Tilera Goes After AMD, Intel, With 100-Core CPU

Page 4 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
[citation][nom]Zoonie[/nom]Hurray for more competition![/citation]

Exactly my thought when I read this article. More pushing to AMD and Intel's sides will lower prices and advance their tech in faster pace normally. Unless ofc the new competitor doesnt aim at us with aproachable prices and everything stays as is. Which I still find hard. Lets wait and see if those numbers and models do come true to production further than papers and such.
 
A better question is, what ISA is it compatible with. If it's using its own ISA, then it might just as well be useless since there wouldn't be any OS that works without porting to it first.

Else, I'd hoping at least SPARC, MIPS or IBM's POWER?
 
processors and gpu's are incredibly complex things to engineer, alot of companies in the past have made applauded efforts at competing with intel, amd, nvidia, ati, time will tell and the proof will be in the pudding, one thing the company isnt short on is enthusiasm, . . . .YET!

i welcome more competition, cant wait to put one through the paces
 
[citation][nom]frozenlead[/nom]I want benchmarks.Even if these things are amazingly powerful, 98/100 cores (on occasion 96) still won't be able to be really utilized. Software just isn't that advanced yet.[/citation]
Depends if they've made a design where it splits up workloads on a hardware level instead of a software level, theoretically.
 
People need to understand the differences between CPU architectures, also the differences between CISC and RISC processing. Intel/AMD/Via x86/x64 are CISC CPU's, meaning they have large instruction sets that get decoded inside the CPU itself into smaller instructions that are then executed by various resources (FPU / ALU / MMU / ect..). This means a monolithic core capable of running complex operations in serial, the cores are much larger and require large amounts of cache to buffer the data. This allows the company to crank up the speed to reach impressive performance in serial, or minor parallel applications. For the past two decades this has been the standard because most desktop and light to medium servers required this nature.

RISC systems use a much smaller instruction set with a much simpler CPU design. The CPU's directly execute the instructions their sent, usually very fast with little wait times. But complicated tasks require more instructions to be sent, with each fetch / store being a separate instruction. But because of this the programmers can optimize to only execute the instructions the absolutely need and only fetch / store the exact memory they need. There isn't as large a demand for cache memory. Also since the cores are smaller you can fit multiple of them on a single package. But because of that design its much harder to ratchet up the clock speeds on these.

In short, RISC systems allow you to combine a massive amount of simple cores together to execute dozens (now hundreds) of simultaneous threads at once. CISC systems allow you to preemptively predict and execute out of order instructions from a single application. Both have their PRO's and CON's.

CISC
PRO: Able to execute single applications with high rates of speed and responsiveness.
CON: Complicated and doesn't scale well with parallel processing.
USE: Gaming, light to medium server work
EXAMPLE: Moving windows, copying files, browsing internet, writing emails, doing office automation work.

RISC:
PRO: Able to execute massive amounts of parallel instructions at once without needing complicated translations.
CON: Suffers in single application performance.
USE: Graphics rendering, Scientific research, Large database operations.
EXAMPLE: Rendering a animated movie using water refractions and ray tracing. Calculating physical interaction of an earthquake, weather patterns. Calculating the mass of the universe based on known physics models, calculating the effects of a Black Hole on a star system a few light years away.
 
Maybe it can scale better ? What kind of bus does it have for the motherboard?
32 cores ha ha ha. Just imagine running each part of the screen as its own subset. It might just speed up game play. Not enough information to understand anything yet. 😛
 
Status
Not open for further replies.