Tandy 1000 SL overclocking

[CarlosTex]

Hey there fellas,

i recently got a loose Tandy 1000 SL motherboard, including both satellite boards. I've always wanted a Tandy 1000 system. The Tandy i'm really after is the TL/3, but these are a little hard to find. Let alone the cost of shipping one from the US to Europe. Canada is even worse...

I've always wanted a Tandy system and i also wanted a 8086/V30 system. Generic 8086 XT boards are extremely hard to find, so i thought i would kill 2 birds with one stone. However an 8MHz V30 is a very small improvement on a 12MHz V20, and since the desired speed for a V30 system for me would be 10MHz i'm wondering if it's easy to replace the crystal on a Tandy 1000 SL.

From the pictures i can see the board has 2 crystals, although i can't see the values, but i assume the 2 crystals at least gurantee one is used for the 14.318MHz ISA bus / timer etc... and the other for the CPU.

I assume the CPU is using a 24MHz crystal which will give 8MHz divided by 3, so i'm wondering if changing to 30MHz will overclock the CPU to 10MHz without affecting the system to a point of instability. Usually 10MHz is well tolerated.


Has anyone tried to overclock a Tandy 1000 SL? If so, what were the results?

 

[CarlosTex]

OK, i found the tech manual for the Tandy 1000 SL.

It seems to be a dead end to try and switch the crystal. It seems that the 24MHz clock goes into the IC 8079024 (U41), chip which generates the CPU CLK and others. This might now be a good idea, but it would still be fun to try it out. It will probably muck up the locks for other stuff.

 

[Agent Orange]

Hey there fellas,

i recently got a loose Tandy 1000 SL motherboard, including both satellite boards. I've always wanted a Tandy 1000 system. The Tandy i'm really after is the TL/3, but these are a little hard to find. Let alone the cost of shipping one from the US to Europe. Canada is even worse...

I've always wanted a Tandy system and i also wanted a 8086/V30 system. Generic 8086 XT boards are extremely hard to find, so i thought i would kill 2 birds with one stone. However an 8MHz V30 is a very small improvement on a 12MHz V20, and since the desired speed for a V30 system for me would be 10MHz i'm wondering if it's easy to replace the crystal on a Tandy 1000 SL.

From the pictures i can see the board has 2 crystals, although i can't see the values, but i assume the 2 crystals at least gurantee one is used for the 14.318MHz ISA bus / timer etc... and the other for the CPU.

I assume the CPU is using a 24MHz crystal which will give 8MHz divided by 3, so i'm wondering if changing to 30MHz will overclock the CPU to 10MHz without affecting the system to a point of instability. Usually 10MHz is well tolerated.


Has anyone tried to overclock a Tandy 1000 SL? If so, what were the results?

Might be a neat experiment, but I doubt that you'll see much of a gain in performance. Take into consideration that when you OC a modern PC, you're working with GHz and the performance increase can be considerable. I have a 486 133 that is OC'd to 150 MHz, and you can realise a small increase in performance courtesy of some mobo settings and slightly increased chip voltage. No xtal mods involved. About the only time you're going to see a major difference would be in gaming or number crunching, and the 8086 pretty much handles most of the period software fairly well. Back in the day, the mods mainly consisted of adding a 386 or for 486 adapter to your 286 system. I don't recall too much in the way of hot-rodding the 8086 other than the V30, as the CPU architecture was rapidly changing in favor of the 386/486 and Pentium/Athlon. Anyway, good luck and I hope it works out for you.

 

[wesleyfurr]

I did luck into acquiring a TL/3...I have never seen a PC boot so fast (it has DOS in ROM). Been a while since I messed with it, but as I recall, it was at a command prompt before the hard drive had fully spun up. Drive was flaky and took a couple of power cycles before it came to life...which made it slow down and wait a bit at boot time.

As to the original question, not something I have considered on the SL. As picky as it seems to be with certain upgrades installed, I would be afraid to mess with it.

Wesley

 

[keenerb]

I did luck into acquiring a TL/3...I have never seen a PC boot so fast (it has DOS in ROM). Been a while since I messed with it, but as I recall, it was at a command prompt before the hard drive had fully spun up. Drive was flaky and took a couple of power cycles before it came to life...which made it slow down and wait a bit at boot time.

As to the original question, not something I have considered on the SL. As picky as it seems to be with certain upgrades installed, I would be afraid to mess with it.

Wesley

That's my favorite part of the boot-from-ROM Tandy machines. Power switch, boom you're ready to go. There's not actually much benefit to DOS later than 3.3 (or 2.11 for the HX/EX machines) supposedly unless you've got an EMS/XMS/UMB board installed...

 

[CarlosTex]

So it seems the Tandy 1000 SL (and probably other models too) use some sort of dual frequency crystal oscillators. I never seen these in other machines. What i did see in the Tandy 1000 SL reference manual is that there is the FFE9 port which comprises of a single byte. The individual bits can be set to set memory wait states, and most interesting of all is that bit 8 is OSCIN select which allows to select from 24MHz or 28.6MHz. Could this be used to make the CPU work at 9.54MHz?

A V30 at 9.54MHz instead of only 8MHz would provide a decent speed boost on these types of machines and would make it IBM AT (first revision) speed equivalent.

Other option would be replacing the 16MHz - 24MHz Oscillator by a 16 - 30Mhz one, it would be pretty easy for me to desolder the Osci and install a socket with my newly acquired Hakko FR300 desoldering vacuum gun, but i can't find these dual MHz oscillators anywhere, or maybe i'm not looking properly.

Anyone of the gurus has an opinion on this?

 

[Cloudschatze]

So it seems the Tandy 1000 SL (and probably other models too) use some sort of dual frequency crystal oscillators. I never seen these in other machines. What i did see in the Tandy 1000 SL reference manual is that there is the FFE9 port which comprises of a single byte. The individual bits can be set to set memory wait states, and most interesting of all is that bit 8 is OSCIN select which allows to select from 24MHz or 28.6MHz. Could this be used to make the CPU work at 9.54MHz?

It's funny that you're asking this, as I've spent the last several days investigating what FFE9, bit 7 actually does. The Technical Reference is vague about its purpose, merely stating that:

BIT 7: Must be 0 when OSCIN is equal to 28.63636MHz
Must be 1 when OSCIN is equal to 24 MHz (default)

I'd mentioned the benefits of changing the wait-states via FFE9 a few years ago in the following post (and have been running with FFE9 set to 0x80 for some time besides), but had assumed that bit 7 was best left alone, unless the Buffer Blue chip was actually supplied with the 28.63636 MHz OSC signal.
http://www.vcfed.org/forum/showthread.php?47587-8086-vs-V30-in-the-Tandy-1000-SL

Fast-forward to last Saturday, when I decided to flip that bit to 0, while the OSCIN supplied to the Buffer Blue IC yet remains the standard 24 MHz.

The result? A slight improvement:

FFE9 0x80:

[UID7D9417F3E]
MemoryTest=1184
OpcodeTest=594
VidramTest=603
MemEATest=743
3DGameTest=570
Score=13
CPU=NEC V30
CPUspeed=8 MHz
BIOSinfo=unknown
BIOSdate=19880622
BIOSCRC16=7D94
VideoSystem=VGA
VideoAdapter=VGA, Tseng ET 4000, DRAM, 1024kb Video Memory
Machine=Tandy 1000 SL
Description= 
Submitter=Cloudschatze

FFE9 0x00:

[UID7D9413C14]
MemoryTest=1177
OpcodeTest=587
VidramTest=598
MemEATest=739
3DGameTest=568
Score=14
CPU=NEC V30
CPUspeed=8 MHz
BIOSinfo=unknown
BIOSdate=19880622
BIOSCRC16=7D94
VideoSystem=VGA
VideoAdapter=VGA, Tseng ET 4000, DRAM, 1024kb Video Memory
Machine=Tandy 1000 SL
Description= 
Submitter=Cloudschatze

So, what does FFE9 Bit 7 actually do? I suspect it may alter the DRAM refresh, as a means of compensating for the timing difference between the two "supported" OSCIN signals. This is just a guess though, as even DRAM refresh is a bit of a black-box with the SL. (Unlike other Tandy 1000 systems, including the TL-series, the DMA controller is integrated into the SL's Buffer Blue IC, where DRAM refresh isn't driven by the output from PIT Channel 1 - which isn't even connected in the SL!).

In any event, supplying the Buffer Blue IC with the 28.63636 MHz signal needed to drive the CPU at 9.54 MHz is next on my to-do list, having also just swapped the 120ns and 150ns DRAMs with 100ns versions. Oh, boy.

TL:DR - FFE9 Bit 7 alone won't drive the CPU at 9.54 MHz.

Other option would be replacing the 16MHz - 24MHz Oscillator by a 16 - 30Mhz one, it would be pretty easy for me to desolder the Osci and install a socket with my newly acquired Hakko FR300 desoldering vacuum gun, but i can't find these dual MHz oscillators anywhere, or maybe i'm not looking properly.

There's already a 28.63636 MHz OSC signal on the mainboard, used by the video controller. The route I'm going to take is to simply lift the Buffer Blue OSCIN pin and patch into that.

 

[vwestlife]

That's my favorite part of the boot-from-ROM Tandy machines. Power switch, boom you're ready to go.

It takes only 2.2 seconds for a 1000RL to cold-boot to DOS in ROM:

 

[CarlosTex]

It's funny that you're asking this, as I've spent the last several days investigating what FFE9, bit 7 actually does. The Technical Reference is vague about its purpose, merely stating that:



I'd mentioned the benefits of changing the wait-states via FFE9 a few years ago in the following post (and have been running with FFE9 set to 0x80 for some time besides), but had assumed that bit 7 was best left alone, unless the Buffer Blue chip was actually supplied with the 28.63636 MHz OSC signal.
http://www.vcfed.org/forum/showthread.php?47587-8086-vs-V30-in-the-Tandy-1000-SL

Fast-forward to last Saturday, when I decided to flip that bit to 0, while the OSCIN supplied to the Buffer Blue IC yet remains the standard 24 MHz.

The result? A slight improvement:

FFE9 0x80:

[UID7D9417F3E]
MemoryTest=1184
OpcodeTest=594
VidramTest=603
MemEATest=743
3DGameTest=570
Score=13
CPU=NEC V30
CPUspeed=8 MHz
BIOSinfo=unknown
BIOSdate=19880622
BIOSCRC16=7D94
VideoSystem=VGA
VideoAdapter=VGA, Tseng ET 4000, DRAM, 1024kb Video Memory
Machine=Tandy 1000 SL
Description= 
Submitter=Cloudschatze

FFE9 0x00:

[UID7D9413C14]
MemoryTest=1177
OpcodeTest=587
VidramTest=598
MemEATest=739
3DGameTest=568
Score=14
CPU=NEC V30
CPUspeed=8 MHz
BIOSinfo=unknown
BIOSdate=19880622
BIOSCRC16=7D94
VideoSystem=VGA
VideoAdapter=VGA, Tseng ET 4000, DRAM, 1024kb Video Memory
Machine=Tandy 1000 SL
Description= 
Submitter=Cloudschatze

So, what does FFE9 Bit 7 actually do? I suspect it may alter the DRAM refresh, as a means of compensating for the timing difference between the two "supported" OSCIN signals. This is just a guess though, as even DRAM refresh is a bit of a black-box with the SL. (Unlike other Tandy 1000 systems, including the TL-series, the DMA controller is integrated into the SL's Buffer Blue IC, where DRAM refresh isn't driven by the output from PIT Channel 1 - which isn't even connected in the SL!).

In any event, supplying the Buffer Blue IC with the 28.63636 MHz signal needed to drive the CPU at 9.54 MHz is next on my to-do list, having also just swapped the 120ns and 150ns DRAMs with 100ns versions. Oh, boy.

TL:DR - FFE9 Bit 7 alone won't drive the CPU at 9.54 MHz.


There's already a 28.63636 MHz OSC signal on the mainboard, used by the video controller. The route I'm going to take is to simply lift the Buffer Blue OSCIN pin and patch into that.

It's kind of a risky mod for me to do considering Tandy hardware is practically non existant on my side of the pond and it gets pretty expensive importing from the US, specially via eBay.

I feel the urge to do it though, even though i do not have an hot air rework station, i might get a friend to help me lift the pin without damaging the pad. Easiest way to do it is to apply some fresh solder on the pin and lift the pin with hot air.

But there's another question. The other oscillator is a 32.5 - 28.6 Mhz dual frequency oscillator. One must make sure to tap the right pin. Whether you tap one or the other frequency both are very supported by a V30. 32.5Mhz will give a clock speed only slightly superior to 10Mhz which would make such system to surpass an IBM AT 6Mhz and almost into Tandy 1000 TL territory.

I ask you to keep in touch and publish your findings in here.

 

[CarlosTex]

I've took a look into the Tandy 1000 SL manual and into the schematics and there seems to be a 1/8W 10ohm resistor between the oscillator 24MHz output and the OSCIN in the buffer blue chip. The buffer blue technical info states that the OSCIN accepts a 30Mhz signal max at 50% off duty cycle. This doesn't seem to be a very complicated mod, except the buffer blue custom IC is a 100 pin quad flat type package and the pin pitch seems pretty small, so it feels a bit risky, there's not much space to work in there.

It's not clear where pin 1 actually is, and by the schematic pins seem to be all over the place, don't seem to go in order.

 

[keenerb]

I've took a look into the Tandy 1000 SL manual and into the schematics and there seems to be a 1/8W 10ohm resistor between the oscillator 24MHz output and the OSCIN in the buffer blue chip. The buffer blue technical info states that the OSCIN accepts a 30Mhz signal max at 50% off duty cycle. This doesn't seem to be a very complicated mod, except the buffer blue custom IC is a 100 pin quad flat type package and the pin pitch seems pretty small, so it feels a bit risky, there's not much space to work in there.

It's not clear where pin 1 actually is, and by the schematic pins seem to be all over the place, don't seem to go in order.

You could try finding a spot to sever the trace on the board and splice into it. Still quite risky though.

 

[CarlosTex]

You could try finding a spot to sever the trace on the board and splice into it. Still quite risky though.

So... I've checked Tandy 1000 SL manual, and that trace that goes under C80 and connects to R24 is the actual 24MHz output from the crystal oscillator. So i could cut that trace indeed, and then do 1 of 3 things:

1- Get some thin gauge wire and link the 28.63636MHz signal from the second Osci on the board and connect it to R24. Theoretically that would give me a 9.54MHz CPU clock.

2- I could get a new 30MHz OSCI, provide it with 5V and GND signals from the board and tap the output to R24. That should give me a 10MHz clock.

3- Same as above but with a 32MHz OSCI instead. Problem is that the Buffer Blue could crap out with a 32MHz signal. The SLtech manual states clearly for the OSCIN pin: Clock input, 30 MHz max, 50 % duty cycle.


I don't think there's much risk on killing the board regarding cutting of a trace, since it will be easy to patch it back in case i want to revert the mod. Anyone thinks the Buffer Blue chip could be screwed?

 

[CarlosTex]

Well i actually did this today. I ran a wire from the 28.6 MHz oscillator directly to the R24 resistor. I cut the trace between the 24MHz clock output to R24.

It seems to be working although it hasn't been all smooth sailing. I actually completed a lot of benchmarks, and the CPU speed seems to be properly picked up. The speed bump ia actually quite noticeable with CheckIt 3.0 reporting a Dhrystones result of 1195. TopBench gives me a score of 15 which is quite a boost compared to a normal Tandy 1000 SL.

Well all text mode programs seem to run fine, the problem is when i try to run a game i can't actually run anything, . The machine seems to be unable to change to graphic modes, so apparently the video chip does not like to share the signal with the CPU. Tried to run a lot of games but ONLY text mode games would work.

So i'm gonna change my approach. I'll unpatch that wire and use a separate oscillator for the CPU clock. Problem is that at the moment i only have 24 and 32MHz oscillator available, which one is the actual default clock and the other goes over the blue buffer chip max clock input by 2MHz.

If the Buffer Blue accepts the 32MHz signal without crapping out the CPU will effectly be overclocked to a whopping 10.6 MHz which would put the system close to IBM AT territory.

I haven't screwed around with the 0 waitstate stuff yet, but a 10.6Mhz V30 with 0 waitstates could be close to a pinnacle of performance by an XT Class system.

 

[CarlosTex]

OK so i made the mod something else. I'm now using an external OSCI to provide the CPU clock. I tried the 24MHz one and it runs like before at 8 MHz. As is suspected the system refuses to boot with the 32MHz crystal which clearly craps the buffer blue chip out. I don't have a 30MHz one to try, i'll have to order one.

Now i remember that when i first tried the motherboard some games wouldn't run either, so i don't think this has nothing to do with the mod i'm doing. I tried some more games and i ran Starflight, California Games, Digger and a few others, but others still don't work. I'm wondering if its because i'm using a CF card with 6.22 from my other V20 system. This card works fine with my XT, but i'm wondering if it is causing any problems. I noticed when trying to launch games the led activity on the card reader would just stop. I don't know why this only happens in games though.

I'm gonna try this again by formatting another card with DOS 3.3. Maybe this will work better with the Tandy system. I also noticed that my BIOS is version 1.04.00, is there any new version available?


EDIT: Just noticed something else... When trying to run DeskMate it actually freezes...unless i remove the lo tech xt ide. I tried in both C800 and D800 and result still the same, Deskmate just hangs at TC logo. If i remove it runs fine.

So it seems this system doesn't like lo tech XT-ide much.

 

[keenerb]

OK so i made the mod something else. I'm now using an external OSCI to provide the CPU clock. I tried the 24MHz one and it runs like before at 8 MHz. As is suspected the system refuses to boot with the 32MHz crystal which clearly craps the buffer blue chip out. I don't have a 30MHz one to try, i'll have to order one.

Now i remember that when i first tried the motherboard some games wouldn't run either, so i don't think this has nothing to do with the mod i'm doing. I tried some more games and i ran Starflight, California Games, Digger and a few others, but others still don't work. I'm wondering if its because i'm using a CF card with 6.22 from my other V20 system. This card works fine with my XT, but i'm wondering if it is causing any problems. I noticed when trying to launch games the led activity on the card reader would just stop. I don't know why this only happens in games though.

I'm gonna try this again by formatting another card with DOS 3.3. Maybe this will work better with the Tandy system. I also noticed that my BIOS is version 1.04.00, is there any new version available?


EDIT: Just noticed something else... When trying to run DeskMate it actually freezes...unless i remove the lo tech xt ide. I tried in both C800 and D800 and result still the same, Deskmate just hangs at TC logo. If i remove it runs fine.

So it seems this system doesn't like lo tech XT-ide much.

My XT-IDE cf adapter is actually running at D000, there is an offset ROM floating around that addressed some issues with RL systems that might apply to your SL as well.

 

[Krille]

So it seems this system doesn't like lo tech XT-ide much.

Is XTIDE Universal BIOS configured to run in "Full mode"? If so, you probably need to increase the amount of "stolen" RAM. This is from the manual (you did RTFM didn't you? );

kiB to steal from RAM [default=1] This menu item will appear only when "Full operating mode" is enabled. Leave it at the default unless you need to enable "Full operating mode" on Tandy 1000 models with 640k or less RAM. Setting this to 33 (almost always enough) or 65 (always enough) will reserve the top of RAM to Tandy video circuitry in addition to the XTIDE Universal BIOS variables thus avoiding a conflict between the two.

 

[CarlosTex]

Thanks Krille, i'll try that when i get a chance. I'm using the binaries from James Pearce lo tech site, i don't recall exactly how they are configured.

I'm still waiting for the 30MHz OSCI to arrive, and i want to replace all the RAM on the board for 100ns versions. Mucking with port FFE9h for 0 waitstate RAM will yet bring another small boost.

 

[CarlosTex]

Some pics of the attempted mods. I also successfully was able to have High Density Drives working in the system. By looking at the pics it is quite visible what i've done to achieve that.

The pure 10MHz speed i'm aiming for won't be a big performance boost vs 9.54MHz, but should be noticeable. I suspect i should achieve around 1250 Dhrystones in Checkit benchmark, and with 0 wait state memory, memory operations should also see a good boost which could help improve the overall system speed.
The simple 9.54MHz mod vs the regular 8MHz operation showed a significant boost in performance, which was immediately noticeable with something as simple as executing a dir command in DOS. Not as fast as a TL or a TL/2, but close.

I often wonder what a 16MHz (NEC V30HL) could do. It would probably give 12MHz 286 systems a run for their money. Unfortunately, a 32MHz OSCI didn't work with the Tandy motherboard as expected, as the technical reference manual makes it clear that 30MHz is the max the buffer blue chip can take.

 

[CarlosTex]

My gglabs CGA2RGB converter arrived this week, so i had another reason to take the Tandy motherboard out of the box and test it with a 15KHz VGA monitor. Unfortunately the damn thing is dead, so it was 35USD thrown in the bin.

Just for fun i ran a wire from the 28.6MHz crystal on the board to overclock the CPU again, and in no time problems started to come back again. Like before i started to witness some text corruption in graphics modes. I'll post some pictures later, but it really seems that it is not a good idea to connect the clock that drives the video chip to also feed the CPU clock. Either that wire is introducing a lot of noise to the circuits mucking everything up, and/or the memory can't handle the CPU speed. I'm suspicious of this because the RAM on the board is an assortment of different brand 46464 DRAM, with half being rated for 150ns and others for 120ns. This could eventually explain the graphics corruption and occasional system hang.

I think i might have to replace the serial EEPROM. Every now and then feels like settings are somehow changed between power cycles. This was driving me nuts, leading to a point where a fixed disk couldn't be booted and DOS in ROM would be default again. I had to remove the Zilog floppy disk controller so that it would force serching for fixed disks so i could run SETUPSL. When i ran SETUPSL in advanced mode i noticed that RAM was set for 3 waitstates, which explains why the system was so slow in memory operations when i benchmarked it.

So eventually i fixed the severed trace, restored 8MHz operation, and soon enough everything came back to normal. I already ordered some NEC 46464 RAM rated for 100ns. When the 30MHz oscillator arrives i will try it externally with a socket, get 5V and GND from the power supply and feed the output to the board. If that still results in graphics corruption and weird behaviour, then it will be clear that the RAM just can't handle the speed.

 

[Trixter]

Unfortunately the damn thing is dead, so it was 35USD thrown in the bin.

Which is "the damn thing" that is dead? The CGA2RGB, the Tandy motherboard, or the 15KHz VGA monitor?