Slashdot Mirror

Seems worth checking out: GPGPU.ORG - "General-Purpose Computation Using Graphics Hardware"

> AGP does a lot better taking data in, but it's still pretty
> costly sending data back to the CPU.
I've heard that mentioned a few times, is it true?

From the AGP 3.0 spec:
The AGP3.0 interface is designed to support several platform generations based upon 0.25m (and
smaller) component silicon technology, spanning several technology generations. As with AGP2.0, the
physical interface is designed to operate at a common clock frequency of 66 MHz. Its source
synchronous data strobe operation, however, is octal-clocked and transfers eight double words
(Dwords) of data within the span of time consumed by a single common clock cycle. The AGP3.0 data
bus provides a peak theoretical bandwidth of 2.1 GB/s (32 bits per transfer at 533 MT/s). Both the
common clock and source synchronous data strobe operation and protocols are similar to those
employed by AGP2.0.11

Later on Page 96:
Traditional AGP devices can demand up to the maximum bandwidth available over the AGP ports.
However, the AGP system does not guarantee to deliver the requested bandwidth, nor does it guarantee
transfers will take place within some clearly specified request/transfer latency time. ...
This is done by the system guaranteeing to process a specified number (N) of read or write transactions of a specified size (Y) during each isochronous time period (T). An AGP3.0 device can divide this bandwidth between read and write traffic as appropriate. Further, the system transfers isochronous data over the AGP3.0 Port within a specified latency (L).
(emphasis mine)

I'm no expert, just asking if the "low upsream bandwidth" assumption is true. If it is, there could still some applications (eg: simple data compression) that could use it. Also, maybe output from VGA/DVI ports could be tapped.

Half of what you want is cpu documentation, Intel makes nice free guides to their 32 bit x86 cpus. Sometimes they even will send out the manuals free in book format, right now it looks like you can download them or order a free cd. Check back often because the free books(shipping free too) come and go. http://developer.intel.com/design/pentium4/manuals /index2.htm

The other main half is BIOS info, check our Ralf Brown's int list: http://www-2.cs.cmu.edu/afs/cs.cmu.edu/user/ralf/p ub/WWW/files.html

For the rest you just need lots docs about various hardware like floppy controllers and such, check out my site for OS developers since I probably have more links listed then anyone else. Besides this stuff you might want to re-include java in your google searches because most of the research into virtual machines that I've seen uses java but alot would still apply.

They'll market the Celeron M line.

1. Make a 64-bit challenger to Athlon64. If it means butchering the Itanium die and adding a 32-bit co-processor, so be it.

2. Enable SMP on something faster than Tualatin.

3. Wake up to the fact that Intel can no longer dominate the CPU market on name recognition and MHz rating alone.

Well...they may say they are giving up the numbering but their marketing department didn't give it up.

Check out the table on the Intel product page.

benedict

Don't know if google spiders /. pages, though.

Yeah, datasheets for things like Ethernet controllers are real hard to get from Intel. Those poor BSD coders, they've only got the complete databook without any NDA conditions attached to work from!

Yeah, datasheets for things like Ethernet controllers are real hard to get from Intel. Those poor BSD coders, they've only got the complete databook without any NDA conditions attached to work from!

"Moore's law" does not refer to the speed of processors doubling, but that the number of transistors that can be fit into a given area doubling every 18 months or so.

The shrinking of transistor sizes has lead to smaller, cooler, faster, more powerful chips, but the speed increase is just a side effect of the smaller transistors. Were chip engineers more interested in packing more operations each cycle onto a chip, then you would see slower clock speeds with similar densities of transistors on larger areas (with more heat buildup being the speed limiter) -- something akin to the PowerPC chips vs the high-speed Pentiums. Similar densites of transistors and the PPCs actually do more floating-point operations per second (flops) than a Pentium that runs at about twice the clock.

My wife used to work for Intel and they sent her and a bunch of cow-orkers to a class on trademarks. She told me about how you're not supposed to refer to "a Pentium", but it's called "a Pentium processor" or some such because the trademark was not to be used as a noun. Since then I try hard to use my best, trademark-correct language for humor's sake, as in "Could you please hand me a Kleenex(TM)-brand facial tissue?" or "Curse these blasted Band-Aid® brand adhesive bandages that never stick for more than an hour!" It's interesting (and difficult) to try and figure out what these kinds of products are actually called other than Kleenex, Aspirin, Band-Aid, etc.

And more from Intel:
http://www.intel.com/research/silicon/leadfree.htm

And more information from AMD:
http://www.amd.com/us-en/Processors/TechnicalResou rces/0,,30_182_4040,00.html

Disclosure: I don't work for, or own stock in AMD or Intel. I haven't purchased an Intel chip since the Pentium came out.

-ted

And more from Intel:
http://www.intel.com/research/silicon/leadfree.htm

And more information from AMD:
http://www.amd.com/us-en/Processors/TechnicalResou rces/0,,30_182_4040,00.html

Disclosure: I don't work for, or own stock in AMD or Intel. I haven't purchased an Intel chip since the Pentium came out.

-ted

Wow, tautological perfection. I've never seen such.

Yes, we can always show some (incomplete) "proof" that we can't do X. And then we usually end up doing X in a novel and unexpected way.

Lather. Rinse. Repeat.

And, BTW, FYI, FWIW, Moore's "law" is more of an empirical observation than any sort of real law, much less one that would apply in this case of magnetics without a transistor in (relevant) sight. I don't mean to detract from the clever, albeit obvious in hindsight, prediction of Moore. He simply observed (and presciently predicted) that there is (and will continue to be) a sustained exponential growth in the number of transistors per integrated circuit (that's "switches" per "chip" to you and me).

That has absolutely not one goddamn thing to do with this topic or the cited article, so STFU or RTFM first. Please.

A norwegian company (I think) has joined forces with Intel to provide polymer storage within the decade. Exiting stuff: Opticom

Well, these guys, these guys, and these guys seem to have no trouble selling 386-based hardware. Not everybody needs the full feature set (or the cost and power requirements) of a Pentium.

It may have been a still birth, but Intel is trying it again. Look at EFI sometime. Byte code drivers stored in firmware. I believe it's only for storage devices, but it's still happening. The boot process needs to be modernized, but I have my doubts that EFI is the correct modernization.

Not one could get the sound working for more than brief periods.

You know what, I'd definately have to bite as "troll" on this one. Not only does the article not state the model of the soundcard (Intel card on an Intel board is the best you get), but it doesn't adequately describe the problem.

I'd say that the author either trolled around for a sound device he *knew* wouldn't work well. And excuse me, but doesn't it seem odd that this supposed new PC and new motherboard worked with the driver using "built-in" OS support all the way back to windows 95?

Now, let's go here. Every board I've checked either has a "soundmax" (I have one which doesn't autodetect on XP much less 95, not sure about 'nix) or an AC'97 (works just fine with 'nix on any mobo I've used that has it, NOT in windows 95 in many cases). I'm sorry, but this is just BS.

So, even if he got the Intel wrong what chance in hell is it that this MainStream board would be using a card based on an (technologically speaking) ancient model? Hell, I'm sorry but many times 95 won't even work well with my motherboard let alone the onboard sound

Intermittent problem: Maybe Fred forgot to set the mixer volume to pop up above 0 on bootup? The sound won't just "work sometimes," either it works, or it doesn't, or you're doing something weird.

What I would do is give Mr. Langa a well-built linux distro - installed - and windows 95. Then allow him the choice between the two after a month.

This isn't a legitimate criticism of Linux by far... it's a criticism made by a moron who either handpicked criteria or made it up.

Intel's C++ compiler is also free for non-commercial use. It even runs under Linux.

If you're concerned about code speed, the Intel C++ compiler for Linux has been available for free for a while, and AFAIK producess better code than the Microsoft one.

The Windows version of Intel's compiler is still commercial only, but maybe they'll change that now.

http://developer.intel.com/software/products/globa l/eval.htm

This reminds me of some similar hardware predictions (which I sincerely hope won't be analogous to this case): 1. Researchers have continually predicted that we will reach the fundamental limit of Moore's law in about ten years. (i.e. in 1970, it was ~1980, now, it's ~2014). You can look at a chart of the predicted end of Moore's Law as a function of time, and it's almost a straight line! The thing is that we've continuously found ways around it, seemingly in the nick of time.
I suppose the scary analogy here is that Microsoft will continually innovate just a little bit faster than the open-source community. While I'd like to say that the open-source community can out-innovate MS, let's not forget that MS has a LOT of money to throw at it, and they'll do whatever it takes if they feel threatened. 2. Gallium arsenide is the silicon of tomorrow, and it always will be. The carrier mobilities in GaAs are MUCH higher than silicon (i.e. the electrons move quicker), and we can incorporate several similar (III-V) semiconductors on a single chip to make ridiculously high frequency HEMTs (can you say 400GHz?), but it never caught on for VLSI applications. Primarily, this is because silicon has a nice native oxide (SiO2) layer that allows us to make high quality MOSFETs, which are better for VLSI integration for various reasons. GaAs can make some REALLY fast transistors and lasers, and gets plenty of use in cell phones and optics, but it's just too hard to integrate into VLSI, and all-optical processors are a long way off, so we're still using silicon in our processors.
The analogy here is that Linux may be well suited to the server environment, where it's more configurable and secure, but may never catch on on the desktop because of its the very same complexity that makes it so useful for servers.

VIA's random number generator greatly increases the speed to generate the keys (ie faster than p4 2.6 ghz on VIA's 1Ghz proc).

Don't fall for the Via FUD. They're making big "wooo, look at us, aren't we innovative" noises, but conveniently overlooking the fact that Intel chipsets have had a hardware random number generator since about '99/2000 with the old 810 chipset..

Details here and here.

Their claimed performance numbers, I suspect, are based on software generation on the P4 system. Regardless, having a hardware RNG is not novel.

VIA's random number generator greatly increases the speed to generate the keys (ie faster than p4 2.6 ghz on VIA's 1Ghz proc).

Don't fall for the Via FUD. They're making big "wooo, look at us, aren't we innovative" noises, but conveniently overlooking the fact that Intel chipsets have had a hardware random number generator since about '99/2000 with the old 810 chipset..

Details here and here.

Their claimed performance numbers, I suspect, are based on software generation on the P4 system. Regardless, having a hardware RNG is not novel.

Are the Most Unwired Airports and Most Unwired Cities lists.
Also, do these lists just count wireless access points that Centrino supports? It almost sounds like some sort of propaganda...

Are the Most Unwired Airports and Most Unwired Cities lists.
Also, do these lists just count wireless access points that Centrino supports? It almost sounds like some sort of propaganda...

The DRM + security features are those of the Group formerly known as the TCPA. TCPA has frequently been discussed on Slashdot.

From http://www.intel.com/design/pca/prodbref/253820.ht m>:

"The Intel PXA27x processor family incorporates the Intel® Wireless Trusted Platform that is designed to provide platform trust and robust security services required for today's wireless devices. Built around the concepts developed by the Trusted Computing Group* (TCG) industry forum..."

Although there were rumors about an Intel Yamhill 64-bit x86 part for many years, they didn't announce an 64-bit x86 architecture extension until February 18, 2004, and it was announced sheepishly as a very minor point in a press release rather than amid great fanfare as AMD had done. Intel still has not released any product incorporating this extension. Thus they've had more than 3 1/2 years to develop their own 64-bit x86 based on the AMD specifications. No need whatsoever for reverse-engineering. In fact, reverse engineering would have taken much longer, because they would have had to wait to get their hands on working AMD silicon.

Bill Siu, head of DPG (Desktop Products Group):
"Coming this year, more and more companies are employing hardware-based security. For example, the NX or the no execute feature will be available in our processors later this year."

If not, this should keep you happy:

The Intel® 6300ESB ICH contains a set of registers that shadow the contents of the legacy IDE registers. The behavior of the Command and Control Block registers, PIO and DMA data transfers, resets, and interrupts are all emulated.

If not, this should keep you happy:

The Intel® 6300ESB ICH contains a set of registers that shadow the contents of the legacy IDE registers. The behavior of the Command and Control Block registers, PIO and DMA data transfers, resets, and interrupts are all emulated.

At work we're considering buying a new (low end) Dell server which uses S-ATA hard drives. It's supposed to house a SuSE Linux system in the future. However, I'm not sure how well Linux in general, and SuSE in particular works with Serial-ATA drives, especially when there's nothing but Serial-ATA available - ie. the installer would need to work with it, as well.

The best resource I found was this page, but it doesn't help me a lot, either. The server would be a Dell Poweredge 750 running the Intel 7210 chipset, which supports S-ATA.

The system which the new server should replace is currently running SuSE Pro 8.1, which I am fairly certain does not support S-ATA - but does SuSE 9.x?

1000base-T used all four wire pairs, yes, but it was extremely limited in length. (10m, IIRC). 1000Base-TX goes the full 100m, uses all four wire pairs, and is full duplex on all wires simeltaneously.

Source, please?

1000base-T does go 100 meters.

Okay, I checked and found that 1000base-TX does exist and is TIA-EIA-854 (1000base-T is IEEE 802.3ab), indeed uses all 4 pairs of wires, but each pair is half-duplex. In addition, CAT6 cable is required for the additional (analog) bandwidth.

And the devices on the market are still 1000base-T. See for example Intel and Cisco: both say 1000base-T and 802.3ab, both go 100m at gigabit speed, and 1000base-TX is nowhere in sight.

Considering that 1000base-T & 1000base-TX are incompatible, and devices would not sell if they don't work with Intel and Cisco, I'd say 99% or more of current gigabit stuff on the market are 1000base-T, not TX.

Being integrated with the motherboard doesn't make a performance difference on any board I've ever seen. It still goes over the PCI bus, it's just not using a slot. Creating a separate bus just for the ethernet port would be too expensive.

Not really if the motherboard chipset was designed from the get go to support somthing like that, it was mentioned earlier in the posts about the Intel i875 as well as the nForce3 based systems.

I've been using GbE for home LAN for about a year now. Here's the hardware I use:

Switch:
Linksys Instant Gigabit 10/100/1000 8-port switch
I think I paid ~$200 for this.

Cards:
Intel PRO/1000 MT Desktop Adapter (~$50 ea)
Use the e1000 driver in 2.4.x or 2.6.x.
Netgear GA302T Copper Gigabit Adapter (~50 ea)
Use the tg3 driver in 2.4.x or 2.6.x

The tg3 chipset runs rather hot, the e1000 is tiny and runs cool. I havent noticed a performance difference between either, and both chipsets run fine regardless of whatever PC I put them in.

Motherboards with embedded GbE typically use e1000 (if theyre good), or realtek (if theyre cheap).

Jumbo frames:
See my post on that here.

Cabling:
Hand crimped cat5e. Works fine. One interesting note about GbE, you no longer have to worry about crossover cables -- the GbE spec requires that devices autodetect crossover. You can make all your GbE cables "straight through" cables.

Do pay careful attention to following strict T568 wiring code though. You can no longer get away with incorrectly wired cables which just happened to work for 100bt. Since all pairs are now used in GbE, your wiring order must be 100% spec.

Here's some wiring guides:
http://www.lanshack.com/make-cat5E.asp
h ttp://yoda.uvi.edu/InfoTech/rj45.htm

Personally, I'm still astonished that nobody has mentioned 10 Giabit Ethernet, and said something like "Pfft! Who needs that pokey gigabit stuff anyway?"

Slashdot and people saying "you don't need something that fast" is strange and frightening. I can at least understand people pointing out that you are unlikely to get full gigabit speeds, but then those people go on to say "so you don't need gigabit" - completely missing the point.

Boy this turned into a bit of a tome.

For a switch I went with an 8 port SMC EZSwitch 8508T. I chose it since:
1. It supports jumbo frames. According to my testing it will pass ethernet packets up to 9212 bytes which should correspond to a 9198 byte MTU.
2. It doesn't have a cooling fan. A definate plus since in my experience the little fans in switches such as this can become quite annoying as they age.
3. It comes with rack mount ears.
4. It's affordable. I purchased it from Securemart.com for $139.31 shipped. Ordered it Thursday or Friday, it arrived Monday or Tuesday.

As to NICs, one of my PCs already had an Intel gigabit port on the motherboard. In addition I purchased 4 more Intel Pro 1000/MT Desktop Adapters. Since:
1. They have good driver support on both Linux and Windows.
2. They support jumbo frames. Supposedly up to around 16000 bytes.
3. They're supposed to be pretty fast/efficient. It's kind of dated but you can find a comparison of some 32-bit gigabit NICs here.
4. They'll do 66Mhz if your motherboard supports it and of my systems does.
5. They have DOS NDIS2 drivers so I can use Ghost to make/restore images over the network.

One I purchased through Intel's evaluation program for $35.31 shipped. As I recall it took over a week to show up. The other three I ordered from OnlineMicro for $28 each plus $11.32 shipping. Be sure to change the shipping option from ground to 2 day air if you order more than 1, it's cheaper. They shipped them out the day of my order and they arrived on time.

One of the Intel NICs died about 4 hours after I installed it. I swapped it with another and the replacement has been working fine for a few weeks now. I ran the diagnostics on it and other all but the link test passed. When the OS is booted up the switch shows no link lights but sometimes when the PC is off the link lights do come on. I've also tried it in another PC where it exhibits similar symptoms. I haven't yet contacted Intel about getting it replaced.

I spent a lot of time tweaking various things. Some findings:
1. With default SO_RCVBUF sizes a MTU in the neighborhood of 4000 or so bytes seems to get about the best network/application wide throughput. Specifically the otherwise fast NF7-S system below would lose almost 50% throughput with 9000 byte MTUs with the default SO_RCVBUF size. Linux to Linux lost around 30% as I recall.

In theory you can change the default SO_RCVBUF size on linux by echoing appropriate values to:
/proc/sys/net/ipv4/tcp_rmem
Other than that you appear to have to change this setting in each individual application. One application of note that allows you to easily make this change is samba. See your: /etc/samba/smb.conf

2. If you crank the SO_RCVBUF size up to 200ish k or more then a 9000ish byte MTU can eek out another 5ish percent more bandwidth. Thus for the moment I've decided to just stick with 4076.

3. MTUs that are not of a size of the form 8x+4 cause Linux to behave oddly when it performs path MTU discovery. Namely for jumbo sizes that don't fit that form the discovery decides that the PMTU is 1492. You can read more detail about it in a Usenet post I made here. I still don't have a good picture of what'

because it's the first core logic chipset to integrate a Gigabit Ethernet MAC

What about Intel's 875 chipset? It supposedly has Communications Streaming Architecture that bypasses the PCI bottleneck to allow gigabit ethernet to go faster.

Intel 875P Info

But I'm no Intel fanboy. I'm happy that Nvidia's improving their Athlon 64 offering by providing something with REAL Hypertransport support, none of that half-speed stuff.

Q1. Do Intel products contain lead?
A1. Yes, most of our products contain lead in very small amounts. The use of lead in very small quantities in electronic products is ubiquitous. Lead is found throughout electronic components, component packaging, printed circuit boards, and other products. Intel estimates that approximately 90% of all electronic components contain some lead -- mostly due to the use of solder that contains lead. For a desktop computer with a CRT monitor, Intel's products contribute <1% of the total lead in the computer system. A typical Intel microprocessor contains approximately 0.2 grams of lead. An Intel motherboard contains roughly 2-3 grams of lead. For comparison purposes, a house key contains about the same amount of lead (2-3 g) as an Intel motherboard.

I am confused: I thought StrongARM was an Intel processor

My favorite part is the wild misconception people have that moore's law has anything to do with speed. His real observation was basically "We're gonna have more space on a chip to cram stuff on there than we know what to do with."

Keep in mind this was around the time the concepts of RISC were being advanced, basically advocating a push for simplicity. Moore was saying "you guys are trying to find the most effiecient way to store a dozen small boxes in a 800 square foot room, which by the way will be 1600 sq ft in 18 months."

That being said, from an electrical engineering standpoint RISC techniques were needed for chips to scale to where they are today, but you'll notice they've all left the 'simplicity first' mentality behind with things such as out-of-order execution and branch-prediction units.

Basically RISC said "keep it small simple and efficient" and Moore said "who cares, cram it all in there, there's plenty of room" and the industry said "I'll take both."

Here is the link to his Paper

Uhh.. you mean like these Pentium 4 and Itanium 2 docs?

Intel has top-notch documentation, far and away the best in the industry. The only company that comes close is AMD. IBM's public documentation for their processors is absolutely abysmal in comparison. Maybe they ahve good documentation buried somewhere in the company, but they sure don't like sharing it with anyone.

As it stands now, PowerPC is no where near as "open" as x86 is, IBM has a LONG way to go.

Uhh.. you mean like these Pentium 4 and Itanium 2 docs?

Intel has top-notch documentation, far and away the best in the industry. The only company that comes close is AMD. IBM's public documentation for their processors is absolutely abysmal in comparison. Maybe they ahve good documentation buried somewhere in the company, but they sure don't like sharing it with anyone.

As it stands now, PowerPC is no where near as "open" as x86 is, IBM has a LONG way to go.

Since this just makes the $250 Squeezebox a really expensive remote for you 'puter, you might want to consider some other products instead.

If want you want is just a remote for your computer, then the Keyspan Digital Media Remote provides a $40 dollar alternative.

If you want to stream video to you TV from your Mac, then El Gato's eyeHome will pull this off at the same price as the Squeezebox. It will also stream music from iTunes and do slideshows from iPhoto, with no hacking required. ( I know.. what's the fun in that :) )

If you want to wait for Microsoft Announceware, you could get the Windows Media Center Extender for XBOX and use the XBOX to stream video to your TV.

Or at ~$150 you could get the Linksys Wireless Media Adapter . This is the first round of Intel's Digital Media Adapters , of which there is a good review of the technology at linuxdevices.com

I'm surprised why all eyes on /. are at MS when Intel is moving up with the EFI technology that could take out the need for BIOS (at least in future versions).

From their page:
"The EFI specification defines a new model for the interface between operating systems and platform firmware. The interface consists of data tables that contain platform-related information, plus boot and runtime service calls that are available to the operating system and its loader. Together, these provide a standard environment for booting an operating system and running pre-boot applications.

The EFI specification is primarily intended for the next generation of IA-32 and Itanium(R) Architecture-based computers, and is an outgrowth of the "Intel Boot Initiative" (IBI) program that began in 1998."

/me is betting that MS is involved in this initiative too.

see for example, this link at intel for the faq or this link where intel started defending it or this link, last updated 4/28/2000 talking about the fact that psn's (pentium serial numbers) will not be in the next generation of pentiums that last link has many bits about why the PSN was a headache.

Yes, this nifty hardware was built into Intel's 82802 Firmware Hub(EEPROM+RNG hardware). It was used in early 810 and 815 chipsets.

Most mobo manufacturers chose to use others plain-jane flash chips as they were cheaper and they could use them over their entire line of mobo's (VIA, SIS, etc). This and lack of support by software companies for the RNG resulted in intel discontinuing the "firmware hub".

Datasheet available here

And that should impress me how?

Its a pretty vanilla ARM core with a few tweeks

I wasn't trying to troll with the grandparent comment, but if you look at the CPUs being developed specifically for handhelds by Texas Instruments or hell even intel with their multimedia extensions. These chips open up a new world of apps for PDA/Handheld products with DSP tech built in to them.

This is just another "been there done that" core from motorola.

They say that Intel is still manufacturing 386 processors for those who want them, but using a much smaller scale (perhaps 0.2 micron or whatever) because that's the only fabs they have nowadays.

Intel calls them mature processors, and they are now sold to the embedded market. You can still buy a 486, a 386, or even a 186.

They say that Intel is still manufacturing 386 processors for those who want them, but using a much smaller scale (perhaps 0.2 micron or whatever) because that's the only fabs they have nowadays.

Intel calls them mature processors, and they are now sold to the embedded market. You can still buy a 486, a 386, or even a 186.

They say that Intel is still manufacturing 386 processors for those who want them, but using a much smaller scale (perhaps 0.2 micron or whatever) because that's the only fabs they have nowadays.

Intel calls them mature processors, and they are now sold to the embedded market. You can still buy a 486, a 386, or even a 186.