Tilera 100 core processor

TheRift said:
RISC vs CISC
Click to expand...

I dunno if this applies to all x86 chip manufacturers but way back AMD used something like a risc core with the x86 instruction set wrapped around it so effectively it was a risc processor, dunno if intel did the same though.

EDIT:

http://en.wikipedia.org/wiki/Reduced_instruction_set_computer#RISC_and_x86
Later, more powerful processors such as Intel P6 and AMD K6 had similar RISC-like units that executed a stream of micro-operations generated from decoding stages that split most x86 instructions into several pieces. Today, these principles have been further refined and are used by modern x86 processors such as Intel Core 2 and AMD K8. The first available chip deploying such techniques was the NexGen Nx586, released in 1994 (while the AMD K5 was severely delayed and released in 1995).

Although RISC was indeed able to scale up in performance quite quickly and cheaply, Intel took advantage of its large market by spending vast amounts of money on processor development. Intel could spend many times as much as any RISC manufacturer on improving low level design and manufacturing. The same could not be said about smaller firms like Cyrix and NexGen, but they realized that they could apply pipelined design philosophies and practices to the x86-architecture — either directly as in the 6x86 and MII series, or indirectly (via extra decoding stages) as in Nx586 and AMD K5.
Click to expand...

http://www.tek-tips.com/faqs.cfm?fid=788 good article.
 
Last edited:
I think it's a fine line between the 2 these days anyway. Parts of the CISC architectures probably doing RISC and so on.

RISC was great, but over time the instruction sets grew and grew to where one could look at it like a CISC.

RISC seems to be more on the embedded market these days. I do believe my old favourite MIPS is still in business. Yip, there it is, right in my routerboard. :D
 
Catal said:
With the GT200 chips you can run almost 30 000 concurrent threads, and they are almost doubled with the GT300. One of the Tesla rigs sport a 1 TFlops speed, but I may have seen a figure of 3 TFlops for the GT300 (may be wrong).One of the worlds biggest cloud computing networks, I think for SETI or something, has 400 000 pc's and boasts a 600 TFlops effective speed.
Click to expand...

Folding@home is actually running do significantly better... :D

As of April 9, 2009 the peak speed of the project overall has reached over 5.0 native PFLOPS (8.1 x86 PFLOPS) from around 400,000 active machines, and the project has received computational results from over 4.51 million devices since it first started.
Click to expand...

http://en.wikipedia.org/wiki/Folding@home
 
Yes Fermi can handle C++,c, Fortran and other languages via PTX which is an industry first and opens up a whole new level of programmability that is close too if not equal to what x86 CPUs can do today.
However, Fermi's in-flight threads are actually less than those of GT200. Somewhere around the 24,000 mark. Having said that Fermi is able to reach this theoretical limit on in-flight threads where GT200 couldn't in practice.
(Same story with old G70, dual issue in fragment shaders was possible, but almost impossible to use)

Correct me if I'm wrong, but these are not fully-featured cores? Pretty much like the CUDA stuff.

Call me when you have 100 full featured cores.
Click to expand...
Well to be fair Larrabee's "cores" are vector cores only so they are not fully featured either. Same as PS3's Cell processor is supposed to be 8 threads, but only 7 are enabled and those 7 are vector processors. You can't really say a core is fully featured, depends what it is required to do. One could say a single core on a Phenom II X4 is not fully featured as the core on a single Athlon64 3800+ CPU because it lacks its own exclusive IMC, shares a cache, has global scheduler and can't control its own P-states.

Fermi's stream processors are as full as they need be, as they have access to programmable L1 cache, access to global L2 cache, local scheduler, dispatch unit, and share a special function unit per SM. Each has a separate and independent Floating point and ALU unit of full and double precision as well.
Intel already has an 80 core research chip. They have developed a process whereby they can simply add more cores to the wafer by just plugging them in. If they want a 1000 cores they just connect more. However these chips are supposed to be another year or two away. This means more competition and hopefully more processing power.
Click to expand...
again so many x86 cores could not possibly fit on a single package the size of the CPUs we have today, the process node would have to be significantly smaller and I'm talking less than 0.9nm. 1,000 x86 cores will not be happening in 2 or 5 years for that matter.
Intel has something over Nvidia, it's called x86. Intel (and AMD) both had a license to make processors with x86, and OS like Windows and Mac both use x86 (linux doesn't). So even if Nvidia make a processor that's 1000 times faster, people can't use it with current OS (mac and pc).
Click to expand...
NVIDIA is not trying to make an x86 GPU that would prove very counterproductive. The GPU by design is a parallel processor as graphics rendering is inherently a parallel process. so a GPU executing x86 code would be hideously slow to the point of stupidity. Think also of the ASIC sizes of GPUs, at 3Billion gates clocking will be less than 700MHz, compared to at least twice that for CPUs that are designed with single thread IPC as key.
A GPU however is far more adept at parallel computing than a CPU is and this is why speed ups sometimes to the factor of 100X can be had with GPU computing. Not every process needs to be parallel, and not every process can be made parallel.
 
Last edited:
Most interesting. :)

So, ShockG, what is it you do for a living that warrants knowing this information by heart? :)
 
I'm a semiconductor fan (spin at 2,000 rpm most times heheh... cheesy I know :P)
seriusly though, I write for a publication or two online and in print. So it helps if I have an idea or two, (or risk turning into those people who keep saying the "good old days" Pentium 60Mhz etc...)

Relly great time in semicondutor industry right now from innovation POV. Convergence of CPU and GPU is is going to happen sooner or later (or an even tighter relationship will develop eg. Fermi supports full analogous transfers between host CPU and GPU). To think that the advent of the 3D accelerator was the divergence of serial-general processing and the parallel nature of 3d rendering or co-processing as it was.
 
Last edited:
Joy-Energiser said:
Most of what ShockG said is just way over my head.
Click to expand...

As a consumer, doesn't it irk you that you don't actually know what is in a processor, how it processes information, and how they make it?

For all you know, Intel could be making the processor from cute puppies by putting them in a blender? For all you know, the way it processes information could be by sacrificing babies and pregnant women to the devil to get answers, and to learn how to draw the pretty pictures on your monitor :eek::p
 
Yeah ShockG sounds like teh ubergeek, but I enjoyed reading his post.

The new Moore's law states that processors are not getting faster just wider, but you all know that by now.
 
konfab said:
Would surprise me if the next PS/Xbox would have something like this powering it
Click to expand...

it would surprise you? :rolleyes:

Considering the ps3 has the Cell Processor in it as is, the Xbox 360 has a gpu similar to that of the HD38x0, so I think this type of computer is what will pave the way of the future.
 
Yes that's true. I don't see next gen consoles having anything but a multi core CPU (8 minimum with additional vector threads is needed) and a multi stream processor GPU especially for the PS3 which is using an older type of GPU that isn't as programmable as the one in the 360 as its still discreet vertex and fragment units. No support for hardware relief mapping or anything like that.

As for the Xenos in current gen 360 its also quiet old school. More X1950XTX than HD2900. Feature wise it's still fixed function 2005/2006 Esq hardware
The Xenos may have unified vertex/fragment units, but there's only 48 of them and they don't support much past what directX9.0c can do which is why games can come out on both PS3 and 360 and look pretty much the same. The hardware generation is about equal.
 
ShockG said:
Yes that's true. I don't see next gen consoles having anything but a multi core CPU (8 minimum with additional vector threads is needed) and a multi stream processor GPU especially for the PS3 which is using an older type of GPU that isn't as programmable as the one in the 360 as its still discreet vertex and fragment units. No support for hardware relief mapping or anything like that.

As for the Xenos in current gen 360 its also quiet old school. More X1950XTX than HD2900. Feature wise it's still fixed function 2005/2006 Esq hardware
The Xenos may have unified vertex/fragment units, but there's only 48 of them and they don't support much past what directX9.0c can do which is why games can come out on both PS3 and 360 and look pretty much the same. The hardware generation is about equal.
Click to expand...

It's quite ironic then, all this war about wich console is best.
 
ShockG said:
I'm a semiconductor fan (spin at 2,000 rpm most times heheh... cheesy I know :P)
seriusly though, I write for a publication or two online and in print. So it helps if I have an idea or two, (or risk turning into those people who keep saying the "good old days" Pentium 60Mhz etc...)

Relly great time in semicondutor industry right now from innovation POV. Convergence of CPU and GPU is is going to happen sooner or later (or an even tighter relationship will develop eg. Fermi supports full analogous transfers between host CPU and GPU). To think that the advent of the 3D accelerator was the divergence of serial-general processing and the parallel nature of 3d rendering or co-processing as it was.
Click to expand...

You definitely need to direct me to some of your online publications. I've been away from hardware advances for about 6 years and it's been leaps and bounds since then. Would be interesting to focus again on what's going on out there.

I hear you on the "good old days", but I wouldn't want to go back there. Well, not with the size of today's software. I'm on some plak these days to get hold of some old machines I've liked in the past, some of them real monoliths. Things might have even been much more complex and advanced today would those companies still have been in business today.

General processing and 3d parallel processing have long been separated before the arrival of the 3d accellerator, but I think you were referring more to the mainstream segment.
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X