Welcome, Guest. Please login or register.

Author Topic: GVP RAM32 board with A2000-030  (Read 3971 times)

Description:

0 Members and 1 Guest are viewing this topic.

Offline thebajaguy

  • Newbie
  • *
  • Join Date: May 2017
  • Posts: 34
  • Country: us
  • Gender: Male
    • Show all replies
Re: GVP RAM32 board with A2000-030
« on: May 20, 2018, 06:43:19 PM »
Original A3001 had 4 MB and 8MB of Nibble mode SIMMs, placed via AutoConfig, into the 8MB space.  ~1988-1989 IIRC.   Supported DMA from Zorro II.  I think it beat the A2630 out (not the A2620), and was the first to do this.  Ronin, CSA, and any others of the early accelerator group either mapped it high, or didn't support DMA in the AutoConfig space, if they autoConfig'd it in the first place (AddMem!).  It was called the A3001 because it was better (faster) than the A3000 that was yet to appear on the market.  It pissed C= off a tad, at the time, yes.

Nibble-mode memory directly supports 68030 CPU burst mode, which sends 4 bits instead of one per memory access, thereby populating the cache more efficiently for sequential instruction and data accesses.  you needed 4 modules as each 30-pin SIMM was 8-bits wide.  You pay a small penalty if all you needed was 1 32-bit longword.

If you had a sequential need for data, and memory clock penalty was 3 for each 32-bit longword, you got 4 longwords in 12 clocks.  In burst, it takes 6, assuming 3-1-1-1.   Varies based on memory speed, and the clock of the CPU.  I don't recall the settings on the memory boards, but 3-1-1-1 and 4-1-1-1 seem like they applied - distant memories.  50Mhz may have been 5-2-2-2 or 6-2-2-2, but remember this is 50Mhz clocks, not 25/28 or 33Mhz.  

RAM32 memory card and the SIMM32's came out to replace the nibble-mode SIMMs and uses them in groups of 4 to get the same kind of 'burst' from memory that only sends 1-bit (x32).  At the time, 32-bit SIMM standards were not established.  GVP made their own, and the design was for 1, 4, and 16MB modules, but the original boards only supported the classic 8MB AutoConfig.  Same with the 33Mhz.  Note the SIMM32 memory is not 80ns or 70ns like the nibble-mode memory, but 60ns, so boards' memory logic are tuned to that speed, which improved at some speeds.

The RAM32 board is identical between speed variants, but the PALs used  (and jumpers) select different AutoConfig amounts, clock timing, and  addresses the memory is found at.  Memory mapped high is detected by the  v3.x driver ROM and mapped into the free memory list very early on.

Later, the 50Mhz top of the line board got 4/8 Autoconfig with 1MB SIMMs, or the same 4MB with 1MB modules + 4/8/12/16 mapped high with 4MB modules, and the memory clocking was adjusted for 50Mhz.  A later revision got 4/8MB AutoConfig with only 4MB modules, and 4/8/12/16/20/24 mapped high, or optionally all modules mapped high for 32MB.  

 The low end 25/28/33Mhz A3001 sandwich boards were replaced with the Combo 22/33Mhz SMC series, and then all were replaced with the G-Force (PGA CPU) 68030 line (25MhzEC/40MhzEC/50MhzFull).


The G-Force 68040 was the first to use the 16MB variants of the module, but supported the 4MB version.
« Last Edit: May 20, 2018, 06:58:38 PM by thebajaguy »
Former GVP Tech Support 1989-1993 - Bloodied in the original trenches of the Amiga Wild Wild West.
 

Offline thebajaguy

  • Newbie
  • *
  • Join Date: May 2017
  • Posts: 34
  • Country: us
  • Gender: Male
    • Show all replies
Re: GVP RAM32 board with A2000-030
« Reply #1 on: July 16, 2021, 09:03:21 PM »
It can be said that a 50MHz-configured board will run at 33MHz or slower, but it's memory-access timing is not optimal for that clock.

The programmable logic in the PALs, combined with the jumpers (which control clock and bus-signal sources, some being negated, etc, if I am recalling correctly) produce the optimum timing for the board speed it was set up for.  Trying to go faster (clock) on a slower-configured memory board won't work.

Although some logic is common between boards for a few things (the memory-detect and/or AutoConfig details), there were no 'generic' part sets that applied to the address-generation/data-access logic.  It was timed to a clock speed for best performance.
Former GVP Tech Support 1989-1993 - Bloodied in the original trenches of the Amiga Wild Wild West.