For those that teach with Java, I wonder how you teach assembler?
If you're used to C on Unix, you can just use the assembler
output switch on GCC to look at the generated code and get an
idea as to how the assembler relates to your source code.

This, of course, is impossible with Java (unless you want to read
the assembler for an interpreter).

this is nonsense. java compilers generate assembly code just the same as a c compiler generates it. there is a java virtual machine definition. there is nothing which requires java byte code to be interpreted. there have been companies which have implemented java byte code cpus.

java vm isa (instruction set architecture) is not very interesting though. it's a stack based byte code and i wouldn't want to code in it. it is certainly doable though.

if this is the enegera they're talking about, their latest pr is this: http://www.egenera.com/display_pr.php?id=218

you seem to be discounting the existing software base of sparc which itanic doesn't have. a vast majority of that base will not go to x64 solaris or linux for quite a while.

I believe that those who invest in a single core machine after Q1 2006 will have an obsolete boat anchor in short order.

i hope not all your conclusions are this silly. there is no program which a sc 32 bit machine can't run that a dc 32 bit machine can run. on the other hand if you have one 64 program which you need, you're sol with a 32 bit machine sc or dc. not snappy come backs are rational or sensible.

of course this discussion completely ignores the utterly pathetic fp performance of arm and the fact that it doesn't run x86 code.

most expensive gas was sold in Turkey

I just checked it here: http://arabam.milliyet.com.tr/haber.asp?PID=681&HID=1&haberID=275525

this article says that the price of a liter of gas is 2.73 YTL. Given a dollar buys you 1.343 YTL it is ~$2/lt. I don't how that compares with your prices.

why don't you let the article speak for itsel ?

...The alliance is a new step in a years-long effort to encourage broad Itanium use and is an indication of just how much effort its backers are willing to put into getting the processor to catch on. But some believe the effort is late.

...

There are plenty of companies selling and buying Itanium-related products, but not nearly the number Intel and HP projected when the chip was under development in the 1990s.

mobile athlon 64 4000+ is available.

now that's a processor i can upgrade my wuxga+ r3000 with. maybe when the price gets a little bit lower though. at newegg someone mentions that they've upgraded a zv5000z with no problems.

Anyone here upgrade the CPUs on their r3000z's? I'm thinking of going the upgrade route as I can't find good high-res laptops
out there and mine is already high-res.

what speed cpu do you have ? i have a 3400 already and i really don't think another 200mhz (%9) is going to help much. if there were core E 754 pin 90nm chips with SSE3 i would do it but things as are, i don't see the point; at least for me.

please keep us updated on this. i'd be interested as well.

thanks.

i have a 20 inch dell (2001 ?) and it "whines" when i get out of standby. it is annoying enough that i have to put on some music but not enough to try to return it. if 2405s are worse it would be unbearable, at least for me.

WUXGA 17 inch Athlon 64 notebook

any particular brand you're hoping to get ? i currently use a compaq r3000 with 15.4 wuxga and i am quite happy with it but i wouldn't mind upgrading to a 17 incher; with 2.4 ghz cpu and more memory of course. actually i have a quite nicely configured system 1.5g memory, 80 gb 7200 rpm hard disk and 2.2ghz 754 cpu. i am considering upgrading the memory to 2g soon. the only crappy part of this notebook is the video card but as the most video intensive stuff i do is viewing verilog sim waveforms, i really don't need anything better. but of course buying a new computer usually doesn't have anything to do with "need" but "want"

MSVC++ has a bug in that it doesn't encode x >> n ^ x << (32-n) as a rotate. Instead it generates some really horrible code.

but does that horrible code compute the same result as a rotate would ? if yes, it would only be an optimization problem. but i am assuming that by "bug" you are saying that it doesn't.

you mean rc1 ? depending on the quality of rc2, we may yet see rtm in june. the time between rc2 and rtm could be very short. rc stands for release candidate for a reason after all.

The key is that USB isn't exactly the most efficient when it comes to bandwidth utilization

that's why i mentioned "raw bandwidth" in my post. also your example doesn't make that much sense. first of, i don't know what you mean "connect a mouse or keyboard to the same port as the usb hard drive". how do you do that you mean put both devices behind a hub ? only one device can be connected to a port at one time. another thing is that a slow device like keyboard or mouse only consumes some interrupt bandwidth, the bulk allocation is still open for the hard disk. the worst problem with usb is not mixing slow and fast devices but the fact that almost all host controllers are connected like a star at the root so no matter how many ports you have, there is only a 480mb/s link at the root so connecting two hard-disks to two different ports share the same 60mb/s raw bandwidth.

usb 2.0 is 480 Mb/s so the raw bandwidth is 60MB/s which is actually below what a high end hard disk can supply. The 133MB/s ATA standard is of course a number no hard disk read channel can reach but you can transfer at that rate from the on-disk cache i suppose.

here is what i've observed for an r3000z: (of course fedex tracking is not smart enough to see through time zone changes)

Apr 29, 2004 9:07 am
Package status
EMERYVILLE CA
Pkg returned to shipper
5:38 am
Arrived at FedEx Destination Location
EMERYVILLE CA
5:19 am
Arrived at Sort Facility
OAKLAND CA
4:10 am
Left FedEx Sort Facility
INDIANAPOLIS IN
2:06 am
Release for delivery
INDIANAPOLIS IN
1:20 am
Arrived at Sort Facility
INDIANAPOLIS IN
Apr 28, 2004 9:38 pm
Left FedEx Ramp
SHANGHAI CN
4:08 pm
Left FedEx Sort Facility
ANCHORAGE AK
12:54 pm
Arrived at Sort Facility
ANCHORAGE AK
1:10 am
Package status
SHANGHAI CN
Package in FedEx location
1:09 am
Pickup status
SHANGHAI CN
Package received after FedEx cutoff time

3400+, 15.4 wuxga, 30gb hd (to replaced with 7200rpm 60gb). i think it's going to be a nice desktop replacement. my current desktop is a 500 MHz celeron upgraded from a 250MHz pentium pro ii. it is amazing that even that machine is good enough for web and email but i use a 2200+ atlon presario 900z notebook for serious work.

thanks for the heads up. i just changed my r3000z order from wsxga+ to wuxga.

also i've read that at least one person has tried the win64 and other than the pcmcia drivers it seems to work ok. i know a couple people have installed various linux distis. check out some of the posts here: http://www.amd64notebooks.com/phpBB2/viewforum.php?f=19

one UPGRADE last week brings the Mean recommendation DOWN by 0,1

i think you're missing something here. at the bottom of the table it says "(Strong Buy) 1.0 - 5.0 (Strong Sell)". with more and more upgrades the number should be closer to one. it is, as they say, inversely proportional to the upgrades.

at the risk of sounding like a broken Intel-a-bee record

that train has come and gone a long time ago; don't worry anymore.

you seem to be missing relatively large chunks of market for 64 bit processing and memory size. eda is one such example. people are moving to opteron linux machines from sparc boxen to run large sims. also video/audio editing is another big market both for the memory size and the processing efficiencies you get from 64 registers and alus. a single minidvd tape occupies 12gb. pretty soon even consumers will require this stuff.

alas for intel, when you have a physical part which is "obviously" separate from the other physical part, it is more difficult to play that game. and it is not that they haven't tried it either. there has been a lot of talk about intel's bundling problems in the past and chipset manufacturers have complained loudly (and to good effect) about difficult to get licenses etc. bundling of physical items are clearly illegal (ibm vs amdahl) so intel has to be a lot more intelligent (and is) about what they can do.

the difference between intel and microsoft is that intel can't give away their cpus as microsoft can give away internet explorer. each cpu has to be actually produced individually with an actuall cost associated with it. i am sure the fact that amd is still in business doesn't show a lack of effort on the part of intel. many companies have tried to make a business out of x86 market and almost all failed.

ability to go back a few statements and rerun them (with ability to edit the statements) - which is now missing from VS 2003, but Microsoft promised to bring it back in Whidbey
called edit & continue i think. i didn't realize they removed it in vs2003. still using vs6 and very nice to debug dlls which one can edit and recompile without affecting the host .exe.

but you can still open up a register window and change the value of eip to go back a couple of statements to take a branch you missed etc. having a mixed source window open (c+asm) also helps with this.

geforce 5200 and power

i bought a gf5200 because it supported dx9 and its fan died in a couple of months which caused me a lot of heartache because i thought my dell 20" lcd display was dead when it would turn itself off after a couple minutes. it turns out when gf5200 gets too hot, it turns off the dvi block. now i have a fan which is not controlled by the card and the noise is driving me crazy but at least i got my gorgeous display back.

If DELL caves in

aint gonna happen.

HDL Compiler (isn't this a real biggy?)

to be frank i am not sure what "hdl compiler" would mean outside of design compiler. i think it is just a mistake that they didn't include it on synopsys web site. actually the list i have has ia64 too and hdl compiler is included for it also.

synopsys eda tools which support opteron:

AMD Opteron and Athlon, Red Hat Enterprise Linux v3.0

Core products available now:

Astro
AstroRail
Design Compiler (DC)
DesignWare Library (DW)
DFT Compiler
Formality
HDL Compiler
Hercules
JupiterXT
Milkyway
Module Compiler
NanoSim
Physical Compiler (PhysOpt)
Power Compiler
PrimeTime
PrimePower
Star-RCXT
VCS
Vera

Support scheduled to occur end of April 2004:

Pathmill
TetraMAX

in email from mentor:


...
Calibre 2004.1_2 Release Highlights
AMD64 platform supported
...

I have never seen changing chip functional charateristics as the goal, or the result, of metal layer changes

i'm surprised to hear you say that. i am involved both in physical layer full custom designs where the io designed by individual polygon pushing and southbridge portion of chipsets where it's mostly standard cell asic flow. in both cases, if you plan ahead there are lots of opportunities to do metal only fixes to make significant changes. especially in logic i've seen ethernet mac, ide controller bugs fixed and functionality (albeit not major) added. especially it's relatively easy to make state machine changes, sampling phase corrections etc. of course this assumes you have a pretty good base and lots of spares to begin with.

afaik, I don't believe that you can make on die memory controller changes with metal layer changes only.

depends on what the change is. if it involves a completely new io family (ie sstl2 to rdram) then it is not possible but if it is logic, it is most likely and even if it is similar io with say different bias voltage differences or current drive strengths, it depends on how much you've anticipated. if you have a dll with programmable tap select, programmable bias generators and lots of spares, a lot of things are possible but you have to plan ahead a little.

it was meant for those who glibly state that all it takes is another level of metal or a simple spin, or ...

i guess this is meant for my post. do you understand what a metal only change is (as opposed to base change)? what makes you think that it means another level of metal or that it is a simple spin ?

all i can say is many of you folks have very little idea of how hard it is to do cutting edge integrated circuit design.

memory controllers that really work in a variety of layout conditions in multimillion unit volumes are very difficult to design.

you just posted two statements, first of which is probably true but you have no idea who is and who isn't; the second is just a statement of the obvious because of the complexity involved. more or less your post is content free. so instead of making snide comments by vague statements, why don't you say something meaningfull which adds to the discussion ?

especially given that, if designed right, memory controller changes can even be metal only changes with no base impact which would reduce the mask costs substantially.

i have it from good authority that acer "should start selling this unit, or one like it, in late March/ early April time frame"