The Odd Story of Manufacturing unit-Downgraded 486s – X86.FR
Counterfeits CPU have been quite common within the mid-90s. The worst interval was between 1993 (simply after the launch of the Intel 486 DX2) and 1998 (when the Pentium II began to be multiplier-locked). It was extraordinarily simple for tricksters to take away the unique marking and reprint one other one with a better frequency score. Many DX4-75 have been remarked to DX4-100, and much more Pentium 133/150 have been remarked as Pentium 166 or 200s.
Real factory-remarked CPUs additionally exist, however they’re usually unusual. Essentially the most well-known instance is the double-sigma (ΣΣ) signal added on early 386s after they’d been examined bug-free from the notorious 32-bit multiplier bug. Some uncommon Intel 486 SX have been additionally later remarked with a better velocity grade. Listed below are two of them:
As for all factory-remarks, the addition is sort of apparent. Intel in all probability binned twice these CPUs once more on the request of a giant buyer (IBM?) and added the second score later. In the present day’s story about factory-remarks is far more uncommon as a result of it considerations customary fashions.
Am486DX4-100SV8B (remarked 5×86)
After I printed this analysis some weeks in the past, a reader instructed me he had a wierd Am486DX4-100 that appeared to be a AMD 5×86. After a cautious take a look at the printings that seemed 100% real at first sight, he was form sufficient to lend it to me for additional investigation with the UCA. Right here it’s:
The “9626” date code tells us it was manufactured in late June or early July 1996, which is sort of late for a Am486DX4. I instantly seen the 25544 bundle code, solely used for the 350 nm die. This die was the idea of all Am486DX5 and Am5x86. The “C” stepping was additionally uncommon because the Am5x86 relies on the A-step (from November 1995) or B-Step (from March 1997). A “C” Stepping construct in 1996 is incoherent with the 5×86 line, however very coherent with the 486DX4 (later 486DX4 within the newest “C” Stepping have been constructed on the 25498 bundle in Might/June 1996). So it was time for a check on the Common Chip Analyzer:
WOW! There is no such thing as a doubt: this CPU is absolutely based mostly on the usual 350 nm die with a completely enabled 16 KB Write-Again L1 cache and a working 4x multiplier. Really, it could possibly even be overclocked simply to 133 MHz. All specs, together with energy consumption and CPUID (0x4F4), make it indistinguishable from an AMD 5×86. This CPU can in fact additionally work with a 3x multiplier like an AMD 486DX4-100 (CPUID drops to 0x494).
After some analysis, evidently all CPUs based mostly on the 25544/C bundle are marked as 486DX4-100SV8B whereas being actually DX5 SV16B (5×86). AMD produced them for fairly a while between February 1996 and March 1997. They in all probability stopped the manufacturing of the outdated 500 nm die in early ’96 however nonetheless had some demand from prospects for DX4s, so they simply used the brand new 350 nm die and marked these CPUs as DX4-100s. So long as you employ the default x3 multiplier, they behave precisely just like the outdated one … aside from the cache measurement.
Has Intel additionally accomplished such bizarre issues? I may have sworn no manner. I used to be fallacious…
Intel 486DX2-66 SK080 (remarked DX4)
The identical reader additionally sends me a DX2-66 that may very well be “actually a DX4-100”. That sounded odd and actually unlikely to me as a result of Intel has a strict coverage on S-Spec. Intel DX4s even have a selected CPUID to assist distinguish them from DX2s by software program. In contrast to AMD 486s, this CPUID does NOT change with the multiplier used, so it’s unusual to have a DX2 with a DX4’s CPUID. Right here is the unique CPU:
All the pieces seems to be real right here. SK080 is among the least frequent S-Spec for Intel DX2s. The one different S-Spec starting with “SK” is the extraordinarily uncommon SK058. The SK080 is a 3.3V SL-Enhanced half which appears to have been produced solely between WW18’94 (Might 1994) and WW48’94 (November 1994). Let’s plug in into the UCA:
Superior! That is actually a DX4 factory-downgraded to DX2-66. The 0x480 CPUID leaves little doubt concerning the unique die used right here. The same old energy consumption and the power to work high-quality at 3.3V at 100 MHz let me assume it’s in all probability a DX4-100. With the multiplier set at 2x, the SK080 additionally works at 2×33 MHz as anticipated for a CPU marked as a DX2-66. To be 100% certain, I used to be capable of finding one other pattern to verify these findings.