Inside Intel 167
z71offroad writes: "There is a really interesting article at Anandtech right now showing what goes on inside Intel Labs. Although it doesnt break any NDAs, it is still a facinating look at what goes on inside the chip giant's labs."
Re:Interesting (Score:1, Funny)
"Oh... Makin' Daleks"
Re:Man... (Score:1)
http://www.anandtech.com/printarticle.html?i=1584 [anandtech.com]
Should we feel guilty about this tho, since they write such good articles, and miss out on add revenue?
Re:Intel Inside stickers (Score:1)
Misconception (Score:3, Funny)
No, it's synonymous with silicom.999999999999
Re:Misconception (Score:1)
Heheheh..! I needed that!
If any of you moderators where around when intel released the first Pentiums around 93, you know what to do with those moderation points! =)
at Intel (Score:5, Funny)
Re:at Intel (Score:1, Informative)
Re:at Intel (Score:2)
Re:at Intel (Score:1)
Intel's approach (Score:5, Insightful)
I think Intel does need to jump off it's approach to sales by clock speed.
Maybe instead of constantly worrying about clock speeds they spend more research into being able to add larger amounts of cache or try to achieve one clock cycle access to main memory
Yes chips will most likely continue to follow moore's law but computers are not much faster now than 2 years ago
what their worries should be
Re:Intel's approach (Score:2, Insightful)
Re:Intel's approach (Score:2, Redundant)
the Althon has a double pumped 133Mhz bus (equivalent bandwidth to a normal 266Mhz bus)
the P4 on the other hand, has a QUAD pumped 100Mhz bus (equivalent to a normal 400Mhz bus)
there's a reason the P4 lays the smack down in memory benchmarks guys... ('tis a shame it sits around doing nothing with the data it actually gets from memory though
Althons are faster because the core itself is MUCH faster per clock, lets see..
P4 x87 FPU = single execution unit
Althon x87 FPU = THREE execution units
the Althon is something of a brute force design, but it's an ELEGANT brute force design
Re:Intel's approach (Score:5, Informative)
Bzzzt, wrong! Athlons are faster in terms of useful work done per clock because they have a shorter instruction pipeline. Thus their branch mispredict penalty is lower and they have a higher instruction-throughput-rate (ITR) than Intel chips of equal clock speed. Other factors (exclusive L1/L2 cache, lower memory latency, better die space allocation to ALUs and FPUs) influence AMD's higher performance too, but this is the main one. It's just a better balanced processor design, and it certainly yields higher performance for price.
BTW, current DDR memory speed is 2 x 133 Mhz = 266 Mhz, not 233 Mhz.
Re:Intel's approach (Score:3, Interesting)
Well, that's Itanium, and see what it got them... (Score:3, Interesting)
We all know how that turned out, don't we? Fundamentally, Intel is trapped by their own success. They haven't successfully introduced a really new architecture since the i860/i960, and that was YEARS ago.
People don't want "efficient" ot "elegant" processors. They want MegaHertz.
-Mark
Re:Well, that's Itanium, and see what it got them. (Score:1)
Re:Well, that's Itanium, and see what it got them. (Score:1)
" They haven't successfully introduced a really new architecture since the i860/i960, and that was YEARS ago. "
The i860 and i960 were very different architectures, the i860 was a fp number cruncher and the i960 was an early superscalar-ish architecture (co-developed, iirc, with Siemens, or someone else beginning with S, anyway).
The i960 was a beauty to write code for. i860 wasn't.
Re:Intel's approach (Score:2, Insightful)
Seriously, I liked that approach!
I want good performance for the money (Score:2)
Who cares if Intel (or anyone else) makes chip n+1 perform better than chip n through clock cycle speedups, fiddling with cache arrangements, implementing faster-than-light wiring systems on the die :-) . . . whatever. As long as it works faster and cheaper than last year's, what does it matter beyond idle curiosity?
And why not turn lead into gold while we're at it. (Score:4, Interesting)
I'm afraid that both of these (especially the last one) sound like the infamous "let's just find a way to factor huge numbers" quote. That is - yes, it would be wonderful to be able to do this, but there are good reasons for believing that it's very difficult (not that people haven't tried).
For caches, the problem is that larger caches are slower and more power-hungry. To compensate, you use a multilevel cache architecture, but you still have some penalties. A modern foundry could put as much cache as it wanted on to a chip (look at HP's most recent chip for an example) - but because of architectural tradeoffs, this isn't always a good idea.
For memory, if you can find a way to get single-clock access latencies reliably without a 10x slower clock, sell it to $favourite_company and retire on the proceeds. This isn't likely to happen for _two_ reasons. Firstly, modern memory is optimized for density at the expense of speed (this is why we use DRAM and not SRAM for system memory). Secondly, because of the trace lengths, capacitance (and inductance!), and crosstalk and noise issues, it's one _hell_ of a lot harder to send data at low latency _or_ low bandwidth across a motherboard than just within a chip.
There are ways of pushing the boundaries on all of these things, but while we're doing that, processor speeds are still getting faster, putting tougher requirements on the memory and negating most of the relative gain.
In summary, there's a good reason that Intel (along with everyone else) is pursuing more conventional enhancements while background research into memory and caches is going on.
Re:And why not turn lead into gold while we're at (Score:1)
My point was that they need to focus more on researching those fields rather than on clock speed.
Think about what you said. More conventional enhancments from what you said = faster clock speed. Well think about the biggest bottleneck with the computer. Computers today are not limited by clock speed at all. Its memory bottlenecking. Yes there are some improvements in memory but if there werent any, your processor is just going to get to the idle state much faster and sit there much longer.
The idea is about allocation of research resources. IF they spend more money on caching technologies and main memory speed, it will most likely yield a more efficient processor instead of executing so many useless no-ops remaining idle til memory finds its way into the cache. Each level you go up from the registers in teh memory hiearchy access time increases about 10 fold each level, even more as u get to secondary storage.
Whats the use of a processor that zips along with the few instructions it has in cache only to wait wasting time doing nothing. The processor is constantly waiting for the slow as hell IDE hard disks which are just as fast as they were several years ago, main memory, and its own cache.
Now, im not talking about making lead into gold, im just saying its almost obvious if you want technologically superior chips, you need to invest in the obvious bottleneck, not what the marketing department says is most effective.
Re:And why not turn lead into gold while we're at (Score:2)
[...]
Now, im not talking about making lead into gold, im just saying its almost obvious if you want technologically superior chips, you need to invest in the obvious bottleneck, not what the marketing department says is most effective.
And my point is that Intel will market what they expect to be able to put on the shelves next year. *That's* why you hear about their other processor tweaks.
Of *course* Intel is spending money on improving their memory subsystems! But you don't hear about it, because they haven't made any breakthroughs yet (RamBus was the last proposed improvement they invested in, and that failed spectacularly).
It is also in Intel's interest to invest in improvements that are likely to bear fruit quickly. Improving the memory subsystem isn't the only way to improve processor performance, so they're investigating other methods as well.
Re:And why not turn lead into gold while we're at (Score:2)
Oh, I don't know about that. The PA-8700 has 2.25 MB of L2 cache, which is okay I suppose. The MIPS R14000 processors in the SGI Origin 3000 series have 8 MB of L2 cache per CPU, and they do pretty well, to put it mildly. I think your assertion that large secondary caches aren't always a good idea sounds a little weak.
Re:And why not turn lead into gold while we're at (Score:2)
Large amounts of cache benefit niche applications. You see huge caches on server chips because a) they happen to run this kind of niche load, and b) servers are optimized for performance above all else, which means a performance gain of a few percent is worth the cost of adding more cache (while a consumer would balk at paying twice as much).
Cache benefits servers that are running many, many tasks at once - a context switch won't necessarily end up purging the cache if it's big enough. Consumer machines don't usually have this kind of load (or we'd all be running dual-processor machines).
Some scientific applications will benefit, but only some of them. The rest either have access patterns that don't lend themselves to cacheing, or use blocking techniques to increase locality enough that even a smaller cache will be adequate (incremental return becomes low beyond a certain point).
For the vast majority of applications, we're already well into the realm of diminishing returns. Simple proof of this: We've had 256k and 512k caches on consumer chips for quite a while. Linewidth is fine enough that we have the ability to put much more cache on without taking a yield penalty. If doubling the cache size was a sure-fire performance boost for consumer chips, both Intel and AMD would have done it already.
Somebody didn't read the article (Score:1)
They don't work on that assumption. Do you see the Itanium working at 100GHz?
The reason the Pentium 4 has a longer pipeline is because you need a longer pipeline for SMT to work properly with x86 processors. That, and the chip was designed for multimedia, where most of the time the long pipeline is far more beneficial than a deteriment to performance.
I don't care how many GHz it is, I don't care what its IPC is, what I care about is what the GHz * IPC is.
Maybe instead of constantly worrying about clock speeds they spend more research into being able to add larger amounts of cache or try to achieve one clock cycle access to main memory
They already have the technology to add larger amounts of cache. It's just not economical to do so, simply because of all the transistors it takes...
Improve memory speed
Last time I checked, Intel didn't design system memory.
Although it does (by far) have the most bandwidth available to the CPU out of any consumer CPU out there.
Make instruction set more efficient (ie make alu more efficient, the intel is no RISC)
You mean like ditching x86 and designing IA-64?
reduce production costs
Sure, Intel will magically wave a wand and make their production costs drop like a rock. Doesn't work like that.
Re:Intel's approach (Score:2)
Looks like there are advantages to CISC, huh?
This is why, for instance, the JVM has variable-length byte-codes.
And production costs are pretty low, just not for the latest and greatest. Most of the costs of the latest and greatest, however, more so than production is IP.
And how would you suggest that Intel, as a processor/chipset manufacturer improve memory speed? they don't have the resources (mainly intellectual) to contribute much.
Also, faster clockspeeds seem to be working for Intel. Obviously the P4 isn't great yet, but it has at least one killer feature (hyperthreading/SMT) turned off yet. And it's scaling more incredibly than any processor we've seen before.
-Dan
Re:Intel's approach (Score:1)
so I'd argue that it's not scaling quite so well for performance.
Re:Intel's approach (Score:1)
Do you not think that with close to $30 billion a year at stake, Intel's engineer's just might have considered those things and optimized them to the Nth frickin degree?
They optmize profit, and they do a damn good job of it. I would imagine that the profit equation has hundereds, if not thousands of parameters that are wildy complex and interrelated.
Re:Intel's approach (Score:1)
Intel Labs (Score:5, Informative)
It has some pretty interesting info regarding what goes on around Intel.
Did we really need a /. article on this?
They left off two (Score:3, Funny)
but that was before they decided to get rid of the dancing guys
in the shiny bunny outfits.
1)Performance
2)Power
3)Integrity
4)Functionality
5)Tools and Methods
6)Originality
7)Choreography
My favorite picture (Score:5, Funny)
Where do I sign up? (Score:5, Funny)
All I gots to say is how do I become a CV Engineer. Getting payed to "test" the stability of chips during games.
Uhh, no I dont think 20 hours of straight counter-strike is rigorous enough, we should do at least 20 more, for quality purpouses.
No, seriously I need a job!
Re:Where do I sign up? (Score:1)
Do no work, play games all night
/mlk
Re:Where do I sign up? (Score:2)
lamer
Re:Where do I sign up? (Score:1)
Become a lowly Green badge today! (Score:1)
I worked in Intel's Dupont, WA CV labs.
Sign up for a 1 year contract through CDI (Do a search on Monster for CDI and check the job listings with Intel lab locations)
Quite honestly the job can get quite boring.. Yes sure there is fun stuff like fiddling with tearabyte san systems, "testing" DVD's by watching movies on 37" monitors and Itanium boxes with 64 GIGS of memory but by the time you have loaded Whistler so many times you can recite the CD-Key from memory it gets a bit boring.
Imagine this as a work day.. Your boss hands you a case of hard drives and a stack of SCSI cards.. Your job.. load up 5 drives each in 3-4 Itanium workstations, load Whistler on each one.. run a list of tests.. load Linux on em.. run more tests.. repeat untill you have tested the whole case of drives.. repeat the entire process with 3-4 brands or SCSI cards.. then go get the RAID cards.. repeat again.. Next week you get a new case of drives..
Guess what.. same thing all over again. Sometime in the middle wander down to the other side of the lab and chat with they guys hackin linux kernels to run on the new Mckinleys.
Still I learned a ton and got to play with next years toys last year.
PR fluff dressed up as engineering cred (Score:5, Interesting)
th very first sentence in this article states th perception th article is focussed on diminishing
really? - as a for-profit company, perhaps their shareholders might be interested in them making maximum profit as well?
and who is this 'we' - only a single authour is mentioned at th top of th article - or perhaps his name has simply been appended to a pre-prepared puff piece?another example of rhetorical writing pulled from th first few paragraphs
very talented engineers [who] are focused on pushing the limits of technology
ok - there may be real information contained in this article - but frankly there were enough warning signals in th first few paragraphs to tell me my time was better spent elsewhere
Re:PR fluff dressed up as engineering cred (Score:1)
I agree with the parent post on this. No one is in a for-profit business for the betterment of mankind. They do it to fatten their pockets. You just hope that they appreciate that doing quality work, and excelling in your field is the way to long term success. Marketing will always work in the short term, but eventually people begin to see through the wool, we hope.
I think your missing the point (Score:1)
"another example of rhetorical writing pulled from th first few paragraphs
very talented engineers [who] are focused on pushing the limits of technology"
And while it may be teeny bit fluffy per your above quote, Anand is certainly not going to write how the Intel engineers are know-nothing smelly idiots. Christ they have a relationship with them and just maybe might need that P4 3GHz when it comes out. And no I don't think this makes Anand a non-credible source for cpu/hardware reviews. If you read any of them you know they often praise AMD.
Geez lighten up.
Re:PR fluff dressed up as engineering cred (Score:1, Offtopic)
guess it's time to add one to th list [fnord.org]
fnord is th reason there's no 'e' in th word 'the'
Driven by the Engineers? (Score:1)
Theres nothing interesting in this post, just a view from Portland about what I hear coming out of Jones Farm and Ronler Acres.
Actually, the most revolutionary thing at Intel is the strides they are taking in the field of fab design.
Another inside view (Score:1)
Re:Driven by the Engineers? (Score:2)
10 Ghz ALU = 5Ghz CPU (Score:4, Informative)
And you are still wondering why Pentium 4 is still slower than the Athlon (or awfully close)
Imagine what would happen if the ALU is only running at the same speed as the CPU.
Personally, Intel is losing little ground at a time right now, but remember, Intel can afford to make a couple of mistakes but AMD can't even afford to make on. One mistake will push AMD back to the bottom, again.
Close, but no cigar (Score:4, Interesting)
The 10GHz ALU was a very limited ALU, not part of any modern processor.
Intel is losing little ground at a time right now
Actually, in Q4 2001 Intel gained market share and AMD lost some. But overall in 2001, AMD did gain market share, that's true.
I still think it's because Intel wants to point to AMD and say "See? Competition!".
Intel could easily release faster CPUs right now to totally crush the Athlon, but it doesn't make sense to do so.
Re:Close, but no cigar (Score:2)
Re:10 Ghz ALU = 5Ghz CPU (Score:2)
Imagine what would happen if the ALU is only running at the same speed as the CPU.
You'd have a 16 bit processor?
No, seriously, they are going for a ALU that works on 32 bits per clock, like the one demo'ed. But, I'd be just as happy with a 20 GHz double pumped 16-bit ALU as I would a 10 GHz single pumped 32-bit ALU
AMD Still Has Upper Hand (Score:4, Informative)
Re:AMD Still Has Upper Hand (Score:2, Troll)
Let's drop the act.
Intel CPUs are just as reliable as AMD, if not more. AMD CPUs rely on motherboard logic to stop overheating and subsequent fires/melting, Intel CPU's have them built in. AMD CPUs are FAR more likely to crack/shatter installing HSFs due to how the sockets are designed, too. Your reliability argument is simply nonexistant.
As for performance, there are instances where the Pentium 4 is king, there are instances where the Athlon XP is king. There's no clear performance leader.
Athlons are cheaper, but take some shortcuts. Still damn fast, still damn cheap. Lots of people love that.
This HSF thing is pure BS in my opinion... (Score:2)
Kjella
Re:This HSF thing is pure BS in my opinion... (Score:1)
Wireless? (Score:2, Funny)
You just got to love all those wires for the wireless network =)
really (Score:1, Interesting)
Seems like it could be a security nightmare... (Score:2)
"There is a strong focus on networking and more specifically wireless
networking at Intel. Intel's campus alone is entirely wired for wireless
internet access for their employees."
I was somewhat disappointed, with the article (Score:2, Insightful)
Mostly the article seemed to run like the following, well we met some real cool guys in a lab in oregon who research all the new stuff, then we followed all the production to the guy's who QC the design at some other lab somewhere. While trying not to break a NDA
I thought this was a bit like going to Nike and just interviewing the guy who made the prototype for the new gel heel. Or Ford and interviewing the guy who made the new concept Focus while ignoring everyone from Detroit
Not that am trying to troll, I just wanted some interviews with the average workers at their Indonesian fab plant, maybe finding out how what kewl shit they suppy to their employees. I heard that one plant had a fully working video parlor for instance, and that intel had helped to setup schools in the local area (although this is prob marketing BS). Also, rumour has it that people who work there are all closet overclocking freak's since the price of the chips themselves are so low.
I also thought that it might be interesting to see the who hugh some of these factories have become, that in some cases they are as large as small cities, with entire regions depending on there income.
Throw, in some interesting facts like, most of the chinese in indonesia are not normal chinese but rather Hacka, or chinese gypsies who moved to the country due to persecution on the mainland.
Things like that, together with reports from the labs, would have made a much more interesting article IMHO.
Re:I was somewhat disappointed, with the article (Score:1)
Yeah, that would be interesting...
Do you have any food? I am so hungry! No please
dont beat me, not the stick, i work harder!
Re:I was somewhat disappointed, with the article (Score:1)
Yeah, I know that companies who outsource their labour have a rep for this kind of stuff. However, I think that people in those countries are a lot better off then they would be (The alternative being nothing). You have to remember that countries like America started off this way themselves. Like I said, I wish it was something that the article had looked into and didn't.
This is not to discount companies like Nike who have a really, really bad rep for policies like this. If you don't like it then don't buy their product.
Although, I think that people in these countries have a lot more power then they do in America due to one simple fact: People. If you close a plant in America, you know that all the people who worked there can always get other jobs or go on welfare. Do the same thing over there and you have thousands of desperate people who you are denying their sole means of income with nothing to lose.
If you don't believe me then consider this: Run Amoke is a Indonesian saying.
BTW, In economics, my teacher said "The reason companies set up factories in third world countries is because they wish to sell to the people in those countries someday" I think he would know what he's talking about.
Re:I was somewhat disappointed, with the article (Score:1)
Even if you wished to argue this line, there was a lot better ways of putting it than "My teacher knows best".
Companies set up factories in low wage areas because they want to lower costs. With a lot of industries I'm sure there is some desire to sell to the citizens of the country (e.g. comparatively low cost goods such as Cigarettes) but mainly it's about costs.
Dyson (the UK vacuum cleaner manufacturer) recently announced moving part of it's manufacturing to Asia. Why? Low labour costs (lower wages and associated costs). I'm sure as an associated benefit they will sell more of their product in Asia, but this is _not_ their primary motivation.
Re:I was somewhat disappointed, with the article (Score:1)
I dunno bout that one, true labour costs can be lower, but most manufactoring deals tend to work by the factory, ie you have a factory of around 200 ppl rather then a factory of 20 like you would in America or England things tending to equal themselves out,
You also have shipping costs and more importantly time considerations to factor in as well as the possiblity of having your stock rot on the pier due to a strike.
There are also currency considerations to factor in with most stuff being priced in American dollars.
In short, things never turn out to be as simple as what people make them out to be. I think that selling to the natives to be a better argument then dealing with the headache of importation. After all you set up your factory as close to your punters as you can get away with.
I'm quite glad that this thread isn't turning into a globalization flamefest, where dude's say stuff like "we don't like what nike does to its workers" when what they really mean is "we don't what Indians wearing trainers, they should stay in the stoneage just cause we say so".
Your right I could have put the argument better, I still think he had the right idea, though teachers tend to take pretty weird views, like that, just to prove how clever they are.
Re:It's just a FIB (Score:1)
"Watchout AMD, Intel can FIB! ...they also have a few soldring irons too" :-)
Inside Intel... (Score:3, Insightful)
Let's not forget to take a complete look inside intel..., not just at there technology.
FaceIntel.com [faceintel.com]
Nice company.... NOT!
Interesting Commercial (Score:1)