Can't get them made at JLC
REV 3 - REV 5 - The beginning (ST536)
Re: REV 3 - REV 5 - The beginning (ST536)
Claude been busy designing a breakout fpga board.
Can't get them made at JLC
Will have to use my other fab house. Probably get 5 boards made. Won't be cheap but I need to find a cheap fpga. It's 3.3v with 1.2v core. Its got 8MB of sdram built in which is pretty cool. Plus a proper PLL clock input. Etc.
Can't get them made at JLC
Re: REV 3 - REV 5 - The beginning (ST536)
Once you've built in area mode, did the basic functionality work (ie. no TT-RAM, no IDE)? If not, then yeah, you've got bigger fish to fry, but if it *does*, then you end up with a more stable platform to work with at the expense of now needing to rework the SDRAM controller.
BW
BW
DFB1 Open source 50MHz 030 and TT-RAM accelerator for the Falcon
Smalliermouse ST-optimised USB mouse adapter based on SmallyMouse2
FrontBench The Frontier: Elite 2 intro as a benchmark
Smalliermouse ST-optimised USB mouse adapter based on SmallyMouse2
FrontBench The Frontier: Elite 2 intro as a benchmark
Re: REV 3 - REV 5 - The beginning (ST536)
That basic test without TT-RAM ran fine, but it doesn't tell me much I didn't already know. The trouble is it isn't one problem. I can have a board running all day here, the current build has just done 297 passes of GB6 overnight, then someone else plugs one in and it will hardly boot.
There are clearly other things going on as well. Even without TT-RAM there has been odd spiking on the reset line, which follows the ST bus buffers switching. I filtered it inside the CPLD so it doesn't reset the board, but I never got to the bottom of what causes it.
There are clearly other things going on as well. Even without TT-RAM there has been odd spiking on the reset line, which follows the ST bus buffers switching. I filtered it inside the CPLD so it doesn't reset the board, but I never got to the bottom of what causes it.
Re: REV 3 - REV 5 - The beginning (ST536)
A quick update on the area / density tests and a new SDRAM test.
First, the current working build with the fitter set to density, no code changes. It booted, but the IDE drive test screen didn't scroll properly and it reset a few tests into GB6. The SDRAM address logic actually stayed in one block this time, but the SDRAM state machine itself got slower instead.
Then the same density build with TT-RAM disabled, so the ROM copy fails and it runs from the ROM chips. That booted and the IDE screen was fine, so the basic machine works, it is the SDRAM side that tips it over.
Next I tried running the SDRAM state machine at half rate. The SDRAM clock and all the pin timings stay the same, but the state machine only moves every other 100MHz clock, always in step with the CPU clock. That gives all the internal SDRAM logic twice as long to settle, so in theory it shouldn't care so much when the fitter moves things around.
With the fitter set to density, that build wouldn't boot at all, even with TT-RAM disabled. With the fitter set back to speed it boots, TT-RAM tests fine, and GB6 is running now. The cost is about 10% on TT-RAM speed, 942% in GB6 against about 1050% before.
In the simulation, the half-rate version also gives the same data timing on every read, every time. The old controller's cache burst reads could occasionally land in the wrong place straight after a refresh in the simulation. That isn't proven on real hardware, but it is the sort of thing that would give a random crash every few hours.
So density hasn't helped on its own. Both density builds also had the same slow path on the ST bus side, which I suspect is what upset the IDE screen, so it just moves the problem somewhere else. I will keep testing the half-rate build and see how it holds up.
First, the current working build with the fitter set to density, no code changes. It booted, but the IDE drive test screen didn't scroll properly and it reset a few tests into GB6. The SDRAM address logic actually stayed in one block this time, but the SDRAM state machine itself got slower instead.
Then the same density build with TT-RAM disabled, so the ROM copy fails and it runs from the ROM chips. That booted and the IDE screen was fine, so the basic machine works, it is the SDRAM side that tips it over.
Next I tried running the SDRAM state machine at half rate. The SDRAM clock and all the pin timings stay the same, but the state machine only moves every other 100MHz clock, always in step with the CPU clock. That gives all the internal SDRAM logic twice as long to settle, so in theory it shouldn't care so much when the fitter moves things around.
With the fitter set to density, that build wouldn't boot at all, even with TT-RAM disabled. With the fitter set back to speed it boots, TT-RAM tests fine, and GB6 is running now. The cost is about 10% on TT-RAM speed, 942% in GB6 against about 1050% before.
In the simulation, the half-rate version also gives the same data timing on every read, every time. The old controller's cache burst reads could occasionally land in the wrong place straight after a refresh in the simulation. That isn't proven on real hardware, but it is the sort of thing that would give a random crash every few hours.
So density hasn't helped on its own. Both density builds also had the same slow path on the ST bus side, which I suspect is what upset the IDE screen, so it just moves the problem somewhere else. I will keep testing the half-rate build and see how it holds up.
Re: REV 3 - REV 5 - The beginning (ST536)
Is probably implicit in the above, but I'll mention it just in case you've missed it, but obviously the assertion of STERM can't run at half pace too else it'll be 20ns late.exxos wrote: Tue Nov 17, 2026 3:09 pm Next I tried running the SDRAM state machine at half rate. The SDRAM clock and all the pin timings stay the same, but the state machine only moves every other 100MHz clock, always in step with the CPU clock. That gives all the internal SDRAM logic twice as long to settle, so in theory it shouldn't care so much when the fitter moves things around.
BW
DFB1 Open source 50MHz 030 and TT-RAM accelerator for the Falcon
Smalliermouse ST-optimised USB mouse adapter based on SmallyMouse2
FrontBench The Frontier: Elite 2 intro as a benchmark
Smalliermouse ST-optimised USB mouse adapter based on SmallyMouse2
FrontBench The Frontier: Elite 2 intro as a benchmark
Re: REV 3 - REV 5 - The beginning (ST536)
This is the half-rate build. TT-RAM is about 10% slower than before because the SDRAM state machine now only moves every other 100MHz clock. STERM itself isn't slowed down, it still comes from the CPU clocked registers, so single reads take the same time, but writes end about one CPU clock later and cache burst reads lose a bit more, which is where most of the 10% goes.
I've given 2 more firmwares to @coonsgm to try out. One with no TTram to make sure his build is stable to start with and the slightly slower TTram build.
I'm still plotting a FPGA version. Need to build a test board as timings are not exactly given. But the china FPGA are dirt cheap, problem is they are 3.3V max input. But 3x 16bit IO buffers should solve that.
I've given 2 more firmwares to @coonsgm to try out. One with no TTram to make sure his build is stable to start with and the slightly slower TTram build.
I'm still plotting a FPGA version. Need to build a test board as timings are not exactly given. But the china FPGA are dirt cheap, problem is they are 3.3V max input. But 3x 16bit IO buffers should solve that.

