Question Intel had a 7 GHz CPU years ago

[igor_kavinski]

https://www.reddit.com/r/hardware/comments/wir8ay

213 mm2 die size, 150 Watts, 50 pipeline stages all for one core at >7 GHz.

I'm getting a strong feeling of lust for this CPU. 150W ain't that much. Imagine what this CPU could have been if it had been developed further, with modern branch predictors and huge caches.

 

[igor_kavinski]

Why is that impressive? Apple's dynamic translator does even better at 70-80%. The Itanium's translator is on the low side closer to 50%.

In that era, it was impressive. Remember, Apple has the advantage of more money and thus more motivated individuals. Also, more man-years of research. Another important factor is having attracted the brightest talent to their side, no doubt with the allure of better compensation but also maybe better working conditions like no interference from higher-ups unless they were executing badly etc.

Speaking from my own example. I'm able to do my best when no one has told me to do something. I come up with solutions to problems I see. Now if someone says to me, this is the problem and this is how much time you have got to solve it, then I will most likely fail. I don't like that kind of pressure on me. Stunts my thinking process.

 

[igor_kavinski]

Seems russians were able to develop something more performant on average than Rosetta: https://habr.com/en/companies/huawei/articles/577208/

Currently owned by Huawei: https://threedots.ovh/blog/2021/01/huawei-exagear-x86_64-apps-on-arm64/

 

[LightningDust]

If the Itanium had an x86 hardware co-processor, it probably may have fared better, assuming Intel and HP had brought its price down to Xeon levels.

First, as already mentioned, it did. It was terrible. IA-32 EL, the software solution, did considerably better.

Second, in 2001-2010, in Itanium's target markets, nobody cared all that much about fast x86 compatibility. The lion's share of IA-32 EL use was just running computationally-trivial Windows utility software on Windows Server for Itanium. IA-32 EL existed for Linux too, but at the time there wasn't some massive pile of binary-only Linux/x86 software that the (small) Linux/IPF userbase was itching to run, so it saw little use. And remember - Windows Server and Linux combined never accounted for more than 20% of Itanium shipments, and across the lifetime of the product, it was more like 3-5%. Nearly all Itanium systems shipped with HP operating systems.

 

[Doug S]

Seems russians were able to develop something more performant on average than Rosetta: https://habr.com/en/companies/huawei/articles/577208/

Currently owned by Huawei: https://threedots.ovh/blog/2021/01/huawei-exagear-x86_64-apps-on-arm64/

Where is the "more performant" part? All I see is a link to Geekbench 5 results that show the translator is at best 65-70% of native. Rosetta 2 easily beats that.

 

[igor_kavinski]

Where is the "more performant" part?

As we can see, Huawei ExaGear has a better average performance of 82.4% in comparison to Rosetta 2’s 76.9%, and ExaGear loses only in 4 tests out of 21: Navigation, Camera, Rigid Body Physics, and Image Inpainting.

As we can see, the results from SpecCPU2006 and SpecCPU2017 follow that of Geekbench: on average, Huawei ExaGear beats Apple Rosetta 2, although in a few subtests Rosetta 2 is able to outperform ExaGear.

URL: https://habr.com/en/companies/huawei/articles/577208/

 

[igor_kavinski]

Conspiracy alert!

Someone tried to sabotage P4 but thankfully, a kind soul averted the disaster: https://www.phoronix.com/news/Linux-6.17-Pentium-4-Fix

Yes, Yes, I know what everyone will say but this wasn't a copy/paste mistake. You don't make mistakes when you are dealing with sensitive code. You check and double check your work and then someone else reviews it before it is mainlined.

 

[yottabit]

You don't make mistakes when you are dealing with sensitive code.

Oh my sweet summer child

 

[lamedude]

4GHZ sighted!

https://www.reddit.com/r/intel/comments/1nloltl/found_this_intel_extreme_edition_980_engineering

 

[zir_blazer]

4GHZ sighted!

https://www.reddit.com/r/intel/comments/1nloltl/found_this_intel_extreme_edition_980_engineering

CPU-Z author seemed quite enthusiastic about this find: https://www.cpu-world.com/forum/viewtopic.php?p=330497#330497
The previous PEE (Pun intended) 965 was based on Presler C1 whereas the found one is D0, so Intel may have gotten the chance to push the top bin a little further. Back then Steppings could have a drastic clock speed and power consumption difference, whereas today is rare to see a new Stepping at all.

I recall than 4 GHz Pentium 4 were even on preorder in eRetailers at some point, Intel Ark, etc, then... nothing. Two decades later and this seems to be the very first time an actual unit that ran by default at that clock speed is found. Is ridiculous if you consider than release seemed inminent yet samples are so infinitely rare, makes no sense.

Also, I recall reading on CPU World that Intel at some point planned Northwood on LGA 775, of which no ES ever appeared, either. The only Northwood on LGA 775 is essencially the P4 EE Gallatin (Northwood with Cache L3). However, I recently learned than there are actually 478-to-775 adapters and that some people got Northwoods running fine with some BIOS modding, if for some reason you want to see it with DDR2 (There are also LGA 775 Chipsets with DDR3 support but not sure if any of those actually worked.

 

[igor_kavinski]

(There are also LGA 775 Chipsets with DDR3 support but not sure if any of those actually worked.

Yeah, they work alright: https://forums.anandtech.com/threads/intel-had-a-7-ghz-cpu-years-ago.2608070/post-41491594

 

[zir_blazer]

Yeah, they work alright: https://forums.anandtech.com/threads/intel-had-a-7-ghz-cpu-years-ago.2608070/post-41491594

I mean, if Northwood works on those Chipsets. It seems to work on early ones, I think I saw one with DDR2, but don't remember DDR3 since later Chipsets officially dropped out NetBurst compatibility.

 

[igor_kavinski]

I wish we had a resident P4 collector to give us answers to these burning questions. If I owned a home, there would be at least one room stacked to the ceiling with different P4 generations.

 

[NTMBK]

Saw this and thought of you @igor_kavinski

Tejas wasn't the zenith of processor complexity, but it was a very good lesson in engineering management vs engineering leadership.

I think the same lessons apply 21 years later:

1. don't confuse being right for the wrong reasons for being right ("but we've been making money this way!")

2. you should fire incompetent people the first time they unequivocally demonstrate they are incompetent. (making an honest mistake isn't incompetence. incomptence is incompetence).

3. people will complain no matter what you do, so you might as well do the right thing

4. a crisis is a terrible thing to waste

5. don't confuse being better than wrong for being right ("the EV6/P6-style architecture is the zen of computer architecture")

6. you cannot solve big problems with crappy math ("the geometric mean of the harmonic mean of performance benchmarks" was a thing people unironically plotted on slides back then. total nonsense)

7. you can build big things with "perfectly fine" engineers and scientists but then they assume they are actually good at the job, rather than merely the ones who showed up

8. the best way to get lucky in your career is to make many bets in parallel

9. waiting in line is for schmucks; try to seize as much impact as you can

and my favorite,

10. leaders lead, managers should be legendarily excellent or immediately fired

Lessons from Tejas: engineering management vs leadership | Armaghan Naik posted on the topic | LinkedIn

Tejas wasn't the zenith of processor complexity, but it was a very good lesson in engineering management vs engineering leadership. I think the same lessons apply 21 years later: 1. don't confuse being right for the wrong reasons for being right ("but we've been making money this way!") 2...

www.linkedin.com

 

[zir_blazer]

Saw this and thought of you @igor_kavinski

We NEED a high resolution picture of the heatspreader of that CPU to see if we can get a sSpec or any other info.
Also, the PCB to the right seems Socket 604, is that Jayhawk?

 

[aigomorla]

ack edit... saw this was a necro'd thread from 2022.

 

[NTMBK]

ack edit... saw this was a necro'd thread from 2022.

The prophecy fulfilled:

Yearly necro until Intel comes to their senses