Bad News for Intel--Whitefield cancelled
You know the new dualcore Intel stuff: Yonah/Sossaman early 2006, quickly followed by Merom/Conroe/Woodcrest, H2 2006. These processors are a lot about energy-efficiency thus laptops, a little for desktops, and lame servers. Not very high-end, i.e. workstation-class (Power Mac).
The Whitefield processor would have been a very powerful x86 chip planned for 2007, succeeding Woodcrest, with a multicore design (2 cores and 4 cores models, 65 nm then 45 nm early 2008 ), part of the new "Nehalem" family. The cores would have communicated each others directly via a high speed internal bus, and with the chipset via a wide HF dedicated bus called CSI (Common System Interface), sorta AMD HyperTransport. The chip would also had an integrated memory controller and 16 MB of L2 cache.
Dunnington would have been the successor of Whitefield in 2008, with 4 to 32 cores and initial 45 nm manufacturing process.
But Intel has just officially cancelled Whitefield. They defer the project to... 2009!
So in 2007, they will launch another processor, codenamed "Tigerton", succeeding "Cloverton" (which is 2 Woodcrest dies on one module).
BTW... "Tiger Ton", I find it quite funny
Tigerton will lack the high-speed CSI and the integrated memory controller, its general design will be less attractive than expected with Whitefield. But at least it will feature a speedy internal dedicated inter-core bus.
Power Macs already have HyperTransport with G5 motherboard. I think the crippled high-end x86 Intel projects for the near future and the lack of any HT-style bus for Xeons in 2006, 2007 and even 2008 (!) is a BIG PROBLEM for Apple in the Intel-based workstation category. Apple may (re)consider AMD Opteron though.
Sources :
- The Inquirer
- News.com
- and many other IT web sites
The Whitefield processor would have been a very powerful x86 chip planned for 2007, succeeding Woodcrest, with a multicore design (2 cores and 4 cores models, 65 nm then 45 nm early 2008 ), part of the new "Nehalem" family. The cores would have communicated each others directly via a high speed internal bus, and with the chipset via a wide HF dedicated bus called CSI (Common System Interface), sorta AMD HyperTransport. The chip would also had an integrated memory controller and 16 MB of L2 cache.
Dunnington would have been the successor of Whitefield in 2008, with 4 to 32 cores and initial 45 nm manufacturing process.
But Intel has just officially cancelled Whitefield. They defer the project to... 2009!
So in 2007, they will launch another processor, codenamed "Tigerton", succeeding "Cloverton" (which is 2 Woodcrest dies on one module).
BTW... "Tiger Ton", I find it quite funny
Tigerton will lack the high-speed CSI and the integrated memory controller, its general design will be less attractive than expected with Whitefield. But at least it will feature a speedy internal dedicated inter-core bus.
Power Macs already have HyperTransport with G5 motherboard. I think the crippled high-end x86 Intel projects for the near future and the lack of any HT-style bus for Xeons in 2006, 2007 and even 2008 (!) is a BIG PROBLEM for Apple in the Intel-based workstation category. Apple may (re)consider AMD Opteron though.
Sources :
- The Inquirer
- News.com
- and many other IT web sites
Comments
Originally posted by Anders
So the Apple curse finally hit Intel
It was about time! I wondered how long would it take
Originally posted by Anders
So the Apple curse finally hit Intel
A brilliant move on Apple's part. Announce support for intel, put it off until some later date, and watch intel self destruct. In the meantime IBM, free of Apple's curse, suddenly finds a way to create a 45nm, 15W, 4GHz G5 and Apple says "Oh, I guess we'll stick with PowerPC after all."
It will be interesting to see if IBM can deliver anything that would keep the PPC in some Macs for a while longer. (And Freescale in terms of the dual core G4.)
CSI being cancelled sucks but perhaps Intel has found that they don't really require it with some of the memory tech like FB-DIMM set to improve memory performance making CSI a negligble benefit.
Originally posted by Fluffy
A brilliant move on Apple's part. Announce support for intel, put it off until some later date, and watch intel self destruct. In the meantime IBM, free of Apple's curse, suddenly finds a way to create a 45nm, 15W, 4GHz G5 and Apple says "Oh, I guess we'll stick with PowerPC after all."
Hahahah - A 10 year old dream comming true!
I don't think any Apple-Intel contracts can help us lowly customers; if Intel can't produce good enough chips a contract can't magically fix them.
Intel are doomed!
Originally posted by vinney57
Hey! Lets all overreact to the cancellation of of some event that might have happened in the future!
Intel are doomed!
There sure seems to be a concerted effort by some to inject a little FUD in Mactel. Frankly I'm tired of PPC promises and ready to see what Intel can do for me. I drank the koolaid back in the day but today it seems so abundantly clear that it's a two pony show with Intel and AMD providing the entertainment.
Whitefield simply wasn't going to be something that Mac users were going to bank on. Kind of reminds me of the PPC 620 that never really saw the light of day. Sometimes the Big Iron stuff just doesn't live up to the hype and it gets scrapped.
Originally posted by hmurchison
There sure seems to be a concerted effort by some to inject a little FUD in Mactel. Frankly I'm tired of empty PPC promises and ready to see what Intel can do for me. I drank the koolaid back in the day but today it seems so abundantly clear that it's a two pony show with Intel and AMD providing the entertainment.
TFTFY
Originally posted by hmurchison
I think people have overrated Hypertransport a wee bit as well as ondie memory controllers. I love the low latency but the advantages of a OMC aren't as prevalent when you take straight throughput into account.
CSI being cancelled sucks but perhaps Intel has found that they don't really require it with some of the memory tech like FB-DIMM set to improve memory performance making CSI a negligble benefit.
It really is quite a substantial benefit. I'd say they dumped it simply because they couldn't get it to work and it's time to cut their losses and bring something in for that timeframe.
Something Intel really hasn't been doing well as yet is using the extra silicon in innovative ways to try and gain a leg up. They really are just pushing the same old recipe, which is great when it works, but they stand a chance of losing out a bit.
- "Intel Revamps Xeon Plans to Boost Performance" at Xbit Laboratories
- "Intel announces major 2006/2007 roadmap changes" at Ars Technica.
You can see that Xbit Labs is slightly more optimist than other sites with the changes announced by Intel. They claim the new planned "dedicated high-speed interconnect" to the whole chipset with Tigerton's "Caneland" platform is a) far better than with todays Xeons, and moreover b) could be even better than CSI in some case.
In fact, as for the cancelled Whitefield "Redland" platform, when the configuration would be a mutiprocessor system, each chip was supposed to share the same processor system bus with the other(s) (CSI). Whereas each Tigerton processor is expected to get its dedicated interconnection to the rest of the system. (OK, it's a little blurry, since we don't know any comparison basis between the different theroric data rates yet).
Ars implies that CSI was perhaps vainly intended to unify the design of Xeon and Itanium sockets, in order to have the same motherboard booting "either a Xeon or an Itanium processor with only a BIOS tweak". Maybe this was scrapped due to tehcnical difficulty and high development costs.
They had this rumour since last year:
« The project is being tackled at Intel India and Intel US has flown a number of validation staff and architects to help out with the problems with Whitefield (WFD). As the INQ reported earlier, WFD uses the Tukwila system interface but while that is IA64, Whitefield is X86 based and there are some big differences in approach.
There are serious problems with WFD's architecture and the validation environment also leaves a lot to be desired. Latency numbers are also rather worse than Intel engineers had hoped and the chip giant could be forced to cancel the project, except that there's no obvious candidate to take WFD's place. »
Hope Tigerton will also feature 4 cores as Whitefield.
Originally posted by ghiangelo
why are we all assuming Xeons are going to be used in Mactel? we might be lucky if we get even a P4...
Because ordinary processors from Intel don't support multiprocessing environments. I doubt Apple is going to drop all their shipping products down to single processor only, especially not the Xserve. Apple will be using Xeons in at least some of their lines and I'd be shocked if a P4 ever sees its way into a Mac outside the developer boxes.
Originally posted by Telomar
Because ordinary processors from Intel don't support multiprocessing environments. I doubt Apple is going to drop all their shipping products down to single processor only, especially not the Xserve. Apple will be using Xeons in at least some of their lines and I'd be shocked if a P4 ever sees its way into a Mac outside the developer boxes.
hope that's the case... it's just that Apple's Power Mac offerings have historically been less than workstation level. the G5 was the first high end workstation type PM in the last 7 years. it was only a short while ago that PMs had a 1.5 GB RAM maximum! this at the time a basic home use Dell offering had 4 GB max MBs. that's why the G5 wasn't just a new power Mac, it was a quantum leap from the previous G4 systems.
Originally posted by ghiangelo
[snip]this at the time a basic home use Dell offering had 4 GB max MBs.[/snip]
Originally posted by Hiro
TFTFY
What? When did we become Arstechnica?
Originally posted by Telomar
Because ordinary processors from Intel don't support multiprocessing environments. I doubt Apple is going to drop all their shipping products down to single processor only, especially not the Xserve.
No, but they might drop them all down to a single chip. More than one chip is expensive and unnecessary if all the cores you need are on one chip.
Originally posted by Anders
What? When did we become Arstechnica?
Originally posted by hmurchison
Kind of reminds me of the PPC 620 that never really saw the light of day. Sometimes the Big Iron stuff just doesn't live up to the hype and it gets scrapped.
The 620 was (is) real and was sold by Bull (Escala). At that time the benefits of 64bit were very limited, so maintaining a 64 PPC line didn't make any sense.
End of Line