7 ms·

Performance of the 8088 on PC, PCjr and Tandy 1000

69 points by WoodenChair 8y ago

tssva 8y ago
The lower 128k on the PCjr is shared with the video controller. Contention caused by this dual access is what slows it down.
- rbanffy 8y ago
  I expected the Tandy 1000 to have similar performance - IIRC, their video modes are very similar.
  - tssva 8y ago
    Most of the Tandy 1000s included a DMA controller which the PCjr did not. This eliminated the shared memory issues the PCjr suffered from. Although similar Tandy graphics could utilize different memory blocks for video. This means that the Tandy was compatible with software written for the PCjr graphics system but some software written for the Tandy graphics system is not compatible with the PCjr.
    - spc476 8y ago
      The DMA issue was only a factor for the disk drive. While the PCjr graphics memory was shared with the main CPU memory, the Tandy graphics memory was separate in the same way that the EGA graphics memory was separate---it was never used for program code. In fact, the Tandy graphics was probably more compatible with the normal PC graphics than with the PCjr graphics. Yes, you could still access CGA graphics on the PCjr via $B800:0000 but the graphics chip rerouted the accesses to low memory were it was actually stored. But to use the 320x200x16 graphics mode (PCjr and Tandy supported this, CGA did not) on the PCjr, you needed to not only set the mode, but know the physical address (which could be anywhere in the lower 128K of RAM). When I wrote programs that used this mode, I had to use a custom MS-DOS configuration that set aside a block of low RAM for this mode. On this Tandy, you could avoid this and use the RAM set aside for graphics (starting at $A000:0000 if I recall correctly).
      - tssva 8y ago
        "the Tandy graphics memory was separate in the same way that the EGA graphics memory was separate---it was never used for program code." This is incorrect. The Tandy graphics adapter did share ram with the cpu. The service manual for the Tandy 1000 can be found at http://www.thealmightyguru.com/Wiki/images/7/7c/Tandy_1000_-_Manual_-_Service.pdf http://www.thealmightyguru.com/Wiki/images/7/7c/Tandy_1000_-.... Starting on page 40 is a good description of how this worked.
rbanffy 8y ago
On these machines performance was very consistent. They has memory without any cache, with memory parts with standard timings connected to a memory bus with the same timing as the CPU clock. Differences were in additional wait states introduced by other parts (such as the video on the PCjr) of the machine-specific architecture. It'd be nice to run a similar set of benchmarks on "almost compatible" machines such as the Sanyo MBC-555, the Zenith Z-100 or the Mindset.
klosor 8y ago
hmm some pretty good memes yes
theologic 8y ago
For context, when these machines were being sold, nobody even thought about performance in these terms. It is only later that we are benchmarking old machines to understand the performance gaps, and if you weren't there, you might think this was relevant at the time. Generally, in the mid 80s, you needed base performance (8088/8086) or "3 times as fast" performance (80286) when the AT came about. Compaq (now HP) became the real competitor to IBM PC/XT because the end user could understand that they ran at 8 MHz vs the PC 4.77 Mhz. This is as much performance issues that the users would want to know. Better performance without needing to buy an AT. Then the clones came about, and everybody started to put a "turbo" button on front of the clone PC. My first degree was in finance and accounting, but I fell in love with technology, so I sold PCs out of the biggest gray marketer in Seattle when the PC AT (second real computer for business) was coming out. I have first hand knowledge of the buying behavior because I was on the front lines of computer sales, and made a ton of good money selling $2-5K computers to people walking in off the street. The advent of the PC revolution was all about apps. I'm not saying that there weren't benchmarks, and maybe you'd find a savvy buyer that subscribed to Byte magazine, but these were far and few between. Tandy simply didn't have Lotus 1-2-3. The PC Junior wasn't a performance issue, but the term "Junior" just killed the thing out of the gate. The investment in a PC was a lot of money, and you weren't going to get a Junior, you were going to get a real PC. (For instance, the keyboard was much more impactful than any performance thoughts.) Finally, for those not around, storage was an incredible change agent. A lot of people think an SSD in a client platform is pretty cool. However, you cannot image the step up that a hard drive provided versus a floppy drive. The PC Junior and Tandy had no real hard drive path. While IBM had hard drive specific model (XT), the effort to install a hard drive in a PC was pretty small because of the accessibility. An open architecture, which any home builder takes for granted today, was pretty revolutionary at the time, and IBM's willingness to allow plug-ins (where they followed Woz) caused it to cremate the competition with a peripheral market.
- justin66 8y ago
  > I'm not saying that there weren't benchmarks, and maybe you'd find a savvy buyer that subscribed to Byte magazine, but these were far and few between. In 1982 Byte magazine had 420,000 subscribers. They might not have been the sort who bought from you and your store.
32bitkid 8y ago
I find stuff like this fascinating; I'm old enough to remember computers from this era, and—at least for me—I have a certain nostalgia for them. The first computer I was exposed to was an IBM PCjr, so seeing stuff like this 35 years after-the-fact is pretty mind-blowing, that people still care. I know that a lot of it is just my age, and I remember my father growing up used to work on/rebuild the engine on his '64 VW beetle. I think for very similar reasons: nostalgia, limited complexity, the tangibility of the "API". For me, POKEing a value in memory and seeing the screen change or a sound play is tangible. The opcodes of an 8088 are _just_ enough that I can feel like I understand it. Modern computers feel several orders of magnitude more complex and—admittedly—more powerful, but at a _cost_. I'm sure my dad thought the same about electric fuel-injected engines and the ever increasing complexity of "modern" cars. Currently, on my work-bench in the garage is a half-completed [Micro8088](https://github.com/skiselev/micro_8088 https://github.com/skiselev/micro_8088) and 8-bit ISA back-plane; still waiting for a few parts to come in. Its been a lot of fun to work on, and the intent is to get back into doing some retro-programming. (Hopefully, I have a better understanding of the underlying architecture now than I did when I spent a summer when I was 14 trying to figure out the relationship between AH/AL and AX without a concrete understanding of binary/hexadecimal math)
Shivetya 8y ago
Having owned all three of these; I grew up in an IBM house hold but later on I had two different Tandy 1000 series PCs (EX and the bastardized TX which had an 80286 on a XT board). We had a PC from early days complete with the large hard bound manuals, single sided diskettes, and text only. Now the regardless of processor the PCjr was seriously impacted in that that the keyboard interrupt was not there but actually had to be handled by the processor. See https://www.brutman.com/PCjr/pcjr_keyboard_handling.html https://www.brutman.com/PCjr/pcjr_keyboard_handling.html for a good description of what this incurred. (now for the rambling part) The Tandy was great for their 16 color hack and better sound which made games so much more enjoyable. (star flight being my favorite game ever and it used two 360k diskettes for a persistent environment) fwiw, even before dedicated graphics we had cool games. Even Wizardry which I had on my Apple was available for the PC and I played it with the ASCII character set being used to create the halls in 3D (think plus sign, bars, and minus signs). All the text based games many reminisce about are from this era
mysterydip 8y ago
I know on our PCjr, the first 128k was built in, and above that only worked when we had the memory expander card attached, so there is a location difference. The cynic in me says it was an administrative decision to show the jr as a lower cost lower performance option.
- amorphid 8y ago
  I had a base PCjr with 128K of RAM, one 360K floppy disk drive, and no hard drive. Copying disks took FOREVER. I was so jealous of my friend who had two disk drives on his PC.
magoon 8y ago
I remember the PCjr felt considerably slower than the 5150 (PC) and also had a terrible keyboard, however it was innovative to use infrared to make it wireless. It also had cartridge slots and a small catalog of (mostly) games that came on them, but in practice there was no benefit to that. The biggest problem I saw with Tandy was partial incompatibility with certain popular software.
- jonhendry18 8y ago
  "but in practice there was no benefit to that." Apparently the cartridge ROMs weren't afflicted by the wait states that applied to the first 128k of RAM, so software on the cartridges could actually run faster than software on disk.

69 points by WoodenChair 8y ago

The lower 128k on the PCjr is shared with the video controller. Contention caused by this dual access is what slows it down.

rbanffy 8y ago
I expected the Tandy 1000 to have similar performance - IIRC, their video modes are very similar.
- tssva 8y ago
  Most of the Tandy 1000s included a DMA controller which the PCjr did not. This eliminated the shared memory issues the PCjr suffered from. Although similar Tandy graphics could utilize different memory blocks for video. This means that the Tandy was compatible with software written for the PCjr graphics system but some software written for the Tandy graphics system is not compatible with the PCjr.
  - spc476 8y ago
    The DMA issue was only a factor for the disk drive. While the PCjr graphics memory was shared with the main CPU memory, the Tandy graphics memory was separate in the same way that the EGA graphics memory was separate---it was never used for program code. In fact, the Tandy graphics was probably more compatible with the normal PC graphics than with the PCjr graphics. Yes, you could still access CGA graphics on the PCjr via $B800:0000 but the graphics chip rerouted the accesses to low memory were it was actually stored. But to use the 320x200x16 graphics mode (PCjr and Tandy supported this, CGA did not) on the PCjr, you needed to not only set the mode, but know the physical address (which could be anywhere in the lower 128K of RAM). When I wrote programs that used this mode, I had to use a custom MS-DOS configuration that set aside a block of low RAM for this mode. On this Tandy, you could avoid this and use the RAM set aside for graphics (starting at $A000:0000 if I recall correctly).
    - tssva 8y ago
      "the Tandy graphics memory was separate in the same way that the EGA graphics memory was separate---it was never used for program code." This is incorrect. The Tandy graphics adapter did share ram with the cpu. The service manual for the Tandy 1000 can be found at http://www.thealmightyguru.com/Wiki/images/7/7c/Tandy_1000_-_Manual_-_Service.pdf http://www.thealmightyguru.com/Wiki/images/7/7c/Tandy_1000_-.... Starting on page 40 is a good description of how this worked.

rbanffy 8y ago

On these machines performance was very consistent. They has memory without any cache, with memory parts with standard timings connected to a memory bus with the same timing as the CPU clock. Differences were in additional wait states introduced by other parts (such as the video on the PCjr) of the machine-specific architecture. It'd be nice to run a similar set of benchmarks on "almost compatible" machines such as the Sanyo MBC-555, the Zenith Z-100 or the Mindset.

klosor 8y ago

hmm some pretty good memes yes

theologic 8y ago

For context, when these machines were being sold, nobody even thought about performance in these terms. It is only later that we are benchmarking old machines to understand the performance gaps, and if you weren't there, you might think this was relevant at the time. Generally, in the mid 80s, you needed base performance (8088/8086) or "3 times as fast" performance (80286) when the AT came about. Compaq (now HP) became the real competitor to IBM PC/XT because the end user could understand that they ran at 8 MHz vs the PC 4.77 Mhz. This is as much performance issues that the users would want to know. Better performance without needing to buy an AT. Then the clones came about, and everybody started to put a "turbo" button on front of the clone PC. My first degree was in finance and accounting, but I fell in love with technology, so I sold PCs out of the biggest gray marketer in Seattle when the PC AT (second real computer for business) was coming out. I have first hand knowledge of the buying behavior because I was on the front lines of computer sales, and made a ton of good money selling $2-5K computers to people walking in off the street. The advent of the PC revolution was all about apps. I'm not saying that there weren't benchmarks, and maybe you'd find a savvy buyer that subscribed to Byte magazine, but these were far and few between. Tandy simply didn't have Lotus 1-2-3. The PC Junior wasn't a performance issue, but the term "Junior" just killed the thing out of the gate. The investment in a PC was a lot of money, and you weren't going to get a Junior, you were going to get a real PC. (For instance, the keyboard was much more impactful than any performance thoughts.) Finally, for those not around, storage was an incredible change agent. A lot of people think an SSD in a client platform is pretty cool. However, you cannot image the step up that a hard drive provided versus a floppy drive. The PC Junior and Tandy had no real hard drive path. While IBM had hard drive specific model (XT), the effort to install a hard drive in a PC was pretty small because of the accessibility. An open architecture, which any home builder takes for granted today, was pretty revolutionary at the time, and IBM's willingness to allow plug-ins (where they followed Woz) caused it to cremate the competition with a peripheral market.

justin66 8y ago
> I'm not saying that there weren't benchmarks, and maybe you'd find a savvy buyer that subscribed to Byte magazine, but these were far and few between. In 1982 Byte magazine had 420,000 subscribers. They might not have been the sort who bought from you and your store.

32bitkid 8y ago

I find stuff like this fascinating; I'm old enough to remember computers from this era, and—at least for me—I have a certain nostalgia for them. The first computer I was exposed to was an IBM PCjr, so seeing stuff like this 35 years after-the-fact is pretty mind-blowing, that people still care. I know that a lot of it is just my age, and I remember my father growing up used to work on/rebuild the engine on his '64 VW beetle. I think for very similar reasons: nostalgia, limited complexity, the tangibility of the "API". For me, POKEing a value in memory and seeing the screen change or a sound play is tangible. The opcodes of an 8088 are _just_ enough that I can feel like I understand it. Modern computers feel several orders of magnitude more complex and—admittedly—more powerful, but at a _cost_. I'm sure my dad thought the same about electric fuel-injected engines and the ever increasing complexity of "modern" cars. Currently, on my work-bench in the garage is a half-completed [Micro8088](https://github.com/skiselev/micro_8088 https://github.com/skiselev/micro_8088) and 8-bit ISA back-plane; still waiting for a few parts to come in. Its been a lot of fun to work on, and the intent is to get back into doing some retro-programming. (Hopefully, I have a better understanding of the underlying architecture now than I did when I spent a summer when I was 14 trying to figure out the relationship between AH/AL and AX without a concrete understanding of binary/hexadecimal math)

Shivetya 8y ago

Having owned all three of these; I grew up in an IBM house hold but later on I had two different Tandy 1000 series PCs (EX and the bastardized TX which had an 80286 on a XT board). We had a PC from early days complete with the large hard bound manuals, single sided diskettes, and text only. Now the regardless of processor the PCjr was seriously impacted in that that the keyboard interrupt was not there but actually had to be handled by the processor. See https://www.brutman.com/PCjr/pcjr_keyboard_handling.html https://www.brutman.com/PCjr/pcjr_keyboard_handling.html for a good description of what this incurred. (now for the rambling part) The Tandy was great for their 16 color hack and better sound which made games so much more enjoyable. (star flight being my favorite game ever and it used two 360k diskettes for a persistent environment) fwiw, even before dedicated graphics we had cool games. Even Wizardry which I had on my Apple was available for the PC and I played it with the ASCII character set being used to create the halls in 3D (think plus sign, bars, and minus signs). All the text based games many reminisce about are from this era

mysterydip 8y ago

I know on our PCjr, the first 128k was built in, and above that only worked when we had the memory expander card attached, so there is a location difference. The cynic in me says it was an administrative decision to show the jr as a lower cost lower performance option.

amorphid 8y ago
I had a base PCjr with 128K of RAM, one 360K floppy disk drive, and no hard drive. Copying disks took FOREVER. I was so jealous of my friend who had two disk drives on his PC.

magoon 8y ago

I remember the PCjr felt considerably slower than the 5150 (PC) and also had a terrible keyboard, however it was innovative to use infrared to make it wireless. It also had cartridge slots and a small catalog of (mostly) games that came on them, but in practice there was no benefit to that. The biggest problem I saw with Tandy was partial incompatibility with certain popular software.

jonhendry18 8y ago
"but in practice there was no benefit to that." Apparently the cartridge ROMs weren't afflicted by the wait states that applied to the first 128k of RAM, so software on the cartridges could actually run faster than software on disk.