LAPTOP CPU UPGRADES - SOME REMARKS

(Copyright 1997 by David A. Ross)

1. INTRODUCTION

A common question on this Thinkpad mailing list and in the USENET comp.sys.laptops newsgroup is, "Can I upgrade the 486 processor in my laptop?" This is especially common for owners of IBM's TP701, who are loathe to abandon the unique "butterfly" form factor but find the clock-tripled DX4-75 too slow for a serious number crunching or game playing.

The answer is that in most cases is that you can in fact upgrade a laptop's 486 CPU, but not without some serious repercussions. This note is the result of my own attempt to understand whether such an upgrade is worthwhile, and includes (among other information) a report on a similar upgrade of a comparable desktop machine.

2. SPEEDUP CHIPS

The usual candidate for a 486 replacement is a chip from the AMD 5x86 family (not to be confused with their K-5 chips). The AMD586 (see www.amd.com) is a clock-quadrupled low-power chip designed as a plug-in replacement for the Intel 486 line of chips. Despite its name it is not really comparable to a Pentium at the same speed, though its clever design (including an on-board 16K write-back cache) makes it noticeably faster than a comparable i486DX-4. It is virtually pin-compatible with the Intel 486DX-2, which makes it a relatively simple upgrade for that chip, and it runs fairly cool with high heat-tolerance (at least this is true for the ADZ version of the chip) so is ideal for enclosed unventilated spaces like laptops.

Cyrix also makes a 586 replacement for the i486; while only clock-tripled, at the same bus speed it is computationally comparable to the AMD chip. While popular as a desktop 486 replacement, this chip is for technical reasons not well suited to laptop upgrades, and all references that follow are to upgrading with the AMD chip.

3. UPGRADING PROCEDURE

If your laptop's cpu is socketed (not soldered), runs at low (3.3v) voltage, and can support the chip's features, then the upgrade might be simple: buy the AMD for your machine's bus speed or greater (at this writing the AMD586-133ADZ, for a 33mhz bus speed, sells by mail for less than $40), yank the old chip (carefully! - you might need to replace it if the new one doesn't work) and pop in the new one.

Unfortunately, this might not work. For example, your motherboard might use different signals than the AMD needs to get the clock multiplier right. (In particular, it has to be told to clock double; it then internally multiplies by 4. If you're replacing a clock-tripled chip, then you will likely only get at best clock-tripled performance from the AMD; if you are replacing a clock-quadrupled chip, then the behavior is unpredictable.)

The chip also needs to have its cache activated and set properly to write-back mode. (On my desktop motherboard these are accomplished by jumpering the board to recognize the chip as a clock-doubled Intel P24D. Such jumpers probably don't exist on your laptop's board.) There is software available which can often set the cache properly after bootup. (Incidentally, there are many reports on comp.ibm.pc.chips of incompatibilities between write-back mode and some SCSI controllers. Many laptops are internally SCSI. Be warned.)

If your laptop runs at higher voltage (unlikely!), or is not reconfigurable to recognize the chip, but *does* have a socketed CPU, you can still do a relatively inexpensive upgrade. Gainbery (see below) sells a do-it-yourself laptop upgrade kit based on the AMD586 which takes care of voltage regulation, clock-multiplication, and cache-setting on board. (They also sell a much cheaper kit of this sort for desktops; I assume the price difference is mainly package thickness.)

If your CPU is soldered - most are - then the above options won't work. Several companies offer 486 to 586 conversion for many laptop motherboards, including the TP701. The plus side of such a conversion is that they'll handle both the (de)soldering and the compatibility issues mentioned above. For the down side, see below.

Interestingly, Gainbery claims at their web site to be the only manufacturer of 486 to 86 laptop upgrades. If true, then these other companies are likely resellers/installers of the Gainbery product. (There are,of course, many manufacturers offer similar *desktop* upgrades.)

Is there a reason *not* to upgrade? Yes, several. The obvious one is the cost. While not an issue if you can just pop in a $40 AMD chip, the Gainbery package is nonnegligible, and the custom upgrades weigh in at $400-$600.

Another is loss of warrantee. Any of these upgrades will almost surely void your warrantee (though on the socketed chip replacement you can always pop the other chip back in and feign innocence, if your ethical code permits). In my opinion, this is a serious drawback - unlike desktops, laptops are not very user-servicable, and warrantees are one of the most valuable components. You can be sure that that the day after you do one of these upgrades, your port replicator or hard drive or screen will die - even if not related to the upgrade, you'll be stuck with a huge repair bill.

4. HOW MUCH GAIN?

Vendors of CPU upgrade packages often claim 100% speedup, or performance equivalent to a Pentium 75. Are they right? Well, maybe. My car has a top rated speed of 118 miles/hour, but road quality, weather, and traffic conditions usually keep me well below 100. Your laptop's speed is limited in application by bus speed, disk and video performance, and memory access speed. For the TP701, for example, the disk and video performance are actually quite decent, but the bus speed is stuck at a sluggish 25mhz and the the lack of Level 2 cache and EDO ram slows memory access.

For computers, we need to distinguish between processor performance and 'systemic' (or whole-machine) performance. Only the latter will be visible to you when your usual applications on the machine. Even AMD makes it clear that a big increase in the fomer might have only a marginal affect on the latter (see the discussion of write-back versus write-through caching for the 586 on their web page).

Unfortunately, it is difficult to rate systemic performance, since some applications (especially number crunching) are very sensitive to CPU speed.

I recently upgraded my desktop computer to an AMD586 chip. Before the upgrade my system was somewhat comparable in performance to my TP701. The CPU was a (Cyrix) DX2-66 with no L2 cache, and disk and video performance was not too different from the laptop. I replaced the CPU with an AMD586DX-133ADZ, and ran an exhaustive series of benchmarks, both before and after the upgrade. I summarize these results in the next section, along with some predictions of how they would translate to the laptop.

****However, for the impatient let me say that the systemic speedup, while less than the usual claim, was substantially better than I had expected.*****

5. BENCHMARKS

The benchmarks I ran were:

1. PC-DOS's QCONFIG. Not really benchmark, but it *does* try to infer the speed at which the 486 system is running.
2. Norton SI (from version 5.0) (just the CPU speed test)
3. Norbert Juffa's Comptest 2.59 (which reliably reports CPU whetstones/dhrystones/mflops, as well as cached and uncached memory access times and hard disk speed).
4. BYTE Magazine's windows benchmarks (ByteCPU.exe), which while performing some tasks with a systemic feel to them are largely tests of raw processing power.
5. MATLAB (a numerical analysis laboratory) comes with a benchmark suite demo; these benchmarks are somewhat more systemic than the previous ones, though still heavily CPU-intensive; mainly a blend of CPU and memory access.
6. 3D Benchmark v.1.0 from New Dimension International; this is a test of processor speed (*not* FPU), memory access, and video speed/access. I also ran two application benchmarks of my own devising:
7. S-plus test. S-plus is a statistics language; the benchmark I ran involved a large regression involving too much data to be processed in memory alone, so this test was disk dependant (though I had caching on, so even the disk access involved CPU processing).
8. TeX test. I typeset a large document with TeX.

I ran others besides these, but haven't used them for my evaluation here, since the information they gave was not much different from that given by the tests above.

Note to WINTUNE fans: While I did run Wintune 2.0 both before and after, I couldn't get consistent results from repeated runs, so I omit the results here.

Note to ZDLABS (Winstone/ PCBench, etc) fans: ZDLabs keeps changing their benchmarks; what's the use of a benchmark if you can't compare results between machines? Anyway, I'm a bit cheesed off at them for this.

6. RESULTS

Here are the before and after benchmark results, together with the performance ratio. (Based on several replications of each benchmark; ratios are rounded off.)

[Note: higher numbers are better, *except* for (9) and (10)]

7. CONCLUSION

As you can see, even the systemic/applications improvements were quite good; I hadn't really expected this. For $40, this upgrade was well worth it!

However, there is some question as to how well these results will translate to the TP701. Recall that my desktop's Cyrix CPU was only clock *doubled*, had an internal 8k cache, and a reputedly inferior FPU. The TP701 is clock *tripled* (well, mine is - I know some of you do have clock-doubled DX2-50 chips in your Thinkpads), has a superior Intel FPU on board, and has a full 16k cache like the AMD replacement. Thus, we should expect much less speed improvement for the DX4-75 TP701 (and marginally less improvement for the DX2-50 version because of the cache and FPU).