I decided to start another thread, since my other one got hijacked. I've read some editiorials about Intel's decision to can Tejas. They point to the cancellation of the Tejas, and the speeding up of the development of dual-core Pentium M, signals the end of the philosophy more computing power by increasing clock speed, and the beginning of increasing power by parallelism. Some point that that point is 90nm. Anyway, I'd like to start the discussion by saying that perhaps Sony was right to pursue Cell technology. And did MS know of the problem intel was having with Pentium4?
Significance of the cancellation of Tejas
- Thread starter bbot
- Start date
This really belongs in the hardware section but i will let it stand .
Personaly I believe the future is both.
Faster clock speeds , better ipc , multi core. THe p4 design was allways bad. Intel has most likely known this but the p3 at the time was dead . I honestly though and hoped they would go with amd for the xbox 2 .
ibm could have produced the chip. Mabye even the dual core k8s . This would have help amd a ton as they are really having a hard time and could use a 64bit x86 push
Personaly I believe the future is both.
Faster clock speeds , better ipc , multi core. THe p4 design was allways bad. Intel has most likely known this but the p3 at the time was dead . I honestly though and hoped they would go with amd for the xbox 2 .
ibm could have produced the chip. Mabye even the dual core k8s . This would have help amd a ton as they are really having a hard time and could use a 64bit x86 push
Someone from IBM claiming that scaling is dead starting at around 130nm.
http://www.eetimes.com/semi/news/showArticle.jhtml?articleID=19502091
So Ken Kutaragi was right to adopt Cell, don't you agree? (Aren't you happy Vince?)
http://www.eetimes.com/semi/news/showArticle.jhtml?articleID=19502091
So Ken Kutaragi was right to adopt Cell, don't you agree? (Aren't you happy Vince?)
bbot said:
No i don't really agree at all . Even for the almighty cell chip its going to come down to clock speeds . If they can't clock the cell chip high enough then it wont do well
There needs to be a good balance. The p4 isn't that balance and the cell chip may not be that balance either
Given the current info, do you really think AMD could offer economical solution at that level of performance? Triple core at 3.5Ghz is well, exceptionally high clocked even by P4 standards...
On related note, if Sony projections for 4ghz also come true, consoles are actually gonna be on top or above the Mhz curve relative to current desktop cpu roadmaps (P4 excluded), which I think would be a first time ever to happen.
Which would be less interesting on its own, if it weren't for the very high IPC of both these 'cpus' as well. Looking at it like that you can see why DM is so skeptical.
I have to agree with Faf.
There might be an interesting trend, it was sort of forecasted but never really came about, but that's not due to entirely technical reasons.
Even though IBM and R&Ders of RISC ISAs don't sink as much money as Intel does into an x86 MPU or even AMD most likely, they're starting to come out ahead. You can point to legacy, but that wouldn't be an understatement. I think it's come to a point where x86 is far to expensive, not in terms of money but in terms of time.
I think we've reached a point where, the amount of time one must spend developing an x86 MPU and getting it to run very fast just can't be compensated for by the economies of scales its manufactures leverage. The advantage is/has exhausted. The "easiness" of RISCy (not EPIC, that's TOO RISCy, j/k
architectures are developed quicker, hit their performance marks with greater ease and the time saved can be spent on layout and process improvements, which yields high clock rates.
Mind you there is the fact that IBM is leveraging previous work, so their investment in the MPUs being developed aren't that small, but then one must also consider the time invested by "competing" x86 products.
There might be an interesting trend, it was sort of forecasted but never really came about, but that's not due to entirely technical reasons.
Even though IBM and R&Ders of RISC ISAs don't sink as much money as Intel does into an x86 MPU or even AMD most likely, they're starting to come out ahead. You can point to legacy, but that wouldn't be an understatement. I think it's come to a point where x86 is far to expensive, not in terms of money but in terms of time.
I think we've reached a point where, the amount of time one must spend developing an x86 MPU and getting it to run very fast just can't be compensated for by the economies of scales its manufactures leverage. The advantage is/has exhausted. The "easiness" of RISCy (not EPIC, that's TOO RISCy, j/k
architectures are developed quicker, hit their performance marks with greater ease and the time saved can be spent on layout and process improvements, which yields high clock rates.Mind you there is the fact that IBM is leveraging previous work, so their investment in the MPUs being developed aren't that small, but then one must also consider the time invested by "competing" x86 products.
Fafalada said:
Hm, Vectrex had a rather nippy CPU for its time, the 68A09 @1.6MHz. This chip outclassed the 1MHz 6501 chip found in the C64 (well, I think it was the 6501, there are so many variants!
), as well as other contemporary processors. Not sure if it was exactly leading-edge in 82 when it debuted (or much less bleeding edge which a 4PU BB engine @ 4GHz would be), but it was a damn fast chip for an 8-bit processor.More on it here: http://encyclopedia.thefreedictionary.com/Motorola 6809
Not totally sure which chip(s) you refer to by "both" here, BB and the rumored 3-core PPC-based XCPU2? If "yes", it may be noted the PPC970 architecture does have a very high theoretical IPC, but its design pretty much prevents it from ever coming close to that level. Ars has a nice blackpaper on the chip, and there are many built-in gotchas in how instructions may be/are required to pair up to be dispatched. Luckily this is all done in hardware, but if it wasn't, it is complicated enough to give a programmer a migraine headache just trying to write a "hello world" program in assembler!
It's still fast, just not crazy fast like the theoretical numbers would imply.
Was there ANY contemporary PC whoose CPU didn't outclass C64's?
But no, Vectrex wouldn't really be cutting edge - Z80 has dominated the PC market long before 1982, and at higher MHZ. In 1982 it made it into the first home machine (Spectrum@3.5Mhz). Same year first IBM PCs came out with 8088s (iirc clocking around 4mhz)... etc.
Well I was talking IPC per chip, if XCPU is PPC970 based, you get 3x the IPC
Fafalada said:
However, the 6809 uses lots more of hardcoded rather than microcoded instructions, so it runs much faster per clock than Z80, or indeed, 8088. It also has a multiply instruction, does these other chips have that?
Well I was talking IPC per chip, if XCPU is PPC970 based, you get 3x the IPC
But still only in theory, and in practice it will be even less since they'll be sharing the same 1MB L2 if the leaked stuff is to be believed, and the L2 will likely be single-ported, so only 1 CPU accessing at a time... (Anything else might lead to concurrency issues I'd wager, and lots more complexity).
overclocked
Veteran
Jvd, well i also hoped they would go with AMD as they/almost did with Xcpu.
But i totally disagree with AMD having a hard time, as it look´s now they seem to get almost all right.
Faf i don´t think we will see a 3,5Ghz cpu in XB2.. :?
If MS is going for a triple-core jobbie, it's going to be difficult to find a use for those chips that while are working, do not hit the clockspeed target MS is aiming at (whatever that might be). Either they will go to waste (thus driving up the price of the chips that do work at the required speed), or IBM will have to try to sell them off as some kind of weird embedded CPU or such I'd think.
I wonder why Commodore didn't clock it higher. Almost all the other 8bits at the time had their 6502 variant running at ~2Mhz. Even the four year older Atari 800 was faster.Fafalada said:Was there ANY contemporary PC whoose CPU didn't outclass C64's?
On the 6809, and why it didn't end up in the macintosh (~ a third through the page): http://www.aci.com.pl/mwichary/computerhistory/articles/macintoshsotherdesigners
edit: forgot to paste the link
Re: Significance of the cancellation if Tejas
It would be intresting to get Peter Glawosky (not sure of the spelling) to comment on Intels change of mind on its roadmap. He is the guy from the Microporcessor Report, Deadmeat used in his signature stating something to the effect "CELL isn't a big deal, just marketing hype". I think he also eluded to that Intel was alreadying working on something similar to CELL. Maybe someone should email him this thread and maybe if he has some time he'll comment.
bbot said:
It would be intresting to get Peter Glawosky (not sure of the spelling) to comment on Intels change of mind on its roadmap. He is the guy from the Microporcessor Report, Deadmeat used in his signature stating something to the effect "CELL isn't a big deal, just marketing hype". I think he also eluded to that Intel was alreadying working on something similar to CELL. Maybe someone should email him this thread and maybe if he has some time he'll comment.
Re: Significance of the cancellation if Tejas
The Microprocessor Report lost, IMHO, all credibility after that:
Why?
Direct comparison with desktop situation doesn't (really) apply, Intel are stuck at 3GHz since.... a long time ago.
Off Topic:
Sure, their new CPU is multicore and runing @2GHz (Jonah), but that doesn't mean another architecture multicore cpu can't run higher.... Hey, it doesn't even mean that the intel Jonah architecture can't run higher than 2GHz (except in some "twisted minds"), it's all about supply and demand, if the 2GHz intel Johan can deliver more computational power than the competion can offer, so why Intel would have to play all their cards at once?
MS on the other side have to deliver a machine that must stay competitive until 2010, and that would have to compete with a processor that is trying to push 1Tflops... But i'm digressing.
Brimstone said:
The Microprocessor Report lost, IMHO, all credibility after that:
PR said:
overclocked said:
Why?
Direct comparison with desktop situation doesn't (really) apply, Intel are stuck at 3GHz since.... a long time ago.
Off Topic:
Sure, their new CPU is multicore and runing @2GHz (Jonah), but that doesn't mean another architecture multicore cpu can't run higher.... Hey, it doesn't even mean that the intel Jonah architecture can't run higher than 2GHz (except in some "twisted minds"), it's all about supply and demand, if the 2GHz intel Johan can deliver more computational power than the competion can offer, so why Intel would have to play all their cards at once?
MS on the other side have to deliver a machine that must stay competitive until 2010, and that would have to compete with a processor that is trying to push 1Tflops... But i'm digressing.
Erhm, I don't know how those cpus performed relative to one another, but I am pretty certain that they all used hardwired instructions. Microcode first waded into x86s inside pentium clones that used to to translate instructions for the risc core - which became common practice for all x86 cpus later on.Guden Oden said:
But yeah, although my memory is hazy I'm pretty sure z80 didn't have a mul instruction. Dunno about 8088 - I know i186 instruction set included it, but 8088 was i86.
Edit:
Ok I noticed that motorola page says Z80 was using microcode. Perhaps I remember it wrong, but I'm still skeptical.
Oh sure, efficiency is another matter. But we won't know about that for at least another year so...
When was N64 announced?Cybamerc said:
486 cpus broke 100mhz in 1994 (120mhz), as did Pentiums (100mhz). Note that version of R4000 used in N64 was not really on the same level with P5 class CPUs though - it would roughly compare to 486s, which basically made it a generation behind, regardless of clock speed.
By the time N64 was out, things were already approaching 200mhz, and there was PPro and MMX out...
We could also note there were other desktop processors outside x86 that were pretty fast back then, but maybe Archie or someone can help with those timelines, I'm not really sure.
Fafalada said:
You're thinking of microOPS (or variations thereof). Microcode is quite ancient. MC68000 instructions are built using microcode for example, at least the more complex ones. For example, 68010 shaved off like two dozen cycles from its mul instruction and even more from the div if I remember correctly (68000 division is over 200 cycles, again if I remember correctly).
The Alphas were damn fast back in the day as far as clock speed goes, and still pretty nippy in absolute performance numbers too. They definitely had x86 whupped, that's for sure.
Fafalada said:
No i don't believe that amd could. But ibm could .
Ibm and amd share fabs and tech.
I was hoping that they would liscense a dual k8 or mabye even a k9 if rumors are true and the chip is nearing completion . Have ibm fab the chips for them .
That is what i would have like to have seen. I think ms may have went down the better path for them in the console land but not for the big picture
Megadrive1988
Veteran
In August or September 1993, as Project Reality.
though it wasn't named Nintendo 64 until 1995 or 1996.
