Intel's Larrabee: A killer blow for AMD

[iOsiris]

Wasn't AMD/ATI doing something like this as well. Perhaps AMD knew this was coming and was a reason why they purchased ATI ?

[PyX]

Geh. Starting to really hate Intel... even though I love their processors. They're always making up stories to up their quotes and to lower AMD's.

Just do the f****** stuff before you say you'll put them on the market in 2009, it's still far away.

Anyway, as long as they keep up the good work with their processors... but if AMD dies because of their own speculations, there won't be any more competition and this can only be a bad thing.

[exanimas]

This is in a way reminding me of the movie 300 except Persia is Intel and AMD is the other group i forgot the name.

The other group is the Spartans. Only difference here is, Spartans were the better fighters but they were outnumbered. This was the case pre-Core 2. AMD had better products but was still the little guy. Now, Intel has better fighters and higher numbers (i.e. faster CPUs and more market share).

/Analogy

[Digitalx]

it's interesting this gpu's architecture is like CBE (Cell broadband engine) except giving the access to a full shared L2 ... as apposed to a general GPU you're seeing now... could potentially be a beast if it is i guess anyway... just have to wait and see, but yes hopefully we don't go back to rubbish intel only, just hope if AMD start faltering a private firm can step in and help them out.

But i feel this is a better piece of technology I've heard coming from Intel in awhile anyway

[TheRealDave]

AMD will be working on something big along the same lines after purchasing ATI. Can't wait for AMD's response but intel looks like a strong competitor so far!

[sundayx Veteran]

Intel Inside, Leap Ahead!

[Smigit]

I'd be interested to see what they release in 2009. Sure they might be designing the technology to run with up until 24 Cores but dont forget they also designed Netburst to take them to 10GHz and never got remotly close due to difficulties they never expected. I believe I read that the quad core cpu's use roughly twice the power of dual cores. For 24 cores to be practical the power consumption is going to have to be extremly low otherwise your probably going to struggle to have enough power in a household PC to power them. Infact I seriously would see such a setup making the currently over the top 2KW PSU's cry as they struggle to supply power in addition to any other hardware you may have.

I'd expect the 2009 release to be alot lower than some of the numbers listed here. As usual I'd look at a 10 - 20% boost over whatever else is out att he time. If they can achieve 20 - 30% improvements over 2008 tech then I think that would be a job well done.

I also wonder if 24 core cpu systems would mean the end of the ATX form factor. Thats going to take up a fair amount of room.

ATM the trend seems to be adding cores. Whats the bet that in a few years we will come to a point where more cores is no longer practical and we get back to a race of squeezing more out of each individual core (not necessaily raw clock speeds but). Not saying that isnt a focus now, but it does seem like the companies want to outcompete eachother by getting more cores in a system first.

[Ap0x]

This is in a way reminding me of the movie 300 except Persia is Intel and AMD is the other group i forgot the name.

That made me lol.. :shifty:

[Smigit]

This is in a way reminding me of the movie 300 except Persia is Intel and AMD is the other group i forgot the name.

too bad the "other group" got killed.

This is in a way reminding me of the movie 300 except Persia is Intel and AMD is the other group i forgot the name.

too bad the "other group" got killed. They put up a good fight and the Persian army got slack and underestimated the groups strength. But in the end the fighting took its toll and the little guys got overthrown.

and unlike the movie I dont think theres another 10,000 little guys waiting back in the main land to continue the fight.

[The_Decryptor Veteran]

I want one!

The hardware accelerated ray-tracing is reason enough, add in the 16-24 cores and it will fly (and i don't see why they can't hit 24 cores in 3 years time, they've already made an 80 core prototype)

[joeydoo]

I'd be interested to see what they release in 2009. Sure they might be designing the technology to run with up until 24 Cores but dont forget they also designed Netburst to take them to 10GHz and never got remotly close due to difficulties they never expected. I believe I read that the quad core cpu's use roughly twice the power of dual cores. For 24 cores to be practical the power consumption is going to have to be extremly low otherwise your probably going to struggle to have enough power in a household PC to power them. Infact I seriously would see such a setup making the currently over the top 2KW PSU's cry as they struggle to supply power in addition to any other hardware you may have.

True, there is always the potential for problems. But that 10gz quote was a much further ahead projection than the two years they are expecting for the release of Larrabee. Plus I think Intel has actually been fairly realistic and open about their chips since "core" has come along. The ghz race was halted due to physical limits of the circuits. This newer stuff is architecture based.

Intel also seem to have a determination to stay low power as well. There are constant improvements to reduce power general across the industry. By 2009 Intel will be down to 32nm. That along with the fact those "24 cores" aren't all going to be the same as the ones we have now. Some will be smaller and pure floating point, much like the spe's in the Cell processor.

There is also a move to reduce the power used by motherboards and Graphics cards. By 2009 the power used by PC's will probably be less than it is now.

I'd expect the 2009 release to be alot lower than some of the numbers listed here. As usual I'd look at a 10 - 20% boost over whatever else is out att he time. If they can achieve 20 - 30% improvements over 2008 tech then I think that would be a job well done.

I also wonder if 24 core cpu systems would mean the end of the ATX form factor. Thats going to take up a fair amount of room.

It's better not to think of this chip in regards to current CPU performance. But rather against the stream processing assistance that GPU are just starting to provide. Nvidia's cuda is expected to offer a 10Xtimes speed boost for applications that can make use of it. When you take some of what makes a gpu so fast and put that directly into a CPU, where there is super fast cache and no motherboard and FSB, software layer etc etc to get in the way... PLUS 2 years of CPU multi-core development, die shrinks, multi-threaded software improvements etc etc. A 4 times increase over things like Cuda isn't unrealistic.

The space probably won't increase, the idea is to get all those cores on a single chip. If people can play games that look decent with one of these things then the PC might well shrink due to the fact that people won't need a graphics card. Home PC's haven't really changed in form factor or size in 30 years. It probably won't in the near future. People don't want bigger systems anyway.

ATM the trend seems to be adding cores. Whats the bet that in a few years we will come to a point where more cores is no longer practical and we get back to a race of squeezing more out of each individual core (not necessaily raw clock speeds but). Not saying that isnt a focus now, but it does seem like the companies want to outcompete eachother by getting more cores in a system first.

The trend is actually already changing from maxing out the number of cores. Larrabee is evidence of that. The shift will be from multipurpose cores, to a making cores for specific tasks. The performance potential if you have a custom built processing core is unbelievable. For some tasks maybe 100's of times faster. As there is more and more space available due to shrinking processes, then more and more small cores will be added to CPU's... each for special tasks. The first step is the GPU, as we go along there will be probably be video encoding and decoding, encryption and decryption, physics, sound etc etc. This was the plan for Intel's 80 core chip they demonstrated a few months back. Though they haven't built the specialist cores yet... The thing most of these chip manufacturers are trying to work out is the interconnect tech and intelligence logic to split up what comes in and distribute it to the necessary core. The dedicated cores themselves will probably be the easy part. Making processors do a single task is simple, and many companies already do it for set-top multimedia devices.

[L3thal Veteran]

Intel is really trying to cripple AMD.

[Smigit]

The trend is actually already changing from maxing out the number of cores. Larrabee is evidence of that. The shift will be from multipurpose cores, to a making cores for specific tasks. The performance potential if you have a custom built processing core is unbelievable. For some tasks maybe 100's of times faster. As there is more and more space available due to shrinking processes, then more and more small cores will be added to CPU's... each for special tasks. The first step is the GPU, as we go along there will be probably be video encoding and decoding, encryption and decryption, physics, sound etc etc. This was the plan for Intel's 80 core chip they demonstrated a few months back. Though they haven't built the specialist cores yet... The thing most of these chip manufacturers are trying to work out is the interconnect tech and intelligence logic to split up what comes in and distribute it to the necessary core. The dedicated cores themselves will probably be the easy part. Making processors do a single task is simple, and many companies already do it for set-top multimedia devices.

Yeah we've actually covered in uni the possability of putting an FPGA along side a cpu so that the cpu would be a general processor and you could have an fpga programmed very specifically to do a set task. Certainly not something I expect to really see used seeing as at least at a consumer level thats not that useful but none the less, its another method of what you are describing. I guess your right yet but.

I wont comment on the rest, but thanks for the post, a good read.

[Punio4]

Hmm. But wouldn't the all-in-one type of product, even though superior in performance, be bad for the consumers?

No dedicated components > no possibility for an upgrade.

Also, which RAM would this thing's GPU use? Certainly it cannot fit its own memory onto the core, as one can do with the cache. Isn't the point why GPU has such massive bandwidth because there is no motherboard/software on the GPU?

[XerXis]

it just comes in waves, sometimes amd is better price/quality (or have you all forgot the pentium 4), sometimes intel is better. that's the way a free market works

[Smigit]

Hmm. But wouldn't the all-in-one type of product, even though superior in performance, be bad for the consumers?

No dedicated components > no possibility for an upgrade.

Also, which RAM would this thing's GPU use? Certainly it cannot fit its own memory onto the core, as one can do with the cache. Isn't the point why GPU has such massive bandwidth because there is no motherboard/software on the GPU?

Well dedicated doesnt mean you have one core for graphics and one for cpu (although it could). It can actually mean a number of things. For instance currently CPU's are extremly general purpose. They arent necessarily optimised to their fullest because they have to be able to run a huge set of opperations and do so well. However you could say have one CPU that was optimised for floating point arithmatic and thats all that core would ever do. You would optimise the processor so that core would be lousy as an actual cpu on its own, but when it came to floating point math it would be lightning fast. Then you have another core that does something else specialised, perhaps its job could be to handle opperations regarding stack operations and put that core near any memory registers ect ect. That scenario is one I've thought up quickly and realistically you may not split it into those levels, but I think you get the point.

Basically what your doing is that instead of having multiple cores each of which does everything in tandem, you have the same number of cores but give each a specific subset of work to do and optimise each one individually for it's given task.

You wouldnt eliminate the need for other devices, your just breaking up the role of the CPU. Now it also seems intel is about to attack the graphics sector. bare in mind they have always had graphics solutions and are the number one graphics solution on the market infact. Theres nothing that will stop users from using another graphics card if intels isn't up to par or you just want something different.

it just comes in waves, sometimes amd is better price/quality (or have you all forgot the pentium 4), sometimes intel is better. that's the way a free market works

In many markets yes. In terms of processors? Well it's questionable. AMD is not a small company but they are small comparred to intel. They were ahead for about 6 or so years but I would say that thats really the first time AMD had really held a solid lead, at least in the past 15 - 20 years. AMD's been around a while but had a huge success with it's Athlon series. I wouldnt call it a wave of success but, theyve had their ride but whether they will get on the board for a second time is yet to be seen. I dont think they are going anywhere, but whether they will dominate the market like they did earlier is in question. ATM their acquisition of ATI could be key to their success but we'll have to see how they utilise their new market potential.

Other markets do see a wave effect, but I don't think the phenomena has really applied to the processor market. At least not yet.

Edited June 20, 2007 by Smigit

[Danrarbc]

I believe I read that the quad core cpu's use roughly twice the power of dual cores.

Only because Intel's current quad cores are essentially 2 Core 2 Duos on one die. (IIRC Yorkfield remedies that)

[bgeh]

I'd be interested to see what they release in 2009. Sure they might be designing the technology to run with up until 24 Cores but dont forget they also designed Netburst to take them to 10GHz and never got remotly close due to difficulties they never expected. I believe I read that the quad core cpu's use roughly twice the power of dual cores. For 24 cores to be practical the power consumption is going to have to be extremly low otherwise your probably going to struggle to have enough power in a household PC to power them. Infact I seriously would see such a setup making the currently over the top 2KW PSU's cry as they struggle to supply power in addition to any other hardware you may have.

I'd expect the 2009 release to be alot lower than some of the numbers listed here. As usual I'd look at a 10 - 20% boost over whatever else is out att he time. If they can achieve 20 - 30% improvements over 2008 tech then I think that would be a job well done.

I also wonder if 24 core cpu systems would mean the end of the ATX form factor. Thats going to take up a fair amount of room.

ATM the trend seems to be adding cores. Whats the bet that in a few years we will come to a point where more cores is no longer practical and we get back to a race of squeezing more out of each individual core (not necessaily raw clock speeds but). Not saying that isnt a focus now, but it does seem like the companies want to outcompete eachother by getting more cores in a system first.

unfortunately, as far as intel's slides show, you'll probably only get 8 cores by 2010

http://en.wikipedia.org/wiki/Sandy_Bridge_...architecture%29

(arstechnica states that it probably will appear in 2009 though)

http://arstechnica.com/news.ars/post/20070...ee-part-ii.html

The presentation also includes some details about Intel's 32nm "Gesher" CPU, due out in 2009. In brief, it's 4-8 cores, 4GHz, 7 double-precision FLOPs/cycle (scalar + SSE), 32KB L1 (3 clocks), 512KB L2 (9 clocks), and 2-3MB L3 (33 clocks). The cores are arranged on a ring bus, just like Larrabee's, that transmits 256 bytes/cycle. Gesher is due out sometime in 2009.

from a slide that was taken down

for comparison, the theoretical FLOPs/cycle(double precision) is 4 for conroe (from http://www.xtremesystems.org/forums/archiv...p/t-102951.html 3rd post), giving about 12GF per core from a typical Core 2 CPU

[beniza]

So I am guessing that Intel will have a Larrabee "CPU" and a Larrabee "GPU" ? :laugh: