ATI enters technology development agreement with Intrinsity

WOW;

This could give ATi the edge they need to crush the competition.

Fast14 said:
Fast14â„¢ Technology

Fast14 â„¢ Technology provides a lasting advantage in design technology enabling Intrinsity to consistently deliver faster processors. It is a scalable design technology, proven in silicon with defendable intellectual property.

Dynamic logic yields two to five times the performance of static designs, but the gains in performance have rarely exceeded the increased costs in design time and effort required to generate a functional dynamic design. For years, designers have carefully handcrafted dynamic logic in speed-critical sections of designs when there was simply no other choice. For the first time, Intrinsity's patented Fast14 Technology allows realization of this speed advantage with the design productivity of static design flows for the entire chip.

Fast14 Technology removes the impediments that restrict dynamic logic use while requiring only a fraction of the resources and design time for best-in-class, dynamic circuits. Fast14 Technology more than doubles the realizable frequency in a given process technology.

With over 83 patents filed, over 62 granted and 600,000 lines of EDA software, Intrinsity's Fast14 Technology enables a five times speed advantage over ASIC design flows. Using a standard CMOS process, Fast14 Technology can provide up to three times the performance of carefully crafted, custom, static design methods.
Click to expand...

Intrinsity produces(?) an advanced math coprozessor @ 2.5GHz. As far as I remember the Prozessor is based on an MIPS-Design. Other Mips-Chips have only surpassed 1GHz, so their claims could be real.

Fast14 Technology Abstract

Fast14â„¢ Technology provides a lasting advantage in design technology enabling Intrinsity to consistently deliver faster processors. It is a scalable design technology, proven in silicon with defendable intellectual property.

Developed for use with standard CMOS processes, Fast14 Technology provides breakthrough, multi-GHz performance with the productivity associated with industry-standard, semi-custom, static design tools. Three years of development have resulted in completion of Intrinsity's Fast14 Technology. This technology solves the productivity and manufacturing problems of dynamic circuits. Fast14 Technology consists of a combination of four patented elements:

-Unique clocking style - Addresses race conditions associated with dynamic logic

-1-of-N Dynamic Logic (NDLâ„¢ family) - A new dynamic logic family that dramatically reduces the number of gate delays required for a given function

-Wire Twizzlingâ„¢ process - Reduces noise in routed signals

-EDA tools suite - Fully automates the implementation of dynamic circuits

Fast14 Technology has been used to design and produce a test chip and the FastMATHâ„¢, FastMATH-LPâ„¢, and FastMIPSâ„¢ processors. Analysis of these devices indicates a process margin and voltage margin similar to static circuits, proving the suitability for robust volume production of products incorporating Fast14 Technology.

Fast14â„¢ Technology derives its name from the atomic number of silicon, which directly translates to "Fast Silicon."
Click to expand...

Link : http://www.intrinsity.com
 
vnet said:
mboeller said:
This could give ATi the edge they need to crush the competition.
Click to expand...

Problem is, we probably won't see this before R600 or even later.
Click to expand...


I don't think so.

You can use the normal CMOS-process/technology and enhance the product with the Fast14 - dynamic-logic process. So you design the GPU like normal, but with the Fast14 processenhancements implemented during the design already.

So maybe we will see it already in the R500 cause companies anounce something when the deal is done, not before. This means IMHO that ATi has already tried to implement an Fast14 chip/part of an chip and it worked.

If you look in the console-forum at the moment they "discuss" the merits of 90nm tech. and the problems of it. Fast14 could be an easy way out of this trap. You can design chips up to 4 times faster ( see eetimes article) than with an normal CMOS-process and the same process technology, so an very small 2 Pipeline-Design @ 2GHz would be possible and at least as fast as an 8Pipeline-Design @ 500MHz (even without taking into account the problems with branching, which will favour an implementation with only a small amount of pipelines). So you replace an monster with 107 Mio transistors with an very small chip with maybe only 35-40 Mio Transistors. That opens the way for very high performance embedded chips and would be an hugh benefit for the R500-based XBox2 too. Therefore it would be (IMHO) nonsense to implement this tech only with the R600.
 
What about heat generation using this "Fast14" process? Are we going to get smaller, much faster chips that generate loads of heat and therefore require huge heatsinks and fans? I'd have thought that for a console especially cooler chips would be much more preferable.
 
mboeller said:
If you look in the console-forum at the moment they "discuss" the merits of 90nm tech. and the problems of it. Fast14 could be an easy way out of this trap. You can design chips up to 4 times faster ( see eetimes article) than with an normal CMOS-process and the same process technology, so an very small 2 Pipeline-Design @ 2GHz would be possible and at least as fast as an 8Pipeline-Design @ 500MHz (even without taking into account the problems with branching, which will favour an implementation with only a small amount of pipelines). So you replace an monster with 107 Mio transistors with an very small chip with maybe only 35-40 Mio Transistors. That opens the way for very high performance embedded chips and would be an hugh benefit for the R500-based XBox2 too. Therefore it would be (IMHO) nonsense to implement this tech only with the R600.
Click to expand...

The problem with dynamic logic being design times (which this deal is all about) and power consumption. Dynamic logic eats more power, combine this with leakage problems at smaller linewidths and suddenly you got a very nice heater for those cold winter evenings. Xbox2 - Now with afterburner simulation for flight sims...

Edit: Mariner beat me to it...
 
Aivansama said:
The problem with dynamic logic being design times (which this deal is all about) and power consumption. Dynamic logic eats more power, combine this with leakage problems at smaller linewidths and suddenly you got a very nice heater for those cold winter evenings. Xbox2 - Now with afterburner simulation for flight sims...

Edit: Mariner beat me to it...
Click to expand...


Yes and no....;)

An A64 @ 2GHz has an TBD of 89Watt, but the amount of logic gates is higher. The P-M @ 1.6GHz has an TBD of only around 25Watt (from memory only). [ I know this comparison is not entirely correct, cause CPU's are not GPU's but do you know of any GPU @ 2GHz ;) ]

So an really small GPU with only around 20-25 Mio Logic transistors and an overall transistor-budget of 40Mio should be able to work with only (pure guess) 40-50Watt, especially with ATi's experience to design "frugal" GPU's. So yes this sort of GPU will need a lot of energy when compared with an normal GPU with the same amount of transistors, but on the other side, it will need nearly the same amount of energy as an highend GPU @ 500MHz. The only downside is the small die so it will be difficult to get the heat out of the chip ( Maybe the Coolchips-tech can help here ).
 
You must log in or register to reply here.
Back
Top