Slashdot Mirror
Intel Begins Shipping 64-bit Prescotts
Rucas writes "With a minimum of fanfare, Intel has begun shipping a version of the Pentium 4 with 64-bit instruction set extensions. The news came to light not via an Intel press release, but rather through the spec sheet for a new server from IBM. In the midst of the new IBM eServer xSeries servers based on the recently released 64-bit Xeon is a blade server powered by the 64-bit Prescott. This marks the first product appearance of the new CPU."
A link to the original on Ars might've been nice:
t ml
http://arstechnica.com/news/posts/20040804-4070.h
Though Xeon-Nocona has been available for more than a month it seems there there are no substantial reports on 64-bit performance of Nocona. Is there anyone here who can report anything about the 64-bit performance of Nocona?
Where could I have seen it before.
--- There isn't any problem that can't be solved by a small, low yield nuclear device, is there??
At what point are people actually going to start making 64 bit applications? I'm not talking 64 bit linux or anything like that, I'm talking consumer level apps and games.
Among other things, it should let the OS map more than a few gigabytes of memory into the address space at one time. A 32-bit application will only be able to see 4 gigabytes (or 2, or whatever the limit ends up being after tag bits and OS space are reserved), but the total amount in use can be more, without an application rewrite needed. This is already done to some extent (my understanding is that the 32-bit processors have 36 bits of address space in total, with a 32-bit per process maximum), but moving to 64 bits gives a lot more headroom.
I see a lot of people upgrading to 64 bit chips, but what good does it do if there's nothing to utilize them? Is it just for bragging rights or what?
I'm a programmer and I have yet to see a need to get a 64 bit chip.
It's handy to have native handling of things like 64-bit integers, but addressable memory space is the most pressing reason right now. You'll be able to mmap() a file larger than 2 gigabytes on x86-64 machines (where up to now you had to use a non-x86 platform). You'll be able to hold more than 2 gigabytes of working data in RAM, which is significant if you're doing video editing (or heavy rendering or really heavy image processing).
Consumer apps and games will move into this niche in a few years (there are algorithms that let you trade off memory footprint and speed, and memory is cheap). But there are several places where the ability to address more memory is important _now_, even for user workstations.
It has been revealed that these 64 bit intel chips are not able to address as much memory as AMD 64 bit chips. Specificaly, whereas the Opteron/Althon64 has a 40 bit physical and 48bit virtual address space (not the same as virtual mem, remember that AMD chips each have a memory controler, thus upto 256 Banks of memory, via 256 processors), these intel chips are limited to 36 bits.
Thats right, the same 36 bits that intel has supported via PXE for years...
Thus, total system memory size for these processors is limited to 64GB, meanwhile the per-processor limit for AMD chips is 1TB, 256TB total in a system (max 256 CPUs, if anyone ever makes a board and Hypertransport bridges capable of supporting such a large number of chips).
Anyway, it is a big difference, and it hints that the actual implementation may be the same old slow PXE implementation intel has had for years (since the pentium pro, if I remember correctly).
------------ This post was made while on percocet and no spell checking has been done. deal.
man is machine
There are also the "intangibles". For example right now software can only use about 3 gigs of memory without hacks (PAE and such). This is because there is only 4 gigs of address space and the OS and libraries must be in there somewhere, so most OSes give the OS 1 or 2 gigs of that address space. And you must map a library into each program's view of the address space, possibly into different areas. With a 64-bit address space, you could give a full 4 gigs to tons of programs, all while having lots of libraries loaded and have a simple linear addressing space for everyone. This simplifies things quite a bit. And when you need to use more than 4 gigs of data, you'll be able to without any performance hit.
The biggest difference you'll see are the registers. While it won't help you type faster into a word processor, it could very well help a game out.
Comment forecast: Bits of genius surrounded by a sea of mediocrity.
Also, Anandtech just posted a new roadmap with some info on upcoming 64-bit Pentium 4 CPU/chipsets for the desktop. The Intel 925XE chipset (with 1066MHz FSB) will ship in October along with 64-bit Pentium 4 "F" processors. "F" supposedly means it's a 64-bit Prescott.
TO START
PRESS ANY KEY
Where's the 'ANY' key? I see Esk, Kitarl, and Pig-Up...
The AMD chips themselves are perfectly stable, and platforms are very mature. The bug you mention is ancient history and due to a bug in the way the Live! worked. The last great chipset Intel made was the BX.
Ok, the VIA chipsets from the late 90s and early 2000s were crap, we know.
IT'S NOT A PROBLEM ANYMORE, NVIDIA AND SIS BOTH HAVE QUALITY CHIPSETS FOR AMD AND VIA HAS CLEANED UP IT'S ACT.
I'm sorry for shouting but damn, this hasn't been a valid argument for years and intel zealots are still spouting off about it.
But rumor has it that Intel's 64-bit performance is no faster than its 32-bit performance, and slower than AMD's 64-bit performance.
As EE Times Reports:
Prescott is also Intel's first processor to support a security technology code-named Le Grande. While Intel has not yet detailed the technology, it is believed to provide a protected space in main memory for a secure execution mode required as part of Palladium, a new PC security scheme being developed by Microsoft Corp.
Le Grande is Intel's codename for Trusted Computing. HP's codename is ProtectTools, Cisco's codename appears to be either NetworkAdmissionControl or SelfDefendingNetwork, Phoenix BIOS code name is CoreManagedEnvironment, and of course we all know Microsoft's codename was Palladium and now is NaGSCaB and is slated to appear in Longhorn.
If you scroll down near the bottom of this page you can catch a look at a micrograph of the Prescott from about a year ago. Note that the Trusted Computing core is it's own an entire CPU and memory and support structures, and eats up about 20% of the chip. In other words Trusted Computing core ties up around 25 million transistors of real-estate, or about half of a Pentium 4.
It will support encrypted code (to secure it against you, the owner), it will encrypt RAM access (again, secure against you) and take over a portion of your cache. It will carry a unique key to identify you and your machine, but far more powerful than the old CPU serial numbers. It will forbid you to know your own encryption keys and prohibit you from decrypting your own data. I know it's designed to work with a "secure clock" (wouldn't want you the owner to be able to "tamper" with the time, now would they?), but I'm not sure if the secure clock is inside the CPU or planned to be external.
AMD has their own Trusted Computing project, but I have been having trouble digging out any hard info. It *may* be incorporated into the Opteron processor.
Transmeta has a trusted Computing project too, the TSX system - Transmeta Security eXtensions. I beleive initially appearing in the Caruso5800.
Welcome to tomorrow. Resistance is futile, all your base already belong to us, Slavery is Freedom, and always remember The Computer Is Your Friend.
-
- - You can't take something off the Internet! That's like trying to take pee out of a swimming pool.
Please do not post links to thefreedictionary.com - they are a dodgy site which repackages Wikipedia content, with ads, for profit, whilst stretching the GFDL as far as they possibly can.
Look at that link you posted - you'll see a credit to Wikipedia at the bottom. Now disable javascript in your browser and refresh - ooh, the credit is gone! They insert it in with javascript rather than putting it in the body of the page to ensure that Google doesn't pick it up. Why? Because a link to Wikipedia's article would help lift Wikipedia's pagerank above that of thefreedictionary.com.
Just say no, and if you want to read about PAE, read the original Wikipedia article.