Researchers' Tool Partially Automates CPU Design, Could Cut Development Time From Years To Months

10

Comments

+ Add a Comment
avatar

barbara009

my classmate's aunt makes $78 an hour on the laptop. She has been without work for 10 months but last month her income was $15994 just working on the laptop for a few hours. Read more on this web site NuttyRich.com

avatar

B.A.Frayd

Skynet continues to march towards reality!

avatar

teh 1337 haxxor

Gotta agree with John here. It's not really groundbreaking. If it was really THAT groundbreaking, it would have been published in a more prestigious journal...

After glancing through the article and looking at the results, there's really not anything groundbreaking. They can select and synthesize with different numbers of functional units for each core, but ultimately, the functional units of each core are the same. Only the width and pipeline depth of the cores changes. This is neat, but nothing special, and I don't think it will net a large performance gain.

I think that the processor industry will be headed in the direction of heterogeneous multicores (and while you could technically argue that this produces heterogeneous multicores, I wouldn't call it that). Basically, many different cores that are each good at something different. Offload specific operations from a large general purpose core to smaller specialized cores that are specifically designed to perform common operations (ie encryption/decryption, java accelerator, FFT co-processor etc). Basically, instead of using your general purpose CPU to play MP3's, the task would be offloaded to the FFT co-processor which would be able to accomplish the task very efficiently and quickly. Problem with this approach is the software programmers (most of whom can't even program very effectively for homogeneous multicores right now). Also, OS support is an issue (It would take a miracle for MS to get on-board with this and not find a way to screw it up).

avatar

Alegomaster

Heterogeneous cores are a quite good idea to use, however for now I see that they will be mostly used for research, and high-end processing. I don't believe that AMD or Intel will come out with anything related to it for at least 4 years, but RISC based designs may start to implement them. Imagine a Android phone with hardware accelerated Java on it. Now that is something Nvidia has to do. Maybe Arm too, because the current ISA may be unsuited for additional processing units, so ARM may need to come out with a new ISA too.

I have a feeling that you are going to love my new cpu Architecture that I am making.

avatar

John Pombrio

Hey, wait a minute. I used to be an IEEE member. I was quite proud of it!
Just read about efforts to build a javascript compiler that will output parallel processes (or something close to that, heh). That will be the next punch, getting a bunch of small cores to process at the same time. Rather than tie two or three different processors together like ARM and TI are doing, it will be better to have identical multiple cores. Kind of the holy grail of programming to be able to easily write parallel coding rather than using different types of cores.

avatar

bautrey

Cool. I was thinking about going to NCState for Graduate School in Computer Engineering. Now I have another reason to go there :D

avatar

John Pombrio

Really? Great new idea, eh Brad? Computer automated micro architectural circuit design has been around since the 1970's. By the time I started With HP in embedded processor development tools in '81, Intel was full blown into circuit mapping by machine. Heck, there hasn't been a "hand" designed CPU or embedded chip for close to 35 YEARS. Now these guys are trying to automate mashing together different CORE TYPES, not the nuts and bolts of core design, even the abstract tells you that (which no one has bothered to read here or on other places on the web). Nice idea but limited in implementation to specific classes of cores. Not bad but not groundbreaking. Groundbreaking was done back in the early 80's by Intel's amazing design tools. Watching the circuit design being done on a $100K workstation with a $20K 27 inch RGB monitor and outputting to a huge flatbed plotter back in the day was like watching magic. God knows what they are using now.

http://en.wikipedia.org/wiki/Integrated_circuit_design

avatar

Alegomaster

It's quite amazing on how much EDA tools have advanced. Now you can create complex circuits in minutes instead of hours or days, and you can do it for free (in some cases). And if you choose to use a FPGA, you can reconfigure it at any time!

avatar

Typo91

exactly what i thought. I was gona say... I thought they've been using a computer to map out the nity grity bits of how the paths will lay out on the actual silicon ever since the beginning...

avatar

Sierra11784

If this anyware close to being accurate, this type of software could potential allow several generational advancements in the time it used to take to one. If AMD jumps on this bridging the gap between their CPUs and Intel's won't be an issue, however the same holds for everyone else; ARM, Nvidia, Samsung, Qualcom, etc. Not to mention what this could do for the microprocessor industry in general; memory chips, controller chips, GPUs and Flash storage. If this software is adopted and implemented quickly it could mean Moore's Law is going to be thrown to the wind. Very exciting!

Log in to MaximumPC directly or log in using Facebook

Forgot your username or password?
Click here for help.

Login with Facebook
Log in using Facebook to share comments and articles easily with your Facebook feed.