, this is using a Mechware SCSI Card reader with an 80x Compact Flash card, But I have had the same results using a real HDD too.
Robert.
-
It's a long shot but is your 68040.library installed correctly?
I've got 3.9 with all the Boing Bags installed, as far as I know none of them make any changes to the 68040 library so I'm still running the original C= one. A little googling finds Thor's libraries, but I don't think they'd affect SCSI performance...
-
Perhaps it's the SCSI to CF reader and CF card? I had this exact accelerator back in the day with a 4GB SCSI-II HD and it screamed. Was getting like 6MB/s.
I tried it with a 4.3GB WD Enterprise drive, 7200 RPM, and SysInfo still reported around 1700K/sec.
-
Asynchron or synchron mode?
I'm not sure how I would change that?
My jumpers are set based on the information I found here:
http://www.bigbookofamigahardware.com/bboah/product.aspx?id=179
and on the amiga.resource.cx site. All caches are enabled and burst mode is on. Motherboard is a 6.4 with full ECS, MegaChip, Indivision, and heat sinks everywhere. As far as I can tell everything works great on both the boards except my clock chip, LOL (still struggling to get that working). ;)
-
I'm not sure how I would change that?
HDToolbox or whatever you use to control your HD.
-
Perhaps it's the SCSI to CF reader and CF card? I had this exact accelerator back in the day with a 4GB SCSI-II HD and it screamed. Was getting like 6MB/s.
Interesting trick considering the gvp gforce 040 uses the Am33c93a (same as WD -08/am33c93a in the A3000) which specs in the data sheet at 5MB/s max in synchronous mode,and that's in laboratory setting most likely. so if you even see 4MB/s you would probably being doing well.
the pcd series card reader are not as fast as the mfg specs for the same reason,most likely lab specs. I have attained 3.9 MB/s from it with a class 10 SDHC 4GB card from transcend on the csppc scsi. I doubt anyone will see this on the same speed on narrow scsi cards. Of course the benchmarks are hard to trust.
-
I'm not sure how I would change that?
My jumpers are set based on the information I found here:
http://www.bigbookofamigahardware.com/bboah/product.aspx?id=179
and on the amiga.resource.cx site. All caches are enabled and burst mode is on. Motherboard is a 6.4 with full ECS, MegaChip, Indivision, and heat sinks everywhere. As far as I can tell everything works great on both the boards except my clock chip, LOL (still struggling to get that working). ;)
gvpscsictrl is the utility you need to set synchronous mode. it should be on the gvp install disk i think
Also another note, if you are using any external devices,i seem to recall the gvp does not supply term power.
I wonder if you can do speedgeeks 33c93a scsi chip overclock hack for more speed? feel like being a guinea pig on the gforce mike?
-
gvpscsictrl is the utility you need to set synchronous mode. it should be on the gvp install disk i think
Also another note, if you are using any external devices,i seem to recall the gvp does not supply term power.
I wonder if you can do speedgeeks 33c93a scsi chip overclock hack for more speed? feel like being a guinea pig on the gforce mike?
There is a good chance the 33C93A is already running at 14MHz this being a later model/revision accelerator card.
The 33C93A is rated for 16 MHz and therefore is NOT overclocked at 14 MHz... but it does need to be configured correctly (for reliable synch mode operation) at any supported CLK speed.
-
Interesting trick considering the gvp gforce 040 uses the Am33c93a (same as WD -08/am33c93a in the A3000) which specs in the data sheet at 5MB/s max in synchronous mode,and that's in laboratory setting most likely. so if you even see 4MB/s you would probably being doing well.
the pcd series card reader are not as fast as the mfg specs for the same reason,most likely lab specs. I have attained 3.9 MB/s from it with a class 10 SDHC 4GB card from transcend on the csppc scsi. I doubt anyone will see this on the same speed on narrow scsi cards. Of course the benchmarks are hard to trust.
Well obviously my memory is lapsing a little then. It was 4-5MB/s. This was around 1997, so forgive me. All I know is the machine was very fast and the HD was much faster than a A2091.
-
There is a good chance the 33C93A is already running at 14MHz this being a later model/revision accelerator card.
The 33C93A is rated for 16 MHz and therefore is NOT overclocked at 14 MHz... but it does need to be configured correctly (for reliable synch mode operation) at any supported CLK speed.
Ok, good to know. I should of checked first before speaking-you didn't overclock it, just brought it close to its rated speed.
i have no idea what the Gforce runs it at.
There is a 20mhz version of the chip also, it may be plcc style only(edit- no there is a DIP version). Have you tried these?
I wonder if there would be any gains on overclocking it,assuming it would tolerate it?
-
Well obviously my memory is lapsing a little then. It was 4-5MB/s. This was around 1997, so forgive me. All I know is the machine was very fast and the HD was much faster than a A2091.
No worries, we are all getting old. I wish i could remember everything i knew :)
-
Ha, I briefly looked at that overclocking hack (and recall having recommended it to people in the past) but it seemed more geared to A2091/590 users. Good thing I didn't spend too much time looking into it! ;)
Any tips how to use the gvpscsictrl utility to force synchronous mode? My A2000 is in pieces right now but I'll put it back together later tonight and see if I can find anything, the manual didn't mention how to change it at all.
I'd be happy if the darned thing was just as fast as my A500 with Series II controller and the same card reader. As it is I just went ahead and purchased a cheap A2000 HC8+ off ebay. Another expense I didn't need but it will be interesting to benchmark that card against the GeForce if as Babel's site says, that the HC is faster than the accelerators. :-/
-
Let us know how you get on using a Zorro GVP SCSI Card with a GVP CPU Board that also has SCSI Onboard - I tried that combination but couldn't get it to work.
Robert.
-
Let us know how you get on using a Zorro GVP SCSI Card with a GVP CPU Board that also has SCSI Onboard - I tried that combination but couldn't get it to work.
Ruh-roh!
According to the manual jumpers J4 & J5 should disable the SCSI drive check and Boot ROM on the GeForce. But assuming they don't... anyone want to buy a GVP SCSI board? LOL. :p
-
I think I may have read somewhere about removing the SCSI Rom from the CPU Board to disable it completely, but I had given up and accepted my 1.7mb/s by this point. I've also noticed the SCSI Uses a lot of CPU Time, have you had any such affects on your 040 board? (Mine is 030).
Robert.
-
No worries, we are all getting old. I wish i could remember everything i knew :)
It's terrible isn't it! I can clearly remember being two years old when my Mum was pregnant with my sister in quite good detail (34 years ago) yet if you ask me what I did one week ago I'll struggle to remember. :/
-
Ruh-roh!
According to the manual jumpers J4 & J5 should disable the SCSI drive check and Boot ROM on the GeForce. But assuming they don't... anyone want to buy a GVP SCSI board? LOL. :p
lol
I'm toying with the idea of selling my 128MB Cyberstorm Mk2 060@50+ CyberSCSI module so if I do decide to then that will at least provide you with an option to replace your existing card. :)
-
is your scsi chain correctly terminated?
-
Yes it's properly terminated,
Term on controller - Device - Term :-)
Robert.
-
So after poking around with gvpscsictrl and HDToolbox for a while, it turns out the only way I was able to format one of my spare CF cards into Synchronous mode was with the much maligned HDInstTools. Neither of the other applications had an option for it (gvpscsictrl is really just for slow old drives and removable storage, SyQuest drives, etc.).
After formatting and prepping the card SysSpeed reported that it was faster, but SysInfo said it was exactly the same speed. Really I don't have much faith in either application, or in HDInstTools for that matter, LOL. ;) What tools are kids these days using to measure their Amiga HD speeds?
-
So after poking around with gvpscsictrl and HDToolbox for a while, it turns out the only way I was able to format one of my spare CF cards into Synchronous mode was with the much maligned HDInstTools. Neither of the other applications had an option for it (gvpscsictrl is really just for slow old drives and removable storage, SyQuest drives, etc.).
After formatting and prepping the card SysSpeed reported that it was faster, but SysInfo said it was exactly the same speed. Really I don't have much faith in either application, or in HDInstTools for that matter, LOL. ;) What tools are kids these days using to measure their Amiga HD speeds?
Turning on synchronous has nothing to do with format partitioning-it doesn't have any ill effects on a already formatted or installed hd. gvp tools allowed you to turn on any device as synchronous(many controllers make it all or nothing). gvscsictrl was the tool if i recall.
HdinstTools messes with the rdb in a non standard way.
latest ver of the gvpscsictrl is here: http://www.gvp-m.com/updatejapan.html
beware if you try to use normal rdb standard tools on the hdinsttools prepped drive,i think it may corrupt it.
Btw, are you using sfs or pfs? you will never get good speed on FFS.
-
I think I may have read somewhere about removing the SCSI Rom from the CPU Board to disable it completely, but I had given up and accepted my 1.7mb/s by this point. I've also noticed the SCSI Uses a lot of CPU Time, have you had any such affects on your 040 board? (Mine is 030).
It seems like we have fairly similar systems and are experiencing the same problem, which again points to Babel's being accurate that the SCSI on the GVP accelerators just sucks. I haven't measured CPU speed yet, what tool did you use for reference? Last time I routinely measured my processor utilization was with TinyMeter, about 10 years ago, LOL. ;)
-
Mike,
i just downloaded the gvpscsictrl from the link above, if you look in the tooltypes of the prog it shows SYNCHRONOUS ON/OFF/TOGGLE in there.
i *thought* it had a menu selections to toggle it on off for each device on the chain,but maybe my mem is failing me. the tooltype suggests it is there tho.
Btw, at the end of the day i suspect the guru rom will be the only help for it,which is hard to find. Ralph will not give out the board design to make new guru roms,he claims that there are still stores selling it,but there are none.. probably software hut listing it still makes him think that.
Gvp had a long history of bad scsi roms and i avoided most their stuff except for the A530(which had the same rom/scsi troubles). the 2091 is no peach tho :D
It is amazing gvp could cripple scsi on a accelerator capable of 32bit to where their own zorro2 card was slightly faster.
The best thing would be a blizzard 2040/2060 or DKB wildfire,then you can have near 10MB/s...
-
Turning on synchronous has nothing to do with format partitioning-it doesn't have any ill effects on a already formatted or installed hd. gvp tools allowed you to turn on any device as synchronous(many controllers make it all or nothing). gvscsictrl was the tool if i recall.
HdinstTools messes with the rdb in a non standard way.
latest ver of the gvpscsictrl is here: http://www.gvp-m.com/updatejapan.html
beware if you try to use normal rdb standard tools on the hdinsttools prepped drive,i think it may corrupt it.
Btw, are you using sfs or pfs? you will never get good speed on FFS.
When I clicked the toggle in HDInstTools to turn on Synchronous for the drive it stated that it required a reformat. Then after I finished my speed test and turned it back to "normal" with HDToolbox I had to rewrite the RDB. Don't worry, I have no plans to use HDinstTools on a regular basis and am only using it on a spare CF, I recall all your warnings! ;)
I'm using FFS right now. Not looking for "amazing" speed, just trying to get a grasp on why the performance is so much worse on this GeForce-equipped A2000 than it is on my A500. This is across-the-board performance degradation, it doesn't matter if I'm using a hard drive or one of your card readers, the GeForce is approx. 20% slower than my A500. Both have latest 4.15 GVP ROM, etc.
Thanks for the link, I'd been trying the tools off the GVP install disk and Ralph Babel's website, didn't realize there were newer versions. More testing is in order! :D
-
As a word of caution for anyone who stumbles across this thread in the future, I would recommend avoiding GVP "ExpertPrep" like the plague. It's probably even more tempermental than HDinstTools, and has hosed many a partition just by opening the application. After enough screw-ups I've been able to detect it's pattern - as soon as you open the app it resets all of the partitions to "NOMOUNT", and changes your block size to 512. So as soon as you reboot all of your drives are gone, yay! :( Fortunately it can be fixed by booting off a 3.9 recovery disk and opening HDToolbox, and resetting all your parameters - block size to whatever you had it before, and whether the partition is to be automounted or bootable, with no data loss. Yeah, care to guess how many tries it took me to learn this? ;)
And now back to installing patches on my system to tweak performance. Amiga OS, I love you, LOL. ;)
-
Hey mike,
I gauged CPU Usage by using a program called RSCP From Aminet, it measues disk speed and free CPU Use. But before that I noticed the mouse pointer was very slow whilst transferring files or formatting a partition.
Robert.
-
Well, the HC8+ arrived, I plugged in 8MB of spare memory, flipped a few jumpers around on it and on my GeForce, installed it, plugged my SCSI cable into it, flipped the switch, early startup control showed all boards working and all drive partitions detected, told it to boot... and nothing. Black screen. :(
After some more playing around with jumpers I was able to get into a "boot with no startup-sequence" mode and ran SysInfo, even with no SetPatch and half the caches disabled on my accelerator I was still getting 1,900K/sec, so this looks promising, but it seems like it's going to take quite a bit more fiddling around to get it to boot. I can't even boot off my 3.9 Emergency floppy or a stock 3.1 Install disk, currently. They all just blackscreen.
Or occasionally I get errors like this attached photo. :-p I think it's a conflict between gvpscsi.device on the accelerator and gvpscsi.device on the card. I seem to recall seeing somewhere that it should autodetect and pick up the second one as gvpscsi2.device, but that's not happening, and no amount of swapping jumpers around has found me the correct setting yet. May just try running the accelerator with no ROM at all, what's the worst that could happen, but it seems like with the ROM disabled I was also loosing my fastmem on the accelerator. (according to avail) ???
-
Going back to the old troubleshooting process, I reset all the jumpers on my GeForce and plugged my SCSI chain back into it, exactly as I've been using it, pulled the "boot" jumper off the HC8+, and tried to boot that way. See if my system was even stable with the HC8+ in there acting as nothing more than a RAM card. Nope. Got another garbled error message about something in-use.
Was able to boot with no startup and run a few commands however, and that's when I noticed that $)(%@ GVP used the same manufacturer and board # ID's for both the SCSI/RAM portion of the GeForce board as they did with the HC8+, and it appears to be some incompatibility on a base hardware level.
Anyone more experienced in this take a look at the attached two screenshots and chime in? I can already tell it's reading them wrong based on the amount of fast memory it's showing (should be 24 MB, 16 on the GeForce and 8 on the HC8+, but it looks like it's reading the second card twice). Crud, crud, crud. :(
So if I'm understanding this correctly, it appears my plan can't work because they have identical ID's, but are not identical cards? Then the only way to do this would be to get a different accelerator, or a different HD controller card, that don't conflict?
(disregard date on clock being wrong, I still haven't fixed my clock chip yet either, LOL)
-
The double vendor ID is at this point nothing that troubles me so much. From the former screenshot, the one containing the y's with the dots on top, it rather seems to me that either the RAM on the board is defective, or the communications between the motherboard and the expansion card is defunct. Given that this is an A2000, this is unfortunately not as unusual as you may believe. Which motherboard revision do you happen to have? Trouble is, the expansion bus of the A2000 is prone to miscommunication due to bad electrical termination. A popular fix for this is to add pull-ups at the bus drivers close to the CPU at the right side, but depending on your board, this may or may not have the desired effect. Reasoning: The y-character you see there has the ASCII (actually ANSI) code 255. This can happen because either the RAM is forgetting its contents, is defect or the communication between RAM and the rest of the system creates the wrong signals and thus returns 255 where it should return the actual value. Some combinations of expansions and A2000 generations simply cannot made to work. I had this trouble here with my A2000 for years until I finally gave up on this. Lousy design/engineering of the expansion bus is the culprit.
-
Hi Thomas,
Thanks for the followup. It's a Rev. 6.4 motherboard, full ECS with Megachip. System is stable with everything else installed, just when I try putting the HC8+ in there I can't boot (even off a floppy). I've tried with and without the DTAK pullup jumper on the HC8+, and also tried removing all my other cards and putting the HC8+ in various other slots.
Interestingly when I remove the ROM from my GeForce and flip the jumpers to "no attached drive, no autoboot" I also loose the memory on my GeForce. (seems to be more from the rom being removed than the jumper moved)
You may have a point about it being bad memory on the HC8+, although I used these chips in my A500 GVP controller successfully in the past with no issues. Off to hunt down a memory test program, since from what I read in a thread on Amibay I should at least be able to use the HC8+ as an additional 8MB RAM board, with everything else disabled, and I'm not even able to do that currently. :-/
-
Progress!
Reset my A2000 back to the configuration I had it working at prior to this latest phase of mucking about. Took the HC8+ and removed all it's RAM, removed it's boot ROM, removed the autoboot jumper, wanted to see if it was functional as just a bare card... and the system booted up perfectly. So it appears that all along issue has been with bad ram on the new card. Thanks again, @Thomas, for pointing me in the right direction, and PS - your libraries are awesome, just installed them a few days ago, huge improvement! :D
Now to take a victory break and look at re-adding components to the HC8+ one by one, starting with memory until I find the bad stick, then maybe the drives. Later. ;)
-
Be careful, though. It *may* be the RAM, but if the bridge to the motherboard is the problem, then it *may* only seem to work and problems will appear as soon as you try to attach a drive to it. You then get the phenomenon that disk I/O is *sometimes* broken, causing flipped bits and damaged files. Been there, done that. Other than that, the 6.4 motherboard should be at least a lot better than my 4.1.
-
Ha, yeah, intermittent problems are my life story. ;) I tried the card with 8 different 1MB simms, all of which had worked correctly in my A500 sidecar in the past, in pairs of 8, 4, and 2. At no point with any combination of them (and with the card in different ZII slots) would it boot. I'd get those gibberish messages like posted above, or it would freeze up. However the card works fine with no memory installed, I was even able to use my SCSI card reader and hard drives connected to it (but for whatever reason it wouldn't see my optical drive). The card works with no ram installed and it boots up, but is pretty slow because of the bus. 800K/sec, give-or-take.
Seems like that narrows it down to:
- either one of the first two slots on the HC8+ is bad, where the first memory sticks are required to be inserted. This might be a possibility, and might explain why the seller immediately accepted my relatively low "make an offer" price on ebay. :-/
- some incompatibility between ram on the HC8+ and my GeForce
- some incompatibility between ram on the HC8+ and my motherboard
No real way to test a couple of those options, if I had a spare HC8+ I could swap in, or if I had a complete other A2000 that I could try this one in... Bah. My quest to add a faster SCSI interface to my A2000 project will have to be continued till another day! I should probably also test the HC8+ without my GeForce installed, just to see what happens there...