A4091: Couple of Q's

  • Thread starter Thread starter Damion
  • Start date Start date
  • Replies Replies 16
  • Views Views 2972
D

Damion

Member
AmiBayer
Joined
Mar 26, 2008
Posts
167
Country
USA
Region
Nevada
Hi all,

Anyone here running a 4091 and getting better than 4.5 MB/s? This is with everything enabled/configured as it should, and with drives that do 8 - 10 MB/s on other SCSI-2 controllers. (TekMagic 2060, Blizz 2060, CSMK2 SCSI, for example.) Trying to determine if this is the real limit of the 4091 (which I believe to be the case) or a specific issue with my hardware combo. SysSpeed and/or RSCP benchmarks would sure be appreciated.

Question 2: Anyone still have the old 40.9 ROM on their 4091, and want a FREE upgrade to 40.13? I would like to send you the 40.13 ROM, you send me your old one with the 40.9 code, then I'll dump the code and send your original ROM back. I'll also pay postage all directions. :o

TIA for any and all replies concerning the 4091. :pint:
 
Lots of things can impact the transfer rate, such as the measuring program, async vs sync, the actual HHDs speed (RPM, cache, type), CPU load, DMA, and stuff. Compare apples to apples. I know intimately the A4091's NCR 53C710 and the CS's NCR 53C770. They are different beasts.

Ask for number using the same program (same amount of data used in transfer), same CPU, same OS (yes, they matter), same hard drive, and a machine outfitted the same as yours.
 
Thanks for your input.

Lots of things can impact the transfer rate, such as the measuring program, async vs sync, the actual HHDs speed (RPM, cache, type), CPU load, DMA, and stuff. Compare apples to apples.
Click to expand...
That's exactly what I've done, the other two systems I have with 53C710 controllers (like the 4091) do over 9 MB/s with the same exact drives, OS installs, benchmark programs, and processors, hence the question of why the 4091 achieves half the raw transfer rate of the others. The issue of RDB configuration, termination, etc has been eliminated by using the same drives and cabling between systems. Actually, the 4091 ignores some RDB flags (like sync/async) in favor of the dip-switches on the rear.

I know intimately the A4091's NCR 53C710 and the CS's NCR 53C770. They are different beasts.
Click to expand...
Which is why I've kept my comparison base strictly to fast SCSI-2 controllers, like other 53C710 based cards and the Blizzard 2060/CSMK2 FAS216. (Fastlane is also reputed to do at least 8 MB/s.) I know the CSPPC/MK3 is a different beast.

I have discovered from old usenet posts that the "fast synchronous" mode of the 4091 may have never been enabled in the card's software. If my suspicion is correct, no more than 5 MB/s is possible on any configuration with the above mentioned benchmark programs. I'd sure love to be proven wrong though. ;-)
 
Wow, I'm sorry I answered. You've researched this at much greater depth than I. I really never compared SCSI controller speeds beyond the two I mentioned and a GVP SCSI Impact card.

Go in Peace and Prosper
 
No, I'm sorry, didn't mean to come across the way I have! I certainly appreciate your input. Just wanted to clarify that I believe it's an unresolvable issue with the 4091, and was hoping someone might come along and prove otherwise. :pint:

S'pose I'm just a bit bummed here - spent a bunch of time and money trying to get a decent HD setup on my A4000/CSMK2 '060 setup.

MK2 SCSI: No go, since I'm one of those nasty overclockers (the card is rather sluggish at 50MHz)

DENEB: Read errors between devices on the DENEB, no problem copying to and from other devices (internal IDE, etc). Spent endless hours trying to figure it out.

Fastlane Z3: Arrived b0rken, many more hours wasted

A4091: Works well, nice card minus the transfer rate problem. (Bugs me, since I'm a bit of a perfectionist and want everything to work like it should!)

With all the $$$ spent, I should have just bought a CSMK3 in the first place and called it job done. :Doh:

/whining
 
Dude, no problem. I have a CSPPC A4000T and use the UWSCSI for all the heavy lifting. The A4091 built in handles other devices in the SCSI-2 range where speed doesn't matter. Believe it or not, HD access through the Deneb USB device will give 5+ Mb/s speed. Deneb has issues with Fat95 (and NTFS) file systems and occasionally with larger transfers; haven't worked these out yet.
 
I'm a little bit of a late comer to the whole A4091 game. But the problem is not the software and it is not about SCSI2 or synchronous mode.

The card does not implement Zorro III bursts. Theoretically this could be implemented in the logic code of the board, as the older revisions of that code in the Dave Haynie files still have remnants of that implementation. But it is fair to assume that it was taken out for a reason.
 
Also check out this thread on eab. Turns out speed from 4091 to ZorroIII memory is higher than to mainboard memory. Of course consecutive access to that data is slower on ZorroIII, so it's a bit of a trade-off.

Turns out you can also run your 4091 with a 66mhz or 72mhz oscillator. The 710 is specced to work at 66/33mhz.

We're also getting on pretty well with the planned open source driver for the 4091 but the driver is (for all i know at this point) not the bottle neck.
 
Yeah I tried a lot back in the day to get my 4091 faster than 4MB/s. Seems it's normal for this card. I always wondered why it would never do 10MB/s as advertised.
 
10MB/s is the maximum bus speed on the SCSI2 bus. There are very few if any Zorro cards that will realistically get to that speed, even on ZorroIII.
 
it does not support zorro III burst cycles, transfer rates depend on zorro bus, with C/S MKII 060 50Mhz i get 4.2 MB/s with 66Mhz it is 4.5 mb/s, when i remember correctly fastlane z3 used burst ...
here is what the RE4091 is about
 
The Fastlane is faster because it has onboard Zorro III memory. When you put ZorroIII memory in your Amiga next to the A4091 you can get faster transfers to the ZorroIII memory as well.

See this benchmark with a GottaGoFaZt3r and an A4091 clocked at 72MHz.

There are other ways to speed up an A4091, such as enabling the Early Byte-Lane Option which will get you about 10% speed-up without overclocking.

YMMV
 
When i had fastlanes arround, we never put memory on them so memory can´t be the only clue here, also Fastlane was running at 20 or 25 Mhz (i don´t remember that one exactly), anyway it is great to see the work done here, maybe i am going to try the 33 Mhz and early byte as faster is better ;-)
 
Yeah with the closed design of the Fast lane that is forever lost beyond the existing cards it will be hard to figure out what exactly they were doing.
 
Hannibal said:
Yeah with the closed design of the Fast lane that is forever lost beyond the existing cards it will be hard to figure out what exactly they were doing.
Click to expand...
yeah, sadly also for CS PPC, Blizzard 2060 etc.... they really knew how to do it
 
Correct, the Z3 bus limitations were a problem; IIRC Haynie didn;'t realize that that would limit the transfer rate until we had working HW and drivers. This limitation doesn't exist in the A4000T implementation, which uses the same driver.
 
You must log in or register to reply here.
Back
Top Bottom