Intel following multicore model

Oldbitcollector (Jeff)

I love how Parallax is so ahead of the curve...

"Intel says to prepare for 'thousands of cores'"
news.cnet.com/8301-13924_3-9981760-64.html?part=rss&subj=news&tag=2547-1_3-0-5

Any chance Chip *might* own the patent on the multi-core idea itself?
<smirk>

OBC

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
New to the Propeller?

Getting started with the Protoboard? - Propeller Cookbook 1.4
Updates to the Cookbook are now posted to: Propeller.warrantyvoid.us
Got an SD card? - PropDOS
Need a part? Got spare electronics? - The Electronics Exchange

waltc

Multicores IMS predate Parallax. I do remember attempts albeit unsuccessfully during the late 80's early 90's time frame. It wasn't feasible because of yield problems.

As far patents go, my guess IBM, TI and Sun would hold some of the early patents on multicores.

As far as number of cores go, Rapport and IBM seems to have beat Intel to the punch on number of cores with the upcoming Kilocore.

FWIW

Forest Godfrey

I think the main reason that you're seeing the push to multicore now is that while the number of transistors you can pack onto a chip is going up, the speed really isn't. You can only get so much benefit out of deeper pipelines and better branch prediction units. If you look at a floorplan of a chip like a Pentium 4 single core, you'll notice that the amount of area devoted to the actual processor pipeline is tiny compared to the surrounding gear like the branch prediction, out-of-order logic, and cache.

So, you add cores instead. The problem is that adding cores doesn't always make your application go faster. In fact, we've seen cases on high performance compute applications where a multicore processor was actually *slower* than a single core running at the same clock rate. The reason is that at least one level of the cache is shared between cores. By having twice as many cores, you end up oversubscribing the cache causing more round trips to memory. This is compounded by fact that memory technology isn't keeping pace with processor technology. Back In The Day(tm) memory was as close as 1 clock cycle from the processor. Now, even 1st level cache isn't that close and memory is well over 100 clocks away.

Further, it's not clear how many applications will truely be able to take advantage of the kind of parallelism coming down the pipe from Intel and AMD (who will presumably do something similar). You can already see this in troubles porting some applications to the IBM Cell processors.

Now, the Propeller solves this problem. Unfortunately, its solution isn't ideal either. The Prop's solution is to strictly partition bandwidth between the 8 cores. That is great for predictability because the traffic generated from one core doesn't affect another. However, it's horrible for utilization unless all 8 cores are doing hub operations. If only one core needs access to the hub, you will utilize only 1/8th of the total available bandwidth.

To echo what waltc has said, the Prop, as far ahead as it seems to be in the microcontroller world, isn't the first multicore processor. In fact, the Thinking Machines CM-1 was a multicore chip and that dates back to, I believe, the mid 1980's. There may be examples further back than that, though I wasn't able to find any.

Ken Peterson

Having a multi-core processor within reach of the hobbyist is phenomenal! I don't know about patents, but I think if Parallax can get the Prop II out before anyone else makes a similar product, they will corner the market.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔

hinv

FWIW, the SeaFORTH24a is phenomenal as well. I haven't actually programmed mine yet, but it is really amazing and so tiny. There are some things I would change, and with the size of the die 2.7*2.7mm, they have the room.
It is 18 bit instead of 32.
The stacks could be deeper
The direction registers are switched around on a quite a few of the cores. North should mean north, always. Maybe I will understand why they did this when I start programming it.
There are only 64words of RAM & 64words of ROM per core
But most of all, they don't have the great people at their forum like we have here. You guys are amazing, talented, sharing people!

Tanks,
Doug

Praxis

said...


Kilocore, from Rapport Inc. and IBM, is a high-performance, low-power multi-core processor, with 1025 cores. It contains a single PowerPC processing core, and 1024 8-bit Processing Elements running at 125 MHz each, which can be dynamically reconfigured, connected by a shared interconnect. It allows high performance parallel processing.

Leon

Then there are the XMOS chips, which have four cores each running at 400 MIPS, with eight threads per core (scaleable architecture), and the Tile processor, with 64 cores.

Leon

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle

Post Edited (Leon) : 7/3/2008 3:08:37 PM GMT

Paul Baker

The concept of multicore is not patentable. If you have something which is prior art, having more of the thing is not a patentable distinction. For instance If I have a patent on a dining room chair, you cannot get a patent on a set of·my dining room chairs, because there is no distinctual difference in how the chairs work when you have more than one.·Where patents come into play is how you connect them together, or how they communicate with each other. Since there are thousands of different ways to do each of these, it is possible for thousands of patents related to multi-core to be issued (which there have been) without anyone infringing anyone else.


▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Paul Baker
Propeller Applications Engineer

Parallax, Inc.