• Please review our updated Terms and Rules here

Clock Doubler Circuit for 8088 to use 16 Mhz V20

chjmartin2

chjmartin2

Experienced Member
Joined
Dec 26, 2012
Messages
440
Hi,

Clearly, I am not technically capable enough, but that never stops me from wondering. What do you suppose would be involved in making a clock doubler circuit with cache to use a 16 MHz V20 in an 8 MHz (or even 9.54 Mhz) 8088? Can it work like a 486 DX/4 or something like that. In my simple mind you need a clock doubler (frankly I want a 1.6771488X so it works on 9.54 Mhz capable boards), some cache to allow the processors to go faster than it can send the info back. My challenge is - I wouldn't even know where to start. I'm pretty sure I could research how to use the CLK signal to give me a 16 MHz signal (that's just research) but the cache - total mystery. I'm pretty sure it wouldn't work at all if you just 1.6777X the clock and wired everything else up. Honestly, I would patron somebody if they wanted to take this on as I want it to exist in the worst way.

Thanks,

Chris
 
Short answer: The PC is a synchronous device and "clock doubling" can be pretty complicated, not to mention support chip speed issues. Why not get one of the off-the-shelf accelerators?
 
Chuck(G) said:
Short answer: The PC is a synchronous device and "clock doubling" can be pretty complicated, not to mention support chip speed issues. Why not get one of the off-the-shelf accelerators?
Click to expand...
I have a breakthrough 286 although I think I may have bricked it, going to put it in a known compatible system to confirm. Regardless that is a 286 for me the V20 16 Mhz is still an 8088. I also don't want to use a slot for it, would rather see it exist as a device to fit into the 8088 CPU slot. Isn't just for speed - could just build a 486 or more for that!
 
Chuck(G) said:
On an 8088-bus system, I'm not sure that cache would buy you much. DRAM currently is more than fast enough to operate at zero wait states.
Click to expand...
See, this is where I am just not smart enough. My understanding is that in a DX/4 or something the speed to the motherboard is like 25 Mhz but then the processor runs at 100 MHz. Am I thinking about it the wrong way that the CPU to the board could be 9.54 MHz but the processor could run at 16 MHz or is that just the same thing and you wouldn't get any speed benefit from it?
 
The PC Sprint modification from ages ago does this. The motherboard continues to run at the standard 4.77Mhz speed. An 8284A is added to drive a faster CPU, usually at 2x the clock speed. The system memory is still running at 4.77Mhz so extra memory wait states are need for the CPU, but anything that is in the prefetch buffer of the CPU runs faster, and of course the CPU is faster.

It was one of the first homebrew speedups for PCs.

 
mbbrutman said:
The PC Sprint modification from ages ago does this. The motherboard continues to run at the standard 4.77Mhz speed. An 8284A is added to drive a faster CPU, usually at 2x the clock speed. The system memory is still running at 4.77Mhz so extra memory wait states are need for the CPU, but anything that is in the prefetch buffer of the CPU runs faster, and of course the CPU is faster.

It was one of the first homebrew speedups for PCs.

Click to expand...
That's not correct, the PC-Sprint speeds up the memory and DMA as well. No additional waits are inserted
 
Last edited:
Chuck(G) said:
On an 8088-bus system, I'm not sure that cache would buy you much. DRAM currently is more than fast enough to operate at zero wait states.
Click to expand...
But that is the whole point of the cache. A faster CPU would not be faster if it still had to use the RAM over the slow bus. So there is cache in between to allow faster operation at all.
 
maxtherabbit said:
That's not correct, the PC-Sprint speeds up the memory and DMA as well. No additional waits are inserted
Click to expand...

The version of the PC Sprint that I'm familiar with (designed by Doug Severson) only speeds the CPU and not the rest of the system. From the original documentation:

"PC-SPRINT attempts to solve these problems by running two crystals simultaneously (the original and a higher frequency). This allows changing only the processors clock, while keeping all other frequencies constant. Obviously, the processor must be replaced with a higher speed version."

Are you referring to a different speed modification?

Edit: I think I see the confusion now, and yes, you are right. There is definitely another clock being added there, but I'm trying to work out how the new faster clock is only affecting the CPU and memory and not the other parts of the system. It seems like upgrading the CPU to a faster CPU while trying to overclock everything else 50% shouldn't work too well.
 
Last edited:
Timo W. said:
But that is the whole point of the cache. A faster CPU would not be faster if it still had to use the RAM over the slow bus. So there is cache in between to allow faster operation at all.
Click to expand...
What slow bus? The RAM is on the planar, not on the ISA bus, so I'm not following you. I can see that for an 80386, but an 8088? I doubt that the RAS/CAS shuffle would matter to one.
 
mbbrutman said:
The PC Sprint modification from ages ago does this. The motherboard continues to run at the standard 4.77Mhz speed. An 8284A is added to drive a faster CPU, usually at 2x the clock speed. The system memory is still running at 4.77Mhz so extra memory wait states are need for the CPU, but anything that is in the prefetch buffer of the CPU runs faster, and of course the CPU is faster.

It was one of the first homebrew speedups for PCs.

Click to expand...
Hi,

I have a Faraday FE2010A - so not sure we could just go the PC-Sprint route. The Faraday FE2010A doesn't seem to have a CSYNC. So help me with this thinking:

Create a board with a 8284A using a 48 MHz crystal to use as an input. Wire the output of the FE2010A to the CSYNC on the 8284A. Use the 8284A to drive the CPU clock, then wire everything else standard to the 8088 Port. Only thing I can't follow versus PC Sprint is how he removed the 8284A in order to get to the appropriate connections. I'm not sure what those connections are. Would the DRAM be able to keep up or would you isolate the speed of the DRAM? If an 8284A could output a 16 MHz useful signal to clock a V20 then we may be in business no?

Chris
 
Chuck(G) said:
Short answer: The PC is a synchronous device and "clock doubling" can be pretty complicated, not to mention support chip speed issues. Why not get one of the off-the-shelf accelerators?
Click to expand...
I know about such a circuit, but it was for Motorola 68000. I have one ATARI ST with that accelerator in. The circuit observes the bus signals of the CPU and while the CPU is not accessing adresses, what means, it is in internal operations like calculating, then that logics doubles the clock from 8 to 16 MHz. As this simple accelerator has no cache, the total speedup is very limited, just about 20-30 % faster than original. For an XT it would be much easier than designing such logics to swap the 8088 / 8086 to a NEC V20 / V30.
 
You're missing the big picture. On a system designed to run at 4.77MHz clock with a turbo mode to 8MHz, running a 16MHz CPU at full speed would spend a fair amount of time waiting for memory accesses. Consider also, many device drivers, including the floppy driver in the BIOS used CPU loops for timing. You're not going to get the full performance out of a 16MHz CPU with the current design.
 
Yes, true, it only will speed up operations inside the CPU. So for example the time to add two registers and put the result in another register takes half the time. But register to memory or vice versa will not speed up.

80x86 Integer Instruction Set (8088 - Pentium)

Integer Instruction Set (8088 - Pentium.
www2.math.uni-wuppertal.de

There are some examples like the AAD and AAM instruction which need 60 or 83 clock cycles. They still last 60 or 83 clock cycles, but in half time.
 
1ST1 said:
Yes, true, it only will speed up operations inside the CPU. So for example the time to add two registers and put the result in another register takes half the time. But register to memory or vice versa will not speed up.

80x86 Integer Instruction Set (8088 - Pentium)

Integer Instruction Set (8088 - Pentium.
www2.math.uni-wuppertal.de

There are some examples like the AAD and AAM instruction which need 60 or 83 clock cycles. They still last 60 or 83 clock cycles, but in half time.
Click to expand...
Ok but wouldn’t that still speed up the system? What about clocking the memory higher and just upgrading that?
 
You must log in or register to reply here.
Back
Top