Reply 20 of 31, by Rita's Cafe
I've acquired three DX4-100 from eBay in the past year. All are SK096 write back models. The left one and middle one came from a Chinese seller (sold together). The right one from a seller from USA.
As expected all three perform exactly the same scores on the benchmarks apart from one little oddity . The CPU on the right will not load the Quake benchmark when write back cache is enabled. Every other test demonstrates a performance difference between WB and WT. So I don't doubt that Write Back does work but it will not allow Quake to load. Any theories?
I should point out that all my tests were performed with the CPU overclocked to 120MHz with more than adequate cooling. Chipset is SiS471.
The CPU on the left strangely has a larger die cap than is normal for a DX4. Much like the one in the original post. Is this a late version DX4 but still earlier than the 2006/2007 font change?
When I first tried benchmarking this CPU it would not complete the Doom high settings benchmark. It repeatedly froze somewhere in the middle of the test. I tried write back & write through and both just resulted in a freeze during the test. I just assumed it was probably a remarked DX4-75. After all, I was trying to run it at 120MHz. Curious if these were re-badged I decided to clean the ceramic top of all three CPUs with acetone as well as cleaning the die cap of the left CPU. Absolutely no difference to appearance was observed. However the left CPU now completes the Doom benchmark flawlessly every time.
If I understand the date codes correctly then these are the dates:
LEFT: 1995 or 2005, week 21
MIDDLE: 1999, week 15
RIGHT: 1997, week 51
I'm not really sure what to make of all this.
Are these all "Genuine Intel" CPUs? Is the left CPU a 2005 model with a large die cap? What fabrication process is it, 0.8µm, 0.6µm, something smaller? Has anybody tried to de-lid one of these? And why won't the CPU on the right run Quake?