• Please review our updated Terms and Rules here

XT Turbo board problem

That's interesting. I always found it odd that Norton was the ONLY program that could properly identify my 15MHz V30 as a 15MHz V30. Well, now I know the real reason: pure coincidence!
 
I pulled out some old JDR Microdevices catalogs, and here are the Norton SI 5.0 benchmarks they have listed for the various motherboards they offered ( * indicates the presence of onboard L2 cache):

8088-10: 2.1
286-12: 9.2
386SX-20: 10.0 (?)
386SX-16: 11.1
286-16: 11.8
386SX-33: 15.3
386DX-25: 15.6
386DX-33*: 35.7
386DX-40*: 43.1
486SX-20*: 54.0 (386/486 combo board)
486DX-33*: 71.3 (ISA)
486DX-33*: 72 (VLB)
486DX/2-66*: 92 (ISA)
486DX/2-66*: 99 (VLB)
486DX-50*: 108 (EISA)

They also gave Landmark 2.0 benchmark results, but Landmark appears to be a more simplistic clock speed test because it shows a 486DX/2-66 being drastically faster than a 486DX-50, while Norton SI actually shows the DX-50 being faster than the DX/2-66 due to its higher bus speed.

Also interesting is how easily a fast 286 can match or even beat a low-end 386. It makes sense, because when running real-mode code without any 386-specific optimizations (and without an L2 cache), the 386 is essentially equivalent to a 286 running at the same clock speed.
 
They also gave Landmark 2.0 benchmark results, but Landmark appears to be a more simplistic clock speed test because it shows a 486DX/2-66 being drastically faster than a 486DX-50, while Norton SI actually shows the DX-50 being faster than the DX/2-66 due to its higher bus speed.

But a DX-50 has a bus speed of 25MHz, not 33... Just like the DX4-100's are so named because they have, again, a 25MHz bus. Confused.

Also interesting is how easily a fast 286 can match or even beat a low-end 386. It makes sense, because when running real-mode code without any 386-specific optimizations (and without an L2 cache), the 386 is essentially equivalent to a 286 running at the same clock speed.

Clock for clock, a 386 is still faster. Nearly every opcode runs in less cycles, plus a DX has a 32-bit bus. But I have seen a 286-20 beat a 386-16sx, because both were on equal footing (386sx uses 16-bit bus just like 286).

I am beginning to think that there is slowly increasing demand for vintage PC benchmarking software that doesn't utterly suck. Landmark, Norton, are all quite terrible because they tried to apply a single metric to unknown/future hardware. Today, we have the benefit of how history played out, so a better benchmark tool could be written. I'll think about it in bed tonight and try to work out some more details.
 
But a DX-50 has a bus speed of 25MHz, not 33... Just like the DX4-100's are so named because they have, again, a 25MHz bus. Confused.
No! This is a real 486DX-50 we're talking about. Full 50 MHz bus speed, no clock multiplier! 8)

80486dx-50_sx546_cpuswe.jpg


The DX-50 was well known to each DX/2-66's for lunch in real-world use, but it fell out of favor because it was so difficult to get VESA Local Bus and PCI cards to work at 50 MHz, or even 40 MHz (33 MHz was the design spec speed).
 
It still seems there is confusion out there about the 486DX-50 and the DX/4-100.

There are two different 50Mhz 486 chips. The 486DX-50 uses a 50MHz FSB. There is also the more common 486DX/2-50, which uses a 25MHz bus as you mentioned. However, the benchmark label and score provided seem to indicate a 486DX-50 was used.

DX/4 is a clock tripled, not clock quadrupled as the name implies. Therefore a 100MHz DX/4 has a 33MHz bus. There is also a 75MHz version that uses the 25MHz bus.
 
DX/4 is a clock tripled, not clock quadrupled as the name implies. Therefore a 100MHz DX/4 has a 33MHz bus. There is also a 75MHz version that uses the 25MHz bus.
Yes, the "DX/4" should really be a "DX/3". BUT!!! You can also run a DX/4-100 in clock-doubled mode: 2 x 50 MHz instead of 3 x 33 MHz! Intel never really promoted that mode because it'll cause many VLB and PCI cards to crap out, just like the DX-50, but it does work and is not "overclocking" -- 2x50 was part of the design spec.

The AMD 5x86 also supports 3 x 50 MHz mode. They released a very small number of true 150 MHz 5x86s, but you can do it just fine on a "133 MHz" 5x86 (as well as 40 x 4 = 160 MHz).

Am5x86-P75+_150ADW.jpg
 
About intel's policy on 50MHz bus on their DX/4 chips....

actually not all of their DX/4s were rated to work on the 50MHz bus. They used to have this up on their website, but actually only a certain stepping officially supports 2x50MHz mode. I believe that model was of the write-thru variety. Though, all of their chips seem to work with the 2x multiplier. In the official documentation for the chip there was also mention of a 2.5X setting, but it doesn't work. Rumour has it the 2.5X multiplier probably would have been used on said DX/3.

50MHz bus generally isn't worth it on 486 boards because of all the extra wait states that need to be added for the cache and main memory.
 
The 2.5x multiplier was the failed DX/3. I dunno why it didn't make it, but the DX/4 was released in its place.
 
Last edited:
It made sense not to release a chip in between the DX/2 and the DX/4. The difference in performance would have been almost undetectable. The bigger mystery is why they kept the name DX/4 and didn't rename it DX/3 instead after the real DX/3 was canned. I've heard people say that "DX/4" gives the impression of being twice as fast as a DX/2.
 
It made sense not to release a chip in between the DX/2 and the DX/4. The difference in performance would have been almost undetectable. The bigger mystery is why they kept the name DX/4 and didn't rename it DX/3 instead after the real DX/3 was canned. I've heard people say that "DX/4" gives the impression of being twice as fast as a DX/2.
Perhaps it was because IBM beat them to the punch by releasing the first clock-tripled, 100 MHz x86 chip: the "Blue Lightning" 486BLX3-100.

Number-jumping was also in vogue back then: witness Microsoft's MS-DOS 6.0 and 6.2, versus IBM's PC DOS 6.1 and 6.3.
 
Intel has always been kinda cryptic and strange when it comes to code names and version numbers. Who knows why they went ahead and released it as the DX/4. Probably just wanted to stick to the version line. Version 3 didn't work so release the improved (?) Version 4. *shrugs*
 
BTW, I just got a Tandy 1000RL, and for an XT-class machine, this thing is QUICK. Its 9.54 MHz 8086 handily beats my CompuAdd's 9.54 MHz V20 in benchmarks (2.6 vs. 2.1 in Norton SI; 2.8 vs. 2.2 in PC Tools SI). The only downer is that it is soldered in place and can't be easily upgraded to a V30. The speed of its 20 MB Seagate IDE-XT hard drive is also well above average for an XT -- and of course, booting up DOS and DeskMate from ROM is even faster. :D

DeskMate even includes a crude form of MOD-like music tracker: you can compose and play multi-voice music, and instead of synthesizing the tones, it uses digitized samples of real instruments, like piano, cello, clarinet, etc. You can even record and edit your own audio samples using its microphone/line input. Neat stuff! :)
 
Great find! The TL and RL series were always my favorites with the better graphics, better audio, and DOS in ROM. They're great for writing hairy assembler programs -- when you lock up the machine, just hit reset and you're back at the DOS prompt in 2 seconds.
 
Great find! The TL and RL series were always my favorites with the better graphics, better audio, and DOS in ROM. They're great for writing hairy assembler programs -- when you lock up the machine, just hit reset and you're back at the DOS prompt in 2 seconds.
Try SETUPRL /A ... which lets you adjust all the nitty-gritty details of the machine, just like advanced CMOS setup on a 386. It even lets you switch the built-in IDE-XT controller between PIO and DMA modes... that's the first time I've seen that on anything less than a 486! :eek: (PC-Check claims the stock 20 MB IDE-XT hard drive is running at 1:1 interleave, which is amazing for an XT-class machine!)
 
Perhaps it was because IBM beat [Intel] to the punch by releasing the first clock-tripled, 100 MHz x86 chip: the "Blue Lightning" 486BLX3-100...

More commonly known as the 486DLC3-100. It's close to the PQFP 386DX (AMD used this packaging for their last 386DXs) pinouts, and designed to more easily implemented into that class of motherboard (or "planar", as IBM did have unpopulated solderpads that were for this CPU on a few of their late-model PS/2s). In fact the "D" in "DLC" denoted that it was intended as an upgrade for the 386DX, just as IBM's 386SLC, 486SLC2 (clock doubled, most common as 50MHz than 66MHz), and 486SLC3 (clock tripled, but more common as 75MHz than 100MHz) were upgrade CPUs for 386SX-class systems (more correctly stated as 16-bit data bus, 24-bit address bus systems).

The wild thing is that the same package can be configured in it's hardware implementation as a 486DLC3 or 486SLC3! IBM increased the L1 cache size on their CPUs for better performance (the 386SLC had 8Kb, the 486SLC2, 486SLC3, 486DLC2, and 486DLC3 had 16Kb). Despite the name, the 386SLC is able to run all Intel 486SX instructions too.

And to be fair, "IBM beat them to the punch" by being licensed to modify the original Intel 386SX/386DX codemasks. All of the CPUs I've mentioned here have a dual copyright printed on top: "1985 Intel, 1992 IBM". Yes, IBM did vastly improve the performance, but it is derivative from an originally Intel work.
 
not sure if somebody else already said this on these 4 pages, so if this is a repeat sorry. does your board have a jumper that lets you modify the RAM access speed? if so, play around with that and see what happens.

i have an XT turbo board made by PC's Unlimited that has a jumper that switches from 100 to 120 ns RAM speed.
 
Back
Top