Previously on this blog we have discussed both what I think is an “ultimate” or at least near “ultimate” 386 PC in the Anatomy of a 386 article and we also talked about pushing certain sockets beyond their intended scope such as with the 486 and Socket 3 motherboards and pushing their limits and I wanted to do something similar with the 386. In this article were going to take a look at the Cyrix 486DLC chip, a drop in replacement chip meant to upgrade a 386 class motherboard with a 486 class CPU. Were going to look at one of the more common DLC chips, the Cyrix DLC-40, run some benchmarks and put it up against the king of the 386 chips the AMD 386-40.
First lets take a look at the test PC and then the CPU’s will be testing.
This is the 386 PC will be testing the CPU’s on.
I could not identify the motherboard but it’s based off the Opti 495XLC chipset and is loaded with 128kb of L2 cache and 8MB of FPM RAM. For video I am using a Tseng Labs ET4000 ISA video card and I have a Sound Blaster 2.0 installed as well although it will be unused for the benchmark tests. My memory settings are fairly standard and I didn’t mess with them much from the defaults. AT bus divider set at 5, memory timings set at 2-1-1-1 and wait states set to 1.
So lets take a look at the two CPU’s will be testing today.
First up is the king of the 386 chips, the famous and beloved AMD 386DX-40.
The 386DX-40 was a very reliable 32 bit 386 chip made by AMD and is very well regarded for its speed and reliability. This is the fastest true 386 made and is overall very capable. It though, like all true 386 chips has no L1 cache memory on board.
The second CPU we will be putting it up against is the Cyrix 486DLC-40.
The DLC-40 created by Cyrix was intended to be a drop in upgrade CPU for 32 bit 386 class motherboards which in essence gave 386 PC users a 486 CPU upgrade. It essentially is a supped up 386 with 486 code inserted and 1kb of L1 cache added on the chip. There was also a more common 33mhz version like this guy.
as well as later DLC chips made by Texas Instruments and IBM which added even more L1 cache. The fastest of these chips is an IBM triple clocked DLC rated at 100mhz that supposedly is about equal to an Intel 486DX-66. Due to licensing agreements though I believe IBM was only able to sell these chips imbedded as part of a motherboard or upgrade kit.
There are also SLC variants which are 16/32 bit hybrids designed to be more compatible with older motherboards where as DLC variants are true 32 bit.
The on board L1 cache memory of the DLC is a huge advantage over the AMD 386-40 so on all the benchmarks I tested the DLC chip twice, once with the cache enabled and one again with it disable so we can see how much of a difference this makes.
The unidentified 386 class motherboard I’m using is a later model and supports the DLC chips in BIOS. Here is a shot of the specs as I boot up.
Notice the BIOS recognizes the DLC chip as under “main processor” it is listed as a Cx486DLC. Many older 386 motherboards may not recognize the DLC chip properly and simply report it as a 80486SX. You may also need to run a setup utility with these older boards to fully utilize the DLC chip.
My BIOS also has the option to enable/disable the internal L1 cache on the CPU though for reasons I cant figure out this option does not work and the l1 cache is enabled automatically regardless of how it is set here.
Failing this you can always disable/enable the internal L1 cache via software. I used the Cyrix cx486 utility which allows you to easily enable/disable the L1 cache by the C button. Just be sure to hit W to write your choice to the registers before exiting the program.
So without further adieu we have the benchmarks. The programs I used were all pretty standard, 3D Bench, PCP Bench and DOOM. Below is a bar chart to better illustrate the performance of the two chips and I also added a 33mhz Intel 486 results from a similarly spec’ed 486 machine using the same amount of L2 cache and the same video card to compare.
The results are basically what we would expect with the DLC handily defeating the AMD 386-40. DOOM seems to benefit the most from the L1 cache as the difference between the 386-40 and DLC with cache disabled is only .6 FPS but with cache enabled its a 2.5 FPS difference.
you’ll also notice the DLC does pretty well compared to the Intel 486DX-33 even tying it in PCP Bench. After some optimization I managed even more performance gain.
After setting wait states to 0 and the AT bus divider to 3 I was able to get the following results out of the cache enabled DLC-40
3D Bench – 22.1
PCP Bench – 5.5
DOOM – 10
You can see with these results the DLC surpasses the DX-33 in all but DOOM.
The folks over at Red Hill Computing whom I often refer to in my research seem to really love this chip and claim it to be “vastly superior” to a 486DX-33 system though I would have to somewhat disagree. In a very optimized system the DLC can match or beat the DX-33 but I wouldn’t say it’s as fast clock for clock and I wouldn’t call a few frames per second advantage “vastly superior”. At the time though they would of been correct in that verbiage from a fiscal point of view as the DLC was vastly cheaper then a Intel 486DX-33 allowing the buyer to purchase more RAM.
I’m very curious how much the added cache of the later DLC chips such as the ones from TI and IBM affected performance and that may be the bases of an eventual revision to this article. The DLC does do what it claims to do and does increase performance of a 386 to 486 levels as well as being vastly cheaper then an Intel 486 but keep in mind it was still restricted by the 386 architecture and limited to the amount of RAM it could use. Possibly not a bad option for the budget user that already owned a 386 machine but holding out for the pricy 486DX-66 would of been a smarter move for the long run in my opinion.
ancientelectronics 