Tejas any information?

[Lepton87]

Are there any leaked Tejas benchmarks or any information about it that's more than pure speculation? This was the planned successor to Prescott. There's a slight possibility that there are some working samples in the wild because the project was cancelled after it had reached its tape-out phase. I wonder how much a working sample would go for on ebay.

 

[Charles Kozierok]

There were some samples, Anand had a piece showing pictures of one of them. No idea if or how well they worked.. and they'd be years old and pretty slow by today's standards anyway.

 

[alyarb]

the tape out date does not really imply anything about the disbursement of samples, if they worked, or just used to show off at trade shows... do you have any other reasoning to suggest there are working samples in the wild? The 10 mentioned by Anand were likely all returned and thoroughly accounted for. Someone would have published by now otherwise.

 

[IntelUser2000]

Tejas was supposed to bring back having 3 decoders as opposed to having one in previous Netburst architectures.

Also it was thought to have a module-like concept similar to Bulldozer: http://ascii.jp/elem/000/000/632/632999/index-3.html

Except with Hyperthreading added of course.

 

[podspi]

It couldn't have been that good, otherwise it wouldn't have been cancelled.

Still, I agree it would be interesting to see what could have been.

 

[soccerballtux]

It couldn't have been that good, otherwise it wouldn't have been cancelled.

Still, I agree it would be interesting to see what could have been.

it could have been a large improvement, but then they found those guys in Israel who gave us the Core architecture, bought them out and that was that.

 

[Ferzerp]

it could have been a large improvement, but then they found those guys in Israel who gave us the Core architecture, bought them out and that was that.

How exactly does a company buy itself out, pray tell?

 

[Magic Carpet]

Tejas @ 10 Ghz would have been a killer cpu

 

[soccerballtux]

How exactly does a company buy itself out, pray tell?

hm can't find anything on it now. I read a story a long time ago about how Intel purchased a research firm in Israel and that's where Core came from.

 

[Homeles]

Tejas @ 10 Ghz would have been a killer cpu

It also would have started a fusion reaction, at those operating temperatures.

 

[alyarb]

Intel had a group working in Haifa since 1974. Any products with code names inspired by nearby landmarks will have come from them (Banias, Dothan, Timna, etc). For instance, Wilamette, Northwood, Nehalem and Tukwila refer directly to history and geography of the pacific northwest. The original codename for SNB was Gesher, which was also the name of a political group in Israel so intel changed the codename and I think then they gave up on the geography-based names.

Anyway, probably not many of you remember Timna but it was supposed to be a Tualatin-based Celeron with an integrated Rambus controller and an even worse derivative of the i740 on a single die. It was to run at 600 MHz and be some kind of AMD Duron/embedded product competitor. Rambus prices remained too high for the target Timna was supposed to hit. In fact, Rambus prices were too high for the soon-to-be-released Willamette platform and intel was already working on a retrofit solution for DDR and SDR DRAM to interface with existing platforms. This was to be called the memory transfer hub and came to be the i820 chipset, which had such serious hardware errata that it was recalled and forced intel to can Timna as well.

I don't know what Haifa was doing for the following two years but it became clear after Prescott began to tape out that no amount of tweaking could get it into a mobile form factor. The Haifa group had already done the work on Timna so they were asked again: put the Penium 4 bus interface on a Tualatin core and build out the front end. This was called Banias at 130nm and completely replaced the mobile netburst line as Dothan at 90nm.

Similar thing happened with Tejas, which they had hoped to launch on 90nm at 4 GHz. Samples running at 3 GHz using over 150 watts made it clear that no amount of tweaking would get it into a desktop form factor. So Haifa was allowed to continue evolving their core, new branch predictor, increased issue width, macro ops fusion, Tejas new instructions etc. The group was owned and run by intel the entire time, but nothing about the technology achieved "pet project" status until they realized Tejas could not be a product. Then they got the idea for tick-tock and now Haifa is churning out flagship products every couple of years as a matter of routine.


I would still be interested to read a Tejas overview, in the way that anand writes his uarch overviews. No one seems to agree on what it was.

 

[nismotigerwvu]

http://www.anandtech.com/show/1217

Looks like they were working, 2.8ghz but pulling 150 watts. Granted these were very early chips (and running on 90nm silicon) and there would have been room for improvement in that regard that's still a truckload of power draw.

 

[Idontcare]

http://www.anandtech.com/show/1217

Looks like they were working, 2.8ghz but pulling 150 watts. Granted these were very early chips (and running on 90nm silicon) and there would have been room for improvement in that regard that's still a truckload of power draw.

Sounds familiar, isn't that what torpedoed 45nm Larrabee? Pulling way too many watts at way too low performance in return.

 

[Magic Carpet]

It also would have started a fusion reaction, at those operating temperatures.

Possibly. It should have met the original design goals, though... but it didn't. And now we see the history repeats with AMD

Sounds familiar, isn't that what torpedoed 45nm Larrabee? Pulling way too many watts at way too low performance in return.

Nobody seems to have had any issues with GPUs pulling 200-300w at the wall, though.

I would still be interested to read a Tejas overview, in the way that anand writes his uarch overviews. No one seems to agree on what it was.

Yeah, it would be interesting to read more on the subject :thumbsup:

 

[Arkaign]

That was another step in the Netburst path, wasn't it? A "tock", as it were? They did make the Prescott replacement, but although it was improved, it was ineffective against the rapidly improving AMD64 single and dual cores.

A lot of people don't remember very clearly, but up to around the 3200+ Athlon 64, the P4s were still hanging tough. There really wasn't any big reason to own one over the other. Once the 3500+, 3800+, X2s came out though, it was game over for Netburst.

Anyway, it was P4 Cedar Mill and PD Preslers that took the Prescott 90nm failure to 65nm, and largely offered what Prescott was supposed to deliver in the first place, albeit too little and too late. It's remarkably similar to what happened to AMD with Phenom, and later Bulldozer. Phenom II delivered what the Phenom was supposed to be, and it looks like revisions to BD will be what it was supposed to be initially.

Timing is everything!

 

[grkM3]

Id love to see a 300watt gpu in a ultrabook lol

to game on it you need to plug an external psu and carry that around with you plugged into a gas generator

on a plus side you could use your laptop to heat up a few rooms in your home durning winter lol

there are lots of people that have problems with 300watt chips and every single company that sold prescotts had heating issues that were a warrenty mess.

I would love to see how far intel could push prescot now on there tri gate tech,Im sure they could easily break 12ghz on a single tejas core and keep it under 150 watts.

There fabs could not keep up with there tech at the time and I remember reading on how intel is doomed when they would need to drop under 90nm and how next to impossible it would be to keep leakege down and today we have quad core 22nm today hitting almost 5ghz on all cores when they couldnt even get 5ghz on a single core that was made to clock high wth less ipc.

Intel is hitting 4ghz today on 4 cores and for shits id like to see what they could clock out of a huge single core monster at 22nm.

imagine what they could do now on a single core with a tdp of 150 watts and no gpu

 

[nismotigerwvu]

Sounds familiar, isn't that what torpedoed 45nm Larrabee? Pulling way too many watts at way too low performance in return.

I'm still rather curious of how these chips performed though. I mean, the answer is obviously "poorly" or at least "poorly compared to dothan/whatever larrabee was ever intended to compete against". I mean what could it hurt to release some numbers, or better yet, an old mobo/chip that wasn't destroyed. Much akin to the voodoo 5 6000's that slipped out. Also, it would be really neat to see any data on the original netburst-esque K10 (grayhound I think?) that was rumored to be deeply pipelined, SMT'd amd all that jazz. Clearly a lot of these ideas got rolled into bulldozer, but still, early data is fascinating

 

[Maximilian]

Well there were 10 sample chips shipped out to "friends of intel":

http://www.anandtech.com/show/1217

Considering what other rare chips go for on ebay i imagine these samples if one ever ended up there would fetch quite a price. CPU collectors have their rare 4040 chips, soviet replicas of things, FDIV big pentium 1's but none of them have a tejas to my knowledge so they would be all over a tejas should it appear.

 

[Magic Carpet]

Intel must have forced everyone to produce a valid Death Certificate. It would have popped out by now otherwise...

 

[nismotigerwvu]

Oh I know, but I really don't see the point in the "it must be destroyed" policy on all their ES chips, especially for designs that see the light of day. Why not directly auction them off to collectors as display pieces with no warranty or support of any kind? I know the bit of revenue they would generate from these sorts of things would be like 0.00000000000000000000000000000000000000 *takes a breath* 0000000000001% of net income, but it would be better than free PR (making money rather than taking it).

 

[Idontcare]

Oh I know, but I really don't see the point in the "it must be destroyed" policy on all their ES chips, especially for designs that see the light of day. Why not directly auction them off to collectors as display pieces with no warranty or support of any kind? I know the bit of revenue they would generate from these sorts of things would be like 0.00000000000000000000000000000000000000 *takes a breath* 0000000000001% of net income, but it would be better than free PR (making money rather than taking it).

Had the benches come to the public domain the results would have only added more ammo to AMD's marketing machine, as well as any other media outlet looking for webtraffic and subscriptions (NYT, LATimes, etc) for yet-another-story on Intel's engineering failures.

All of that bad press would have weighed down Intel's stock price far more than it would have weighed down its earnings or revenues. And for executive decision makers whose compensation packages are far more weighted towards the market valuation of INTC than the market valuation of CPUs, maintaining a tight reign on anything that would damage the market valuation of INTC is going to be daily-activity.

You don't let your follies reach the front page of the New York Times and keep your golden parachute that nice golden color. If you are a nerdy geek who likes benches for obscure canned projects then you really only have one choice - work for the company and on the specific project in question, or discreetly throw a lot of beer at someone who did

 

[pantsaregood]

All of the existing dies probably reached autoignition temperature and ceased to exist.

 

[nismotigerwvu]

Had the benches come to the public domain the results would have only added more ammo to AMD's marketing machine, as well as any other media outlet looking for webtraffic and subscriptions (NYT, LATimes, etc) for yet-another-story on Intel's engineering failures.

All of that bad press would have weighed down Intel's stock price far more than it would have weighed down its earnings or revenues. And for executive decision makers whose compensation packages are far more weighted towards the market valuation of INTC than the market valuation of CPUs, maintaining a tight reign on anything that would damage the market valuation of INTC is going to be daily-activity.

You don't let your follies reach the front page of the New York Times and keep your golden parachute that nice golden color. If you are a nerdy geek who likes benches for obscure canned projects then you really only have one choice - work for the company and on the specific project in question, or discreetly throw a lot of beer at someone who did

But do you really think that a near decade old scrapped design would really be a source of negative PR though. If they set the auction date far enough away from relevance it should pose little to no risk. Besides I doubt Intel has much of a for phd level biochemists. :-D

 

[Idontcare]

But do you really think that a near decade old scrapped design would really be a source of negative PR though. If they set the auction date far enough away from relevance it should pose little to no risk. Besides I doubt Intel has much of a for phd level biochemists. :-D

Remember, Intel's motto was/is "only the paranoid survive". Arguing that nothing bad can happen is not how risk is managed, proof that nothing bad will happen is how risk is managed.

Further, at the individual level, there is no financial incentive for any given person within Intel's employ to expend their time managing all that must be managed in creating/releasing the information.

But there is financial incentive at the individual level to pursue the exectution of goals and projects already underway.

I know if it were up to me I would not take on any secondary work regardless how kewl it might be if that secondary work stood a chance of slowing down my primary work and then knocking my year-end bonus down say 10% or 15%.

 

[nismotigerwvu]

Remember, Intel's motto was/is "only the paranoid survive". Arguing that nothing bad can happen is not how risk is managed, proof that nothing bad will happen is how risk is managed.

Further, at the individual level, there is no financial incentive for any given person within Intel's employ to expend their time managing all that must be managed in creating/releasing the information.

But there is financial incentive at the individual level to pursue the exectution of goals and projects already underway.

I know if it were up to me I would not take on any secondary work regardless how kewl it might be if that secondary work stood a chance of slowing down my primary work and then knocking my year-end bonus down say 10% or 15%.

Oh, I understand and completely agree with the point you are making, but I think I might have expressed mine clearly enough. What I was getting at was more along the lines of altering the ES policy from "preview, return, destroy" to "preview, return, auction on website 7 years later". It's just personal opinion (and like I said, I'm a biochemist, not a EE or even finance guy) but it seems like the relative value added from destroying the chips is always 0 (which the support for is that this is a guaranteed non-negative value) versus the potential free PR that could come from auctioning off even a portion of those ES chips (say maybe 100 are set aside from the "destroy" pile). I'm sure they have a broom closet somewhere that they could set these chips away on.