• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

4 GHz ALU's and 1.2 GHz Alpha

VladTrishkin

Senior member
Ace Reports:




<< Samsung also fattens DRAM densities with a 4-Gbit double-data-rate SDRAM. The architecture eschews the 0.14-micron technology node used for 1-Gbit devices in favor of 0.10-micron CMOS to realize a 0.10-micron2 cell. Samsung researchers used gain-controlled pre-sensing and active calibration of bit-line reference voltages to improve sensing margins. The company said the design is the first time 4 billion transistors have been contained on a single IC.

Intel will present two papers detailing its recent innovations in both the IA-32 and IA-64 segments. Attacking low-power/high-performance dilemmas for handheld applications, Intel will describe an enhancement to the IA-32 architecture that offers up to 1,000-Mips performance at 800 MHz but runs at operating voltages of 0.7 to 1.65 V.

Intel engineers also will describe a 4-GHz integer execution unit deployed in 0.18-micron CMOS. The design, which runs the ALU at twice the frequency of the rest of the processor to reduce latency, is designed to boost desktop and server speeds while retaining instruction-set compatibility. Circuit techniques that Intel deployed in an implementation of the ARMv5TE allow power and performance to be widely adjusted for use in mobile devices. The device, to be described at the conference, is a 16.77-mm2 hard macro intended for use in system-on-chip (SoC) designs.

At the IA-64 level, Intel bumps up by 1.5 times the frequency of its current 64-bit processor in a 1.2-GHz design that exploits 3.3 Mbytes of on-die cache, organized among four separate arrays in a three-level hierarchy. The lowest levels of cache achieve a low enough access time to enable zero-penalty access for integer instructions, Intel will report. Engineers from Compaq Computer Corp. are keeping pace on the high end with a 1.2-GHz Alpha processor that claims a 44.8-Gbyte/s chip-pin bandwidth. The design incorporates a 1.75-Mbyte L2 on-chip cache.

Sun Microsystems Inc. will describe a version of its MAJC processor, including two complete four-way-issue VLIW processors running at 500 MHz. And IBM Corp. will describe a Power4 processor running at more than 1 GHz with two four-way out-of-order-issue processors.
>>



-Intel is gaining speed, 2Ghz is not that far away now..
 
just a question vlad.....what happens when you run out of microns...how small can we get before we need a completely new interconnect process.....
 
Microns is a standard unit of measure. You can't really run out of them. In essence, the industry is already past microns demonstrated by the fact they are using fractions/decimals instead of whole numbers. There are varying opinions on when a whole fabrication technique will be needed or what the smallest possible process is. There have been a few supposed physical limits to creating smaller processes which have all been surpassed. So who knows where it will end.
 
I think the limit is around 20 nanometers or 0.02 microns. I read somewhere that IBM has come up with a process they call the &quot;V-Groove&quot; which may allow 10 nanometers, but won't be used in cpu's until about 15 years from now. So Moore's Law will still be alive for many more years.
 
the athlon uses the EV6 bus protocol... wonder if that means you can work an alpha into an athlon board somehow... would be super sweet.
 


<< a completley new die fabrication process >>



-I dont have a degree in Advanced EE (yet), but it?s a fairy simple question. I am sure PM can always add, as he has experience here.

You cant really &quot;run out&quot; of physical space (in microns), you can always (in theory) move smaller and smaller. To move smaller you need to implement a new etching process. (Extremely thin beams of electrons etch the transistors into the working silicon). For now IBM has reached 0.05 micron, but it?s not the end... IBM is working on a new technology called Prevail (Projection Reduction Exposure With Variable Axis Immersion Lenses), which promises big results in terms of &quot;small micron sizes&quot;.




<< the athlon uses the EV6 bus protocol... wonder if that means you can work an alpha into an athlon board somehow... would be super sweet. >>



LOL, you want to stick a huge Alpha on a tiny Athlon board? It?s just the same name 😉
 


<< LOL, you want to stick a huge Alpha on a tiny Athlon board? It?s just the same name >>



Eh, what is just the same name? The EV6 bus name of the Alpha and the Athlon? It's not just the same name, it is the same protocol. However, there are many other things that would prevent the transition from an Athlon to an Alpha on an Athlon mobo.

Patrick Palm

Am speaking for PC Resources
 
Yes, it?s the same name, but they are as &quot;an apple to a grape&quot;. Of course there are some similarities, but I don?t recall the name to be related directly.
 


<< the athlon uses the EV6 bus protocol... wonder if that means you can work an alpha into an athlon board somehow... >>

Well, the chipset has to support it. Alpha chipests will support the athlon (according to what I've heard anyway), but not the other way around, because the Athlon chipset uses a subset of EV6...or something to that effect. I'm not certain of this, so take that thought with a dose of salt.

As for that nice IA-64 chip....I would venture a guess that is in reference to McKinnely, while the Alpha @ 1.2ghz is the 21364 as designed by Samsung (they have a license which allows them to modify the core as compared to the one that API designed....which is the addition of the ondie L2 cache).

&quot;zero-penalty access for integer instructions&quot;....sounds interesting, wish I knew exactly what that meant (that is a bit vague if you ask me).

<< how small can we get before we need a completely new interconnect process..... >>

I don't know for certain, but I seem to recall PM taking a guestimate that it would be around .05 microns, but that the economic limit was closer to around .08 or so, but that was just an estimate....

<< There have been a few supposed physical limits to creating smaller processes which have all been surpassed. So who knows where it will end. >>

Well, what about the fact that atoms are NOT infentismely small. At some point, there won't be enough atoms to form a barrier between the wires, which I believe is part of the reason for moving to SOI (I won't even begin to claim understanding it fully), and I believe that this is partially to keep the electrons from jumping randomly from one wire to another.

Once you get small enough, that insulating part just can't get any smaller (you need X many atoms wide)...again, take everything I say in this post with a good dose of salt.
 


<< Yes, it?s the same name, but they are as &quot;an apple to a grape&quot;. Of course there are some similarities, but I don?t recall the name to be related directly. >>



The name is directly related, it IS the same bus protocol. The EV6 used by the Athlon is the same EV6 bus protocol that the Alpha uses.

Patrick Palm

Am speaking for PC Resources
 
Back
Top