Mystery NetBurst

Some time ago, a mysterious CPU showed up at the OS/2 Museum:

Intel CPU, S-spec SL7HY

Intel CPU, S-spec SL7HY

It is a Socket 775 CPU with a Pentium 4 label and the following markings: 3.73 GHZ/1M/1066/A4. In other words, 3.73 GHz clock speed, 1 MB L2 cache, 1066 MHz FSB, and A4 power/TDP requirements (should be mere 84W TDP according to Scott Mueller). The S-spec of the CPU is SL7HY. The processor was manufactured in week 22 of 2004, long before the fall of the evil heat-dissipating NetBurst empire.

Okay, so this is one of the faster-clocked Pentium 4 CPUs, perhaps with an “extreme” FSB speed, but what’s so unusual about it?

What’s unusual is that there is no known Intel CPU with these characteristics, and the S-spec is completely unknown. It’s not a qualification or engineering sample processor; the S-spec suggests a standard production CPU, and there are no SECRET/CONFIDENTIAL markings either. The processor is not unheard of, but it’s a little strange.

It’s instructive to see what CPU-Z thinks about the CPU:

CPU-Z statement on SL7HY CPU

CPU-Z statement on SL7HY CPU

Xeon? Not if the labeling on the CPU is to be believed. Socket 604? Uhh, nope. Calling the CPU Nocona is not too far off the mark, if we ignore the fact that there were no Nocona CPUs with 1066 MHz FSB. But Nocona processors were 90nm Prescott relatives with 1 MB L2 cache, and that’s what this one is. If anything, this nicely illustrates that Intel’s product naming (Prescott/Nocona/Irwindale and so on) is rather arbitrary.

Note that CPU-Z thinks the CPU is an engineering sample (ES). It is not clear what that is based on; possibly simply the fact that a CPU with the exact characteristics is missing from CPU-Z’s database. The “Xeon” part comes from the board’s BIOS.

The latest version of Intel’s own CPU identification utility claims that the CPU was “made after the release of the utility” (droll sense of humor there!), in other words the utility has no clue.

According to Intel’s ark site, there were only three released processors with 3.73 GHz clock speed: Xeon 5080, Pentium Extreme Edition 965, and Pentium 4 Extreme Edition with HT. The first two are the same thing, a 65 nm dual-core Dempsey/Presler with 2 × 2 MB L2 cache. The 3.73 GHz P4 EE is a 90 nm Prescott, probably very very close to our mystery CPU. The difference is that the P4 EE has 2 MB L2 cache, and SL7HY only has one megabyte of L2 cache.

So what is the strange processor? Based on the available data, it looks like a variant of the Prescott Pentium 4 Extreme Edition with half the L2 cache. Possibly an unreleased precursor, given that the P4 EE was supposedly released in Q4 2004 and the SL7HY is from the end of Q2 2004.

Given how much heat it generates and how poorly it performs, the SL7HY CPU is not worth running, but it is interesting nonetheless.

Update: HWiNFO is just as confused by this processor as CPU-Z:

HWiNFO statement on SL7HY CPU

HWiNFO statement on SL7HY CPU

Yes, this is Intel’s own Bad Axe 2 motherboard. The board’s BIOS is where the Xeon designation is coming from.

And for good measure, the contents of /proc/cpuinfo from Linux:

processor : 0
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) CPU 3.73GHz
stepping : 2
cpu MHz : 3733.467
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 1
apicid : 0
initial apicid : 0
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca
 cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe
 syscall nx lm constant_tsc pebs bts nopl pni dtes64 monitor
 ds_cpl est cid cx16 xtpr
bugs :
bogomips : 7466.93
clflush size : 64
cache_alignment : 128
address sizes : 36 bits physical, 48 bits virtual
power management:

processor : 1
vendor_id : GenuineIntel
cpu family : 15
model : 4
model name : Intel(R) Xeon(TM) CPU 3.73GHz
stepping : 2
cpu MHz : 3733.467
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 1
apicid : 1
initial apicid : 1
fpu : yes
fpu_exception : yes
cpuid level : 5
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca
 cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe
 syscall nx lm constant_tsc pebs bts nopl pni dtes64 monitor
 ds_cpl est cid cx16 xtpr
bugs :
bogomips : 7466.93
clflush size : 64
cache_alignment : 128
address sizes : 36 bits physical, 48 bits virtual
power management:
This entry was posted in Intel, Pentium 4. Bookmark the permalink.

20 Responses to Mystery NetBurst

  1. John Heritage says:

    Stuff like this is always neat. There were rumors of European dealers getting very small quantities of 4.0GHz Pentium 4 CPU’s as well that also never became official.

  2. Michal Necasek says:

    I’m skeptical that those were ever delivered. The CPU was definitely planned, and there’s even evidence in ark. Note how there’s no S-spec for that one. The CPU is also not listed on CPU-World which makes me strongly doubt that anyone has one (beyond an engineering sample perhaps).

  3. Peter says:

    Perhaps a special for an OEM? Every now and then I see Lenovo and Dell selling entire machines for slightly more than the retail price of the CPU it contains, so clearly the OEMs have some pull.
    In the Netburst era AMD were still competitive with Dell and HP both shipping AMD machines. Perhaps it was Intel’s way wooing an OEM back?

  4. bhtooefr says:

    What does HWiNFO report, out of curiosity? That tool actually will try to guess the qSpec…

    Also, CPU-Z will use heuristics from the steppings, processor brand string, and such to figure out what’s actually going on. That’s where it’s getting the socket and the codename from, they’re not actually known in the CPU. So, the brand string is saying it’s a Xeon, the steppings match with a Nocona…

  5. Michal Necasek says:

    I have no direct experience with Intel CPUs, I do know that AMD had “OEM specials” with OEMs shipping CPUs that were not available through retail channels (or at least not until months later). So yeah, it’s entirely plausible that Intel did similar thing.

    But… if they did that… where are those 4 GHz Pentium 4s? The CPUs wouldn’t just vanish. Given that there’s no evidence those CPUs were ever delivered, I’ll take it as a 99% solid proof that the rumor is just that, a rumor. The CPUs were planned but never materialized.

  6. Fernando says:

    It would be interesting for me the output with linux of cat /proc/cpuinfo, I think that will be closer to the metal.

  7. Richard Wells says:

    The 4GHz Pentium 4s never got the yields necessary to enter the mass market. I think small numbers were sold to specialty customers who needed the performance and were willing to pay an inflated price. The lower clock speed follow-ons with larger caches were faster for most users and lacked the yield problems. Unfortunately, with production totaling only in the thousands, the only way to get one would be to intercept retired systems on their way to the breakers.

    Rumors are that Intel has a similar setup today selling small quantities of 5 GHz chips to deep pocket customers.

  8. Yuhong Bao says:

    I noticed that the 65nm Pentium D was “losing” ~30W of TDP per new stepping at the same clock speed. I can imagine a 125W Pentium D 990 at 4.2Ghz based on the D0 stepping. I wonder if this was the original plan.

  9. Michal Necasek says:

    It’s just as confused as CPU-Z. See updated post.

  10. Michal Necasek says:

    That’s totally plausible, but what were these CPUs? Did they have some off-the-books S-spec? Or were they labeled as engineering samples? Did they also need a custom BIOS? Inquiring minds want to know…

  11. Michal Necasek says:

    Possibly. The problem I think was that such CPUs would have come out in the Core 2 era and would be horribly overpriced underperformers. I doubt such a CPU could keep up with, say, an X6800 Conroe (except in terms of heat dissipation).

  12. GotNoTime says:

    The Xeon model string is coming out of the CPU itself. It isn’t the BIOS since it is a result of a CPUID instruction.

    As for how the tools detect whether it is an engineering sample or not, it appears that it just based on having an internal database of expected values. There doesn’t appear to be any flag that indicates it is an ES CPU. cpu family: 15 model: 4 stepping: 2 doesn’t exist according to the various online lists. Stepping 1 is a Prescott. Stepping 3 is Irwindale, Nocona or Prescott again. Stepping 2 is your mystery CPU.

    As mentioned by other people above, Intel do make weird CPUs as a custom order by somebody big. I have a couple E3 Xeons from eBay that were pulls from something but Intel doesn’t believe they exist at all.

  13. Yuhong Bao says:

    Yea, I know it would not be as good as Conroe.

  14. Peter says:

    Did the extreme editions sell well? Could it be Intel’s way of getting rid of a bunch of unsold Extreme Editions? Somehow disable half the L2 and ship them out? I’m still leaning towards an OEM deal, but haven’t found any evidence for it yet. I suspect they must have wound up in consumer/big box retailer grade machines. That’s where I’ve seen all sorts of hard disk models that don’t officially exist and special (crippled) GPUs configurations.

  15. Michal Necasek says:

    The brand string may depend on a microcode update (that much is documented by Intel). Also, I don’t know if that’s true for Intels, but on AMDs the brand string is usually programmable and needs to be set by the BIOS.

    I have personally seen the reported brand changing after a BIOS update for Intel CPUs (Core 2 IIRC). So it kind of is in the CPU itself and kind of isn’t.

  16. GL1zdA says:

    This CPU actually has a stepping which is nowhere officially documented. There are no 0F42h CPUs in any of the Intel Specification Update PDFs (be it Pentium 4, Celeron or Xeon). There official Prescott 1Ms are: 0F33h (C0), 0F34h (D0), 0F41h (E0) and then 0F49h (G1). On the other hand, by searching for 0F42h I have found a site listing microcode updates with this stepping, so it Intel definitely supported them. I also would test it with a newer motherboard – the 875 did not support 1066 FSB, the 925 was the first to do it.

    My guess is, this was one of the many OEM CPUs Intel released at that time. There were P4s which only made to OEM PC like Dell or HP – for example the Pentium 4 F-series, the first Pentium 4s with 64-bit enabled. I did a quick search for any news item mentioning Dell running a 3.73 GHz CPU, but couldn’t find anything, try it, maybe you will be more lucky.

  17. Michal Necasek says:

    This was a 975X chipset, not 875P. And the CPU did run with 1066 MHz FSB.

    The theory about an OEM-only CPU sounds quite plausible, and would explain why it’s by all appearances a production processor yet “off the books”. A kind of semi-extreme edition, with faster FSB and clock but not the larger L2 cache. (Also a bit older than the 3.73 GHz P4 EE.)

    As for the stepping… it looks that Celeron D 310, S-spec SL8S4, had that same 0F42h TFMS with G0 stepping. So that might be why it’s in a microcode update list. I can not find that S-spec in a specification update, but it is listed on ark (as G0 stepping). Weird!

  18. Dale Smoker says:

    How rare are ‘one-offs’ like this?

    I wonder what its market value is among collectors?

  19. Michal Necasek says:

    How rare… that’s something most likely no one knows, because it’s completely unknown how many such CPUs were made.

    As for market value, based on a sample of one, I’d say not higher than a “known” CPU of similar characteristics.

  20. Denis says:

    Since you can run linux on this CPU … Can you install the most recent Intel microcode update package? On the modern kernels the microcode gets updated the first thing, the moment the kernel loaded initrd. This update might let us identify the CPU better. Even if it does not, it can give us more info based on the fact the microcode revision has changed (or not).

Leave a Reply

Your email address will not be published. Required fields are marked *