Intel Has Double Vision: First Multi-Core Silicon Production Begins

[Rondog]

Dont the Cell processors that are going to be in the PS3's have somthing like 9 parrellel cores? Or is someone pulling my leg?

[dsio]

We will see what happens as far as cell goes. They have quoted amazing specs, insane scalability, huge performance and are calling it a supercomputer. Apple did the same thing with the G5, and it isnt a supercomputer either. The facts from the engineers and the marketing spin i think are worlds apart on cell, and their ambiguity and smoke and mirror demonstrations are kind of getting a little annoying for my liking. If its good, its good. If it isnt, there might be some disapointed customers.

[Sideburns]

If cell chips are good they're good, if not.... YONAS! Dual core centrino goodness.

Isnt Blue Gene, IBM's supercomp meant to be running cell processors at the moment?

[keithwalton]

It looks as if it is going to be a close run thing between AMD and Intel as to who gets dual core cpu's with them both preping for launch on the same day!
It looks as if it's round 1 to intel thou as the chips are with reviewers and previews / early tests are coming out now. (Anandtech got there system on April 1st! whilst they were on there way to visit amd they got a call up from intel).

Round 2 will be to who can ship them in quantities first. I would actually say intel are more favourable to get this done since they have much greater manufacturing capacity and a mature 90nm process in numerous fabs. Whilst amd are really only just starting to ship 90nm's in quantity and only at fairly low speed grades still.

The final part to the war though will be which chip is faster.
AMD's chip should be quicker than intel's initial offering as long as they can get reasnoble clocks out of it. Since the intel chip is allready at 3.2ghz at launch (compared to 3.8ghz of single core) its allready going quite fast. Amd will really need to hit 2.4Ghz + with there chip from the start.
As by the end of the year / start of 2006 intel will be bringing a proper selection of dual core chips with the yonah and presler. Both have been built on 65nm allready! and are in testing / clockspeed ramping

[dsio]

i would tend to disagree there. Intel's 90nm process is terrible really :\. The prescott was essientially a nightmare for intel and lost them more money and ground to AMD than they were losing with the northwood, which really wasnt a bad core.

AMD's 90nm cores are insane. 2Ghz 3200+ runs at 2.6Ghz stock. No added voltage, no extra cooling. It just does it. Watercooled or heavy air cooled winchesters have hit well over 3.2Ghz but are being bottlenecked by their motherboards and low multipliers.

The only reason AMD hasnt launched 90nm high end chips is because there is really no need or benefit. The single core battle has already been won, and this is symbolised by intel's decision to scrap the 4Ghz Prescott. the 4000+ and the FX series are already on top, so there is no reason to extend the lead. As far as changing the FX's over to a 90nm process, there is no need to. the 130nm SOI FX's dont run hot, dont overheat, and dont fail. If it aint borked, dont unbork it.

What i want to know, is how hot are these smithfields running at the moment, and when we are going to see a performance enhancement. Looking at the reviews on the intel 840 chip, i wouldnt touch it.

[Áedán]

There's also the interesting issue with Hyperthreading and dual core when running Windows. It seems like the Windows Scheduler makes some poor choices when working out which logical CPU code should be running on (see here for a longer description). There's a workaround for this, but it has to be applied by the application developers to each application.

[dsio]

yea, i noticed that. It almost seams a bit like 64 bit processing, which could lead to massive performance gains in windows (code breakers and cryptography enthusiasts think athlon64 is a gift from god) however support and optimisation needs to be done at the software developers end. Shader Model 3.0 is much of the same.

It really needs to be understood that this, like SM3.0 and 64 bit processing is not a "drop in" performance boost. But then again, marketing departments dictate what we know these days.

[Áedán]

64bit processing doesn't lead to a massive performance gain except in situations where one or two things is true.
Firstly, heavy processing of 64bit (or 128bit) data - is better suited to 64bit processors, but most 32bit processors can also combine internal registers to more efficiently handle this data.
Secondly, if your application needs more than 4GB of virtual memory, then you see a performance increase by moving to 64bit processing. There are applications that have seen a speed up of over 40%, simply because they need the extra memory space over the 4GB mark. Unfortunately few of those programs are programs we run on our desktops...

[dsio]

thats kinda what i was implying. cryptography programs in particular (one of my good friends is doing his PHD on theoretical codebreaking algorithms and associated high complexity "beautiful mind" kind of maths stuff) and in some of the programs he uses, the performance gains have been over 100% by using an Opteron instead of his old Xeon SGI system.

This would relate to that first situation you mentioned more than the virtual memory cap and this is running 64 bit unix.

[keithwalton]

Well the 90nm' dothan is a pretty fine chip which runs cool and fast, and well I dont follow the school of thought that amd are not using 90nm on there highend chips because 130nm suffices. For starters 90nm is cheaper to produce chips in volume because they are smaller especially on 300mm wafers. I think they're waiting on fab36 before making any big push.

Just reading the hexus review of the intel chip it does seem pretty clear that anything that has been made to take advantage of hyperthreading gets a hughe boost from DC.
AMD are going to be the ones that benefit more from dual core because intel allready have HT. several tests they did the athlon fx went from first to last when a program was multithreaded.

I'm all in favour of these dual'y chips just as i am with hyperthreading. They may not win benchmarks but in terms of real world usage they are a boon. All day long my computer runs more than one thing at a time and doesnt even flutter unlike my old athlon.

With regards to how hot, well they kick out 130w peak. but due to larger package they can dissipate it better and run at 65c. One thing to note all review i've read so far are saying how stable this platform is as good as anything they've tested before.

The most important question is though. how does it fold

[dsio]

The reason i thought they would not change because there was no point isnt because 130nm chips are cheaper to make than the 90nm chips, but that the cost and effort of beginning production of 90nm top end single core chips when their day has been and gone is overwhelming. Even if they save money by using 90nm on their top end chips, they have to make production changes before it can be implemented. AMD does not sell FX's or 4000+s in huge volumes at all, and given that the cost of switching to 90nm production just to save a small amount per chip on chips that already have huge margins and dont sell many doesnt make sense. They *are* however selling 90nm 3000+'s, 3200+'s, 3500+'s and i believe 3800+'s at the moment.

It isnt that the 90nm chips cant hit FX speed. Mine runs at 2.6Ghz (albiet with a 512kb cache) on 1.4V with a stock cooler 260mhzX10. This isnt an anomaly, this is common for winchesters. 3Ghz is possible with many winchesters at 1.5V (same voltage as 3500+ stock).

When AMD announced the 90nm socket 939 chips, they stated that it would be brought in at the bottom of their range first, before they ramped it up through their product line so they could see how the 90nm chips fared at low speed before they became mainstream. If intel had done that with the prescott, then maybe we wouldnt have 3.6Ghz white box machines thermal throttling during the summer because they cant handle their own heat output.

Given the recent direction change to dual core, i would imaging that this is AMD's new aim in terms of production rather than saving the odd buck on ageing single core products that are priced out of the general public market, and are doing fine on their existing technology.

Edit: btw i wasnt intending to intel bash or AMD fan, rather to bash prescotts (they really are terrible, especially compared to dothan) and support the winchester as ive had alot of first hand experience with these things, and it is the best overclocker ive seen since the 2.4C, not to mention the coolest running and most sophisticated chip ive seen in ages.

[Áedán]

Originally Posted by dsio

This would relate to that first situation you mentioned more than the virtual memory cap and this is running 64 bit unix.

Well, there's also clock speed, memory controller and I/O that's changed. When we're (as in where I work) doing precomputed attacks against various types of crypto, we've found that the I/O efficiency makes a huge difference, depending on the type of attack we're performing.

[keithwalton]

The prescotts are indeed bad, its why i'm still running my 2.4'c' which i've found out is coming up 2 years old now! the longest i've ever run a cpu for! and its going to last quite abit longer yet, as i'm going to wait for the next generation of dual core which gets its stuff from the dothan, (yonah / presler)

[SteveI]

Originally Posted by keithwalton

The prescotts are indeed bad, ...

I have to disagree with that.... I have 5 socket A systems (2 tbred / 2 bartons), and 3 P4's... ( a notebook, 2.8C & 3.0E)... I have the 3.0E o/c'd to 3.6ghz, running dual channel in synch, folding 2cores 24/7. It IS faster than my bartons (2.2-2.3ghz). It only looses a step in gaming benchmarks, but in all other respects is significantly faster. A lot of my apps utilitze HT, and I always have 4-8 windows open, so HT is very useful for me... and a gig o ram doesn't hurt either.

My "preshot" runs 52-58c under load. It is the november04 revision and is much cooler than the original prescotts of last year. I had a prescott from early 04, and it ran over 10c hotter.

Honestly, I'm itching for dual core rigs... I haven't upgraded much lately, but quite frankly, my systems are fast enough. Only when dual cores are mainstream will I have justification to upgrade systems (starting with my own ). Just as a hobby, I'm dying to tinker with something.

[keithwalton]

With regards to precott's being bad, that was in comparison to the northwood's that went before it.

Anyhoo new rumour, rumour has it that dell are about to ship the Precision 380, and the 5th generation Dimension XPS. Both run the intel 955x chipset and dual core chips.
Funny that this pops up on the same day someone gets there mits on an opteron 866