Intel cancels Tejas!

Deepak

B3D Yoddha
Veteran
Link

The chips to be canceled include Intel's fourth-generation Pentium 4 chip, code-named Tejas, which was to be sold next year. Also, a new Xeon processor for low-end computer servers, code-named Jayhawk and believed to be based on a similar architecture to Tejas, is also to be canceled, the person said.

***********************************************************

What does that mean for Intel and for entire industry as well??
 
Deepak said:
What does that mean for Intel and for entire industry as well??

Odds are it'll mean nothing, Tejas was massively late (it was actually supposed to precede Prescott) had performance and heat issues. What will probably happen is that there will be a bit more of a gap between when Tejas was scheduled to come out and when the next chip actually does. There might be a quick shrink of Prescott to 65nm if the next Prescott derivative is taking to long to finish.

Actually the death of Tejas has been talked about for a really long time inside of Intel, I'm surprised it took this long to come out. Like I said Tejas was actually supposed to precede Prescott, but when that didn't happened the men in the suits started to ponder its usefulness. It's really a shame they did cancel it since it was supposedly getting really close to tape out, but then again it wouldn't really have a place in Intel's product line.
 
If intel introduce Dothan as an small server processor I will seriouslly think about buy it 8)

IIRC The Pentium-M (P3 core) was supposed to be available for small 1U servers.

I really miss my P3-S. Maybe I will buy it back.
 
From Wall Street Journal:

"Intel Corp. is canceling work on its next chip for desktop computers, the latest sign of technical difficulties at the semiconductor company.

People familiar with the situation said that the company will announce Friday that it is scrapping Tejas, Intel's code name for a forthcoming microprocessor, because of problems that are believed to include the chip's power consumption. Tejas was expected to be introduced in late 2004 or early 2005 and succeed Prescott, another internal name for a version of the Pentium 4 chip introduced in February. Engineers working on the Tejas design will be shifted to other chip projects, these people said.

Intel declined comment.

The Santa Clara, Calif., company has been grappling with a transition to a new manufacturing process that reduces lines of circuitry from 130 nanometers, or billionths of a meter, to 90 nanometers. Prescott, the first chip to use that process, has been saddled with unusually high power consumption, an issue that could also affect Tejas.

The most-powerful version of Prescott consumes 103 watts, compared with 89 watts for earlier Pentium 4 models. Higher wattage requires computer makers to add extra fans or take other steps to cool chips.

Dothan, the first chip for laptop computers that uses the 90-nanometer process, was delayed several months because of problems that caused high defect rates. But Dothan, and a predecessor chip dubbed Banias that was introduced last year, use power more efficiently than Prescott. Dothan will be formally introduced Monday, joining the Pentium M family that was begun with Banias.

Intel is believed to be considering desktop chips that borrow from the Pentium M technology. But one person familiar with the matter said the company may focus on efforts to make chips faster by putting the equivalent of two microprocessors on a single piece of silicon."

8) :mrgreen:

Good to see I am not totally crazy :LOL: http://www.beyond3d.com/forum/viewtopic.php?t=10088&start=0
 
Well a P-M, without the focus that it currently has on power savings could easily be clocked MUCH more aggresively (the Dothan is being released starting at 2 GHz), extra transistors from the improved process will likely be for 64bitness and throwing two cores down.
 
Hi Saem

Yes a faster clock could be possible but there are other possibilities for the future like:
a - faster onchip memory controller
b - improved FPU with single cycle MA 8)
 
pascal said:
Hi Saem

Yes a faster clock could be possible but there are other possibilities for the future like:
a - faster onchip memory controller
b - improved FPU with single cycle MA 8)

Don't forget 64bit mode either, those extra registers certainly wouldn't hurt.
 
Here more about future 4-SMP chips 8) http://www.theregister.co.uk/2004/05/05/whitefield_intel_deets/

Intel's Whitefield goes Banias in 2006
By Ashlee Vance in Las Vegas
Published Wednesday 5th May 2004 20:30 GMT
Exclusive The Register has learned more details about Intel's future Xeon processor code-named "Whitefield", the company's first all-Indian design, that we first revealed last week.

Whitefield is a low-power multicore Xeon processer that places four mobile Banias cores around a shared Level 2 cache. The chip will arrive sooner than we expected last week; our sources say 2006. This chimes with predictions made by Intel India's President Ketan Sampat, who said customers could expect a new Xeon out of India by 2005 or 2006 on a 65 nanometer process.

It should be noted, that Whitefield's design differs from what you would expect with Intel's Tukwila version of Itanium. Tukwila will place numerous cores on a single die instead of connecting cores via shared memory.

The design, in theory, is similar to what IBM did with Power4. In that case, IBM combined two dual-core processors with the help of its multichip module. Intel will be using single core chips but similar module packaging. Basically, this means that Intel will be shipping the dual-core Tulsa chip and Whitefield at about the same time. This gives customers a choice between a faster but more power hungry chip and a slower chip aimed at lower-end tasks such as Web serving.

Intel's secrecy behind Whitefield is not surprising given that the company may appear to be lagging rivals with the design. Sun's Niagara chip, also due in 2006, combines up to 8 low-power cores on a single chip. Both Whitefield and Niagara appear aimed at reducing the time it takes a processor to communicate with memory. This is an ideal strategy for software such as Web and application servers that make lots of requests but don't require terribly high single thread performance.

It could be argued that Transmeta and RLX helped pave the way for such products with their success in the low-power blade server market. ®
 
Back
Top