Slashdot Mirror

moorespaper.pdf

"Cramming More Components Onto Integrated Circuits"
(Acrobat PDF file, 167 KB)
Author: Gordon E. Moore
Publication: Electronics, April 19, 1965

ftp://download.intel.com/research/silicon/moorespa per.pdf

In the last application I was writing for a client that used graphics routines I downloaded Intel's performance primatives libraries and they worked very well. Very fast, quite impressive. I'd love to see AMD release something similar and link against either dynamically from within graphics code.

Here is Intel's press release announcing their WiMax product.

I can think of one reason DELL may be willing to play keep away with AMD - an exclusive supplier agreement.

It isn't illegal to have one and it may be a key into DELL's low costs. By negotiating large contracts with vendors from graphics card to CPUs they can get only such a price break, but if DELL, the world leader in desktop computing, agrees to use your computer widget exclusively, well that is worth another penny or two per thingamajig, which adds up to $M for DELL's net profits.

Intel is savvy enough to get into this kind of pact with DELL. Here's a slighylt off topic story but it speaks to Intel's marketing wit.

Back in the 80's a marketing study was done to see what IT names people thought of when they thought Personal Computers. The usual suspects were there of course, Microsoft, IBM, even Apple. But strangely absent was the one company who's hardware was arguable at the heart of ever IBM compatible PC in the 80's - Intel. No one knew of them and when customers don't even know who you are, you risk being replaced by competition.

So Intel embarked on the "Intel Inside" campaign. They came up with the sticker that adorns just about every PC with an Intel processor as well as the doo-doo-doo tone we hear at the end of DELL, Gateway, HP, and other computer maker commercials.

But how did they get manufacturers to add stickers and 3 seconds of their 30 second commerical on Intel? With Money! Intel offered to share the cost of TV ads, radio, and print if manufacturers would put the sticker on and have the logo and doo-doo-doo tone on EVERY ad.

The campaign worked. Intel had another study done X months after the campaign launched and they shot up to one of the top on the list.

You can find out more from Intel and Google.

http://www.intel.com/netcomms/columns/jimj105.htm (doh! fixed link)

Press Release

Wiki Article on WiMax

Doing the editors jobs so they don't have to!
--
wdd

There are a few possible solutions you might want to look at for a big-RAM server. Now, if you really want 64GB and AMD Opteron processors than you really only have one choice, the HP Proliant DL585. That's the only Opteron solution that I know of which supports 64GB of memory.

If you can get by with a bit less memory then you have some other solutions. Tyan carries quite a number of boards with varying capabilities. The trouble here is that the Opteron processors are limited to 8GB of memory per processor, so to get 16GB you're going to be looking at a dual-processor board (quad processor for 32GB). Since the memory controller is right on the CPU with the Opteron you will actually need a second processor in the socket to use this memory.

For this reason, you might actually want to consider one of Intel's new 64-bit Xeon chips. I know that Supermicro offers some boards that can handle up to 32GB with only a single Xeon processor. Something like the X6DHE-XB seems like it might fit you're bill reasonable well. Fairly inexpensive to get you up to 16GB of memory, though going to 32GB is quite expensive. Crucial has a list of compatible memory for this board, including some 4GB modules.

Of course, if you're not limited to x86 systems then there are other solutions that would work. You could get something like an IBM Power system or Sun UltraSparc system with pretty much any amount of memory you need (or can afford).

Hyper-threading on a dual-core will be possible with the upcoming Intel Pentium 4 Extreme Edition dual-core. But with 4 virtual processors, that is going to require some SERIOUS cooling.

A simple search for moores law on google will give you this page which includes the original paper. Though the original paper is "dumbized" so it doesn't include all the nitty gritty details so . . .

There is a web site that I went to for a more detailed description than he actually put in his paper, but I don't know the URL anymore.

However, whether it was numerically correct from the start may not be the important part. How about just nailing down the shape of the curve? Isn't that worth something by itself?

On the other hand, does anyone actually have a graph of transistors per chip, or transistor size plotted against time, covering the past 40 years? That is, is anybody checking the numbers?

I guess I can do my own Googling:

Gordon's graph paper that shows cost versus number of transistors per chip

Intel processors, a little behind the curve - doubling every two years.

More Intel processors (same ones), but this time doubling every 18 months.

This one is probably the most useless. It looks good (although too large to display the whole thing) until you notice the disclaimer for the vertical (Transistors) axis: "Note: vertical scale of chart not proportional to actual Transistor count." WTF?

However, whether it was numerically correct from the start may not be the important part. How about just nailing down the shape of the curve? Isn't that worth something by itself?

On the other hand, does anyone actually have a graph of transistors per chip, or transistor size plotted against time, covering the past 40 years? That is, is anybody checking the numbers?

I guess I can do my own Googling:

Gordon's graph paper that shows cost versus number of transistors per chip

Intel processors, a little behind the curve - doubling every two years.

More Intel processors (same ones), but this time doubling every 18 months.

This one is probably the most useless. It looks good (although too large to display the whole thing) until you notice the disclaimer for the vertical (Transistors) axis: "Note: vertical scale of chart not proportional to actual Transistor count." WTF?

(pdf link)

Let's say that I did have a copy of this magazine. I would expect to be paid for it based on Moore's Law. Its only fitting. So with that in mind, let's see how much it woiuld be:

Magazine came out 40 years ago. Moore's Law says it doubles every 18 months. That's 26.6 doublings. Let's take 26 to make it easy. So thats 2^26 of the price.

I could not find what the cover price was but let's be fair and say $0.10 (10 cents). So thats 2^26 * 10 / 100 = $6,710,886.40. Thats a good deal more than the $10,000.00 they are offering.

I think its a rip-off.

BTW: here is a link to the original article in PDF format.

For those interested, here is the original paper. While I think it was a valid observation at the time, it's unfortunately become a driving marketing factor for the industry. While processor speed may be increasing all the time, I question the demand for it. Already, computer sales are leveling off as people realize they don't really need more than a 1Ghz to surf the web, send pictures, and listen to music. Even an Intel research paper suggested Moore's Law was coming to an end based on simple technical limitations.

I guess I'm just finding it difficult to imagine what I would ever need, say, 32Ghz for, other than gaming--which would be what my ultra-hip game console would be for.

Eventually, they will be able to run much much faster. But at the time the current chips aren't made to use this kind of transistor. They'll definitly bump up speed in the long run but I'm wondering if it still won't follow Moore's law.

They make a very large majority of their money from flash memory, as i recall.
Most of their money comes from the x86 processor type products (CPU, chipsets, etc). Flash memory falls under Intel Communications Group that actually is losing money

Problem is, Intel manufacturing is so expensive, they can't afford to give Dell any more of a discount
Intel can manufacture cheaper than AMD. If you look at the 2004 financial numbers, Intel has much better gross margins than AMD.
Intel Revenue: $34.2B
Intel Cost of Sales: $14.5B
Intel Gross Margin: 58%
AMD Revenue: $5B
AMD Cost of Sales: $3B
AMD Gross Margin: 40%

I believe they've been hiding under this rock.

(...every time I hear that Intel "chime" on TV I almost lose my mind...)

If you realize that a cache transistor uses about 10% of the power consumed (and hence, heat generated) by a logic transistor, you see why the power/thermals issue IS EXACTLY why Intel is shifting a high percentage of die to cache since it generates excellent performance-per-watt compared to logic. Itanium, by the way lives in a much more disciplined thermal-design-point than most modern 'big' chips (eg. 160 watt Power 5) being 130 / 99 / 62 watt (depending on model) - which is actually better than most Xeons, Pentium 4s and other Prescott-core designs
BTW, Another poster (later on) also correctly mentions that Moore's Law simply talks about number of transistors on a chip - not which type and what they're used for - stigmatize cache when you get your own law :-).

1. Macs are made with higher quality, so their users tend to hold on to them longer than the equivalent PC and they have a higher resale value. Which is why you tend to see Macs (Toyota Avalons) be more expensive than their PC (Ford Focus) equivalent.

2. I can see you are using the archaic clockspeed (GHz or MHz) metric for measuring CPU speed and performance. I would say that a PowerMac G4 w/ 256 RAM and a 500MHz PowerPC processor running Mac OSX 10.3 Panther is faster than an IBM Aptiva w/256 RAM and 500MHz Pentium III running Windows XP SP2, in that same token BeOS would run circles around both Windows and Mac OSes on the same hardware but I digress. My statement is based on the fact we have both in our department and the older Macs with Panther feel faster against older PC of the same clock speed running Windows XP SP2. Now you might think this is a smoke screen and an apologists version of CPU speeds. So let me point to you to the company that made plenty of money on marketing the clockspeed metric years ago. About Intel Processor Numbers

When comparing processor numbers, it's important to keep in mind that there are other key features besides clock speed that contribute to the processor's overall value. For example, there may be a case where the processor number increases because a front-side bus speed increases (e.g. from 400 MHz to 533 MHz), or cache increases (e.g. from 512KB to 1MB), while the clock speed stays constant or even decreases.

A muscle car Pontiac GrandAM may have plenty of horsepower under the hood but it doesn't translate to speed on the road. Compare that to a BMW 3 series which costs more can have the same horsepower but has better performance. You're not comparing two equal products. Take a top of the line Toshiba laptop or IBM Thinkpad and then compare that to a Apple Powerbook. Your comparisons are not comparing the same types of products in my opinion.

HyperThreading (guessing that is like the velocity engine, although I don't know what either term means, probably slick marketing)

Actually they are quite different technologies. SIMD is Motorola's Velocity Engine/AltiVec/VMX is equivalent to Intel's SSE2. Intel's HyperThreading is equivalent to IBM's simultaneous multithreading (SMT) on their Power 5 processors. CPUs like Intel's which have very deep pipelines sometimes benefit from HyperThreading and other times not. The IBM PowerPC (not Power) line of processors don't have as deep pipelines which is why they use a more superscalar architecture like the PowerPC 970 which has four ALUs, two FPUs and two SIMD units. I hope that helps and sheds some light on the confusing technical lexicon.

Here what I would recommend if you can save up for it. An older PowerMac G4 running at 500MHz with 256MB of RAM and 32MB graphics card OR a newer MacMini with 512MB of RAM. Don't bother with any G3 based Mac on eBay, as the SIMD (velocity engine) makes a difference in the performance of the machine and so does any video memory below 32MB. There are other conveniences of the Apple Mac OSX platform which become more apparent with the imminent release of Mac OS X 10.4 Tiger. Good luck which ever direction you go. May your firewall protect you from worms, your web browser from spyware and your operating system not burdened by viruses.

The paper he wrote for Electronics Magazine is here.

Its definately the verification that gives hardware the edge. There are usually 3-4 times as many people verifying just logical correctness alone then there are designing the logic. The situation is the same for the physical design as well. Intel has a great paper on the P4 validation process here.

Sure took a long time. The concept of perpendicular recording is relatively old - I remember work being done on it back in days when Microsoft was a cute little monkey. Here is a nice link that explains the process: http://www.wtec.org/loyola/hdmem/02_03.htm/ Intel has a note on it - no date, but this is for floppies... http://support.intel.com/design/archives/periphrl/ docs/7281.htm Toshiba is working on this technology: http://www.nwfusion.com/news/2004/1214toshitous.ht ml/ The technology was used in tapes in 2002: http://www.internetnews.com/storage/article.php/15 01631

Moore's law is about the doubling of the number of transistors not the doubling of speed.

http://www.intel.com/research/silicon/mooreslaw.ht m
ftp://download.intel.com/research/silicon/moorespa per.pdf

Moore's law is about the doubling of the number of transistors not the doubling of speed.

http://www.intel.com/research/silicon/mooreslaw.ht m
ftp://download.intel.com/research/silicon/moorespa per.pdf

For the ignorant ones, the making of the PCB's themselves is not a simple process. Think about the traces you see on the surface, then place about 4-6 layers on top of each other. The fact that the PCB's are outsourced takes a huge load off the remaining process.

The first Pentium III motherboard I owned (an Intel SE440BX-2) looked like the product of an alien civilization. I still have the computer I built around it; I'm using it now, in fact. Slackware 10 (kernel 2.6.9), 550 MHz Slot 1 Pentium III, 768 MB RAM, 110 GB disk (80 + 30), ADSL. It serves me well.

...laura

Windows Server 2003, released in March of 2003, supported Intel Itanium processors which are in fact 64-bit chips.

Other groups working on optical interconnects: (incomplete list)

Heriot Watt

Cornell University

IBM Zurich

Delft

UIUC

Intel

Stanford

Take a look at this on IBM compilers on mac os x. According to SPEC ratings int performance is 11% to 50% faster using xlc and floating point is apparantly even better. Most of the performance gains are over 50%. Apple of all people can afford a compiler to at least compile their own OS on. The free software side of me in the other hand is happy that they are choosing to improve the gnu compiler instead but it honestly doesn't make any sense to me since they can get a practicaly free huge performance gain on a relatively cheap purchase of a compiler.

-bloo

These new chips are not true 64-bit chips. The EMT64 chips just add support for more memory, and will "support" 64-bit operating systems (they're just doing some fancy emulation).

http://developer.intel.com/technology/64bitextensi ons/faq.htm
http://developer.intel.com/technology/64bitextensi ons/index.htm

We shouldn't see a TRUE 64-bit chip from Intel for a year.

These new chips are not true 64-bit chips. The EMT64 chips just add support for more memory, and will "support" 64-bit operating systems (they're just doing some fancy emulation).

http://developer.intel.com/technology/64bitextensi ons/faq.htm
http://developer.intel.com/technology/64bitextensi ons/index.htm

We shouldn't see a TRUE 64-bit chip from Intel for a year.

Parity VOS's Graphical Call Flow Charter (which was bought by intel and ignored a while ago) is an example of this. Anything other than a very simple call flow ends up being many many square meters of virtual screen real estate.

I will be very very happy once we finally get away from it, and I'm looking forward to a simple perl scripting based replacement.

And the first Google hit for the same question (after the paid ad):

Intel Announces Fastest Pentium® II Xeon(TM) Processor

Wow, Google found a CPU which is 50 MHz faster. A clear win!

You sure about that?

http://www.intel.com/software/products/opensource/

http://www.intel.com/cd/ids/developer/asmo-na/eng/ 52779.htm

http://www.intel.com/software/products/opensource/

http://www.intel.com/cd/ids/developer/asmo-na/eng/ 52779.htm

The minute they cancelled Itanium's whole branch

Sorry, but the Itanium2 is alive and well, though relegated to a relatively small market.

You might want to use the right part numbers when you go an speculate like this. So far, all of the parties have been wrong. It's not negligible, nor is it quite as large as others have stated. Looking up Intel's stated figures from ftp://download.intel.com/design/mobile/datashts/30 526401.pdf, on page 294, we find that the mainline mobile chip, the 915GM has a stated max TDP of 6W. So, if we assume for a moment that they're 100% accurate on this (I know, I know, just bear with me here...) then the aggregate TDP for the CPU and Northbridge is 33W for the Pentium-M model they tested- IF AMD used a laptop with this model setup. If they used a laptop with the "high-performance" chipset, it's liable to be higher. Now, that means that it's largely a push on peak consumption- it boils down to how well they idle back and how often when they're not idled back they're burning the peak power.

Keep in mind that all the other components, such as floppies, DVD drives, etc. consume power and they seriously contribute to the overall run-time of a laptop.

Was it cooking the reviews? Probably not, in reflection. Personally, I wish that the Reg author that made the article of the discussion subject would get some balls and find some integrity because he's pretty much wrong on most of what he wrote about.

Your post seems to indicate you are not looking at the entire Turion line. Indeed, there are chips with only 512K of L2, but there also others with 1MB. Also, the Turion has double the L1 cache compared to the Pentium-M, 64K.

Maybe the reason your last post was modded down was your lack of supporting evidence and merely speculative.

AMD data: http://www.amd.com/us-en/Processors/ProductInforma tion/0,,30_118_12651_12658,00.html

Intel data: http://www.intel.com/products/notebook/processors/ pentiumm/index.htm
http://download.intel.com/design/mobile/datashts/3 0526201.pdf L1 Cache size is stated on page 7, second bullet

Your post seems to indicate you are not looking at the entire Turion line. Indeed, there are chips with only 512K of L2, but there also others with 1MB. Also, the Turion has double the L1 cache compared to the Pentium-M, 64K.

Maybe the reason your last post was modded down was your lack of supporting evidence and merely speculative.

AMD data: http://www.amd.com/us-en/Processors/ProductInforma tion/0,,30_118_12651_12658,00.html

Intel data: http://www.intel.com/products/notebook/processors/ pentiumm/index.htm
http://download.intel.com/design/mobile/datashts/3 0526201.pdf L1 Cache size is stated on page 7, second bullet

For things like the Cell processor, you'd be better off with something like OpenMP that is designed for spreading threads across multiple execution units such as those in the Cell processor. All autovectorization does is help with things more like MMX.

Since the Pentium 4 according to Intel, but it's not a good question as that's Intel's trademarked term for their two-thread implementation of simultaneous multithreading:

Simultaneous multithreading allows multiple threads to execute different instructions in the same clock cycle, using the execution units that the first thread left spare.

By contrast, Niagara is implementing Chip-level multiprocessing:

CMP is SMP implemented on a single VLSI integrated circuit. Multiple processor cores (multicore) typically share a common second- or third-level cache and interconnect.

In other words, Niagara implements in hardware, at greater scale, what Pentium 4 offers as an emulation feature. In theory one could SMP on top of CMP chipsets for even greater throughput. If you find the Sun article too hard, the Wikipedia references I have cited will probably prove much easier to understand.

1. If it didn't work for Sprint, what'd they do wrong? Why will the next company to try the same thing work?

Good points.

First, distance: A single WiMax router won't be practical since a 30 mile broadcast range can easily cover too many users. Instead, overlap them and route from there. Your distance from 1 router will be substantially below 30 miles.

Second: Latency. A quick search shows that Intel is interested in 802.16a. Intel says in a foot note on the linked page: "6 Latency may be unacceptable for real-time IP services such as VoIP during handovers but acceptable for TCP and VPN services as well as store-and-forward multimedia services." They do specifically mention issues with VOIP and games and that these are addressed. Other sources say similar things about latency.

You're correct that the iMac wasn't the first to ship with USB but I don't think you can argue against that it was the iMac that caused USB to become so popular today.

Apple has less than 2% of the PC market. In the broad scheme of things, they are a very small-time player in the industry. They definitely did not cause USB to become so popular today.

Intel is the reason. Intel, with it's 90%+ marketshare in the CPU market, created the USB standard and pushed for it. Since Intel had such dominance in the CPU and chipset market, any motherboards that were designed for their CPU's and/or used their chipsets supported USB. If you were a motherboard manufacturer and designed a new MB at that time, you most likely used an Intel chipset and supported USB. Almost all the new motherboards being made for the PC industry had USB capability, thanks to Intel being the de facto standard.

I was working in the motherboard industry when Intel first started the push for USB. They made it easy for any motherboard manufacturer to include USB on their motherboards. They created the standard, and created the USB Implementers Forum in 1995 to push the industry to adopt its standard.

http://www.intel.com/standards/case/case_usb.htm

http://www.intel.com/intel/finance/pricelist/

What I want to know is... Lets say I swap the cards and it doesn't work, can I revert to the old card without any data loss?

I'm currently using an Intel SRCS14L. Does anybody have any recommendations?

Remember when Intel announced processor identification? They were slaughtered! They ended up shipping P3's with this feature turned off.

You must not remember the different small cases that Intel has demonstrated in the past at IDFs. I distinctly remember one looking like a pyramid, one that was kind of like a wave look, and a few others. All of these were roughly the same footprint size of a Mac Mini but were taller. All of these were examples of more "artistic" looking cases that could be designed. This was in 1998 link Look for the "Aztec" and the "Twister".

Intel has been demonstrating concept form factors at IDF for a long time.

Google knows all.

Intell talks about Infiniband Architecture

Initially InfiniBand Technology will be used to connect servers with remote storage and networking devices, and other servers. It will also be used inside servers for inter-processor communication (IPC) in parallel clusters. Customers requiring dense server deployments, such as ISPs, will also benefit from the small form factors being proposed. Other benefits include greater performance, lower latency, easier and faster sharing of data, built in security and quality of service, improved usability (the new form factor will be far easier to add/remove/upgrade than today's shared-bus I/O cards).

Additionally, InfiniBand Architecture reduces total cost of ownership by focusing on data center reliability and scalability. The technology addresses reliability by creating multiple redundant paths between nodes (reducing hardware that needs to be purchased). It also moves from the load-and-store-based communications methods used by shared local bus I/O to a more reliable message passing approach.

Scalability needs are addressed in two ways. First, the I/O fabric itself is designed to scale without encountering the latencies that some shared bus I/O architectures experience as workload increases. Second, the physical modularity of InfiniBand Technology will avoid the need for customers to buy excess capacity up-front in anticipation of future growth. Instead, they will be able to buy what they need at the outset and 'pay as they grow' to add capacity without impacting operations or installed systems.