Speculation - I7 16 core 32 core cpus?

S

Sunin

[H]ard|DCer of the Month - August 2008
Joined
Dec 27, 2005
Messages
3,421
So when is the question. We have had dual core and quad cores for what feels like ages and the promise of even the last architecture was 16 cores, now it is refreshed in the new architecture. Will it be seen?

Looked at the Intel roadmap and I'm not seeing mention of a 16 core I7 processor yet. I know I'm not jumping that boat until then.

How much longer until they get 32 core versions out 2 more years, maybe early 2010 for the 16 cores and 2012 beore 32 cores are available?

Your thoughts?
 
Before we go 16/32, we need to see a 8-core cpu. This will be a possibility with Westmere for sure next year... For the rest, dunno...
 
I thought 6-core was next, and possibly 8-cores at the end of that? 8 to 32 is kind of ridiculous though, we're just starting to see 4-cores get used a little. I want processors with higher clocks/clockability and less power/heat first.
 
it wont be called the i7.. and we wont see 6 8 16 cores til the move to 32nm.. and we will probably see a 8 core processor from amd first once they make the switch to 22nm..
 
Westmere (originally called Nehalem-C), the successor to Nehalem, will initially be Quad Core and/or Sextuplet Core but most likely Sextuplet only. Westmere marks the move to 32nm technology and scheduled to launch in 2010. I doubt we will see any native 8-cores in the Westmere family. What we we will probably see though in 2010 is a Sextuplet core with hyperthreading and that will be pretty cool.
 
That's not what I meant, since I don't think we are talking about Xeons. We're talking about Westmere and beyond.
 
Tobit said:
Westmere (originally called Nehalem-C), the successor to Nehalem, will initially be Quad Core and/or Sextuplet Core but most likely Sextuplet only. Westmere marks the move to 32nm technology and scheduled to launch in 2010. I doubt we will see any native 8-cores in the Westmere family. What we we will probably see though in 2010 is a Sextuplet core with hyperthreading and that will be pretty cool.
Click to expand...


with the delay on the i5 and p55 chipset from intel id say we are more likely to see westmere late 2010 early 2011..
 
sirmonkey1985 said:
with the delay on the i5 and p55 chipset from intel id say we are more likely to see westmere late 2010 early 2011..
Click to expand...
Most likely, yeah. :mad: Initially, Intel was hoping for late 2009 on some Westmeres which I thought was overly optimistic from the start.
 
the thing with intel is that they seem to keep shooting them selves in the feet with the constant socket size changes and i think its going to piss a lot of people off.. because i know the 32nm cores wont use 1366 either.. so we will have to see what happens when they are released.. and the way amd is going i think am2+ and am3 will have a long life down the road which makes upgrading much easier for people..
 
sirmonkey1985 said:
with the delay on the i5 and p55 chipset from intel id say we are more likely to see westmere late 2010 early 2011..
Click to expand...
Westmere is independent of i5, though. It shares the same socket as the i7s.
sirmonkey1985 said:
the thing with intel is that they seem to keep shooting them selves in the feet with the constant socket size changes
Click to expand...
LGA-775 has stood around for a relatively long time. With i7, they've introduced one new socket. With i5, another. So, two more, total, after a few years.
because i know the 32nm cores wont use 1366 either..
Click to expand...
Are you talking of Westmere specifically? Everything I've read says that Westmere will be 1366.
 
hmm now they are saying it will be 1366.. for a while they were talking about using a totally different socket for the 32nm.. but who the heck knows at this point with intel.. especially since amd isnt giving them any competition.. i think the only hope of seeing westmere is if AMD does come through and sets a realistic release date for its 22nm cores..
 
Current Bloomfield CPUs will be 4-core only. Westmere is the 32nm die shrink, and will be 4-core and 6-core. There may be an 8-core CPU released sometime after that, but 16 and 32-core CPUs are a ways off right now. I wouldn't expect to see a consumer CPU with that many cores until 2011-2012 at the earliest.
 
sirmonkey1985 said:
the thing with intel is that they seem to keep shooting them selves in the feet with the constant socket size changes and i think its going to piss a lot of people off.. because i know the 32nm cores wont use 1366 either.. so we will have to see what happens when they are released.. and the way amd is going i think am2+ and am3 will have a long life down the road which makes upgrading much easier for people..
Click to expand...

Why won't Westmere be 1366? That would be retarted if Intel expected people to change sockets every 2 years.
 
crazjayz said:
Why won't Westmere be 1366? That would be retarted if Intel expected people to change sockets every 2 years.
Click to expand...
Westmere will certainly use LGA1366. It's only a process change and slight tweak compared to Bloomfield, similar to the situation with Wolfdale/Yorkfield and Conroe/Kentsfield.
 
crazjayz said:
Why won't Westmere be 1366? That would be retarted if Intel expected people to change sockets every 2 years.
Click to expand...

I don't think the socket changes will be that ofter ... every 3-5 years is about right ...

Look back at how long 478 and LGA 775 have been around ...

Core i7 required a different pin configuration so the change was necessary ...
 
The biggest thing I really am interested knowing is whether or not Westmere will be supported on the X58 chipset.
 
I see i7 as nothing more than a blend of say AMD and and the best 2.4c intel cpu.

I've held up on building my own i7. I want to see main board revisions and a step on the cpu's / revision. I will then melt some plastic. Its a shame AMD is struggling. and I blame that on there end users. And I also dislike to some degree how 7 was launched. I think the FSB controller is a bit raw and the new so called hper threading is a bit a skew.

Until things are fine tuned. I will keep beating the crap out of my Q6600.:)
 
psyshack said:
I see i7 as nothing more than a blend of say AMD and and the best 2.4c intel cpu.
Click to expand...
I couldn't be happier with my Core i7. It's great. I have it overclocked to 3.5GHz and it's been stable for a month folding at 9200 to 9800 PpD. Temps never go over 60C when folding.
 
psyshack said:
I see i7 as nothing more than a blend of say AMD and and the best 2.4c intel cpu.

I've held up on building my own i7. I want to see main board revisions and a step on the cpu's / revision. I will then melt some plastic. Its a shame AMD is struggling. and I blame that on there end users. And I also dislike to some degree how 7 was launched. I think the FSB controller is a bit raw and the new so called hper threading is a bit a skew.

Until things are fine tuned. I will keep beating the crap out of my Q6600.:)
Click to expand...

Um, kinda? I see where you're coming from with the blend of AMD and the best 2.4c... but I would say that 2.4c is now some blazing fast 4.0. Haha. I do agree though that Intel may drastically improve on their IMC technology in the coming years (seeing how this is the first chip in a while to use an IMC).
 
As long as they can continue to get more performance/clock they can slow down the need for multi-core processors. They will stick to Moore's Law as long as they can. If that requires adding more cores they will do it, but as long as they can sell new product for a premium and outmode old stuff they will continue with what they are doing. I'm not saying it's a bad thing. So long as we get the performance we want and the prices stay sane I'm happy.
 
Problem we face is we have been at the 3.2ghz speed since P4... Either we need some large jumps 6ghz etc or we need more cores. I feel we have seen very little progress in the last couple of years. Yes performance per clock has grown, but not at the pace it was once at.
 
I'm not sure how many of you have looked at the overall economy of late, but it's a bit down. R&D takes tons of cash and these days cash is a bit short, even for Intel. Then add in the entire tech market is down and showing no signs of going up anytime soon.

Intel makes right around 60% profit on the whole sale end (as opposed to say NewEgg who makes 10 to 15%) but that doesn't help if you don't have customers. Unless Big Business decides to upgrade to something fantastic machine wise there isn't any reason to build new product.;)
 
Screw that I want a supercomputer on a chip!!! and I want it now!
 
Sunin said:
Problem we face is we have been at the 3.2ghz speed since P4... Either we need some large jumps 6ghz etc or we need more cores. I feel we have seen very little progress in the last couple of years. Yes performance per clock has grown, but not at the pace it was once at.
Click to expand...

Why can't we just get better processing / Ghz through new architectural design? It doesn't always have to be about raw clocks.
 
crazjayz said:
Why can't we just get better processing / Ghz through new architectural design? It doesn't always have to be about raw clocks.
Click to expand...

How about this, a new OS kernel with properly written gates and properly timed steps? I don't recall the guy's name ATM but he made significant changes in Linux Kernels and got much improved performance. You won't ever see those changed unless you can find one of his kernels on a site somewhere. He pretty much broke all the established rules, took a ton of heat from Linus himself so he said screw it and went on to making money.

I do agree about the architecture idea, as it is we have more wait states then enough on these high clocked CPUS and DDR-3 and poor read write speeds on drives don't help a bit.;)
 
BillR said:
I do agree about the architecture idea, as it is we have more wait states then enough on these high clocked CPUS and DDR-3 and poor read write speeds on drives don't help a bit.;)
Click to expand...

No worries, SSDs are dropping every month. It think once they get to the $1.50/GB price point for something like the Intel X-25, people will start to consider them for boot drives.
 
crazjayz said:
No worries, SSDs are dropping every month. It think once they get to the $1.50/GB price point for something like the Intel X-25, people will start to consider them for boot drives.
Click to expand...

Maybe, but that can't fix the wait states in the ram, and that's a ton of cycles we are throwing away.
 
crazjayz said:
Um, kinda? I see where you're coming from with the blend of AMD and the best 2.4c... but I would say that 2.4c is now some blazing fast 4.0. Haha. I do agree though that Intel may drastically improve on their IMC technology in the coming years (seeing how this is the first chip in a while to use an IMC).
Click to expand...

Thermal constraints and performance. Power doesn't always scale linearly with clock speed; i believe it's more of an exponential increase (which is why when the G5 hit a certain clock speed, apple watercooled it). As for improving the IMC, i'm sure that they'll do that, but how much of a performance benefit will it really bring? I remember reading a review that tri-channel didn't change performance by a ground breaking amount, but that was a while ago.

For the original thread; it has to do with cost. Semiconductors are all about trade offs. You can have a 16 core CPU, but you'd need an exorbitant amount of cache to feed all of the cores. There was a prediction that once you hit around 16 cores, the performance drops off because of the lack of bandwidth. Although, my gut feels differently as we've seen bandwidth increase by several times in just the past year.

Also, it has to do with yield. You have a huge die, and as a result, less die/wafer.

I think the future is a hybrid approach. Intel said that they wanted to go many core, and imo it's going to be a blend of normal cores (like Nehalem) and Larrabee esk cores. Essentially, i think it's going to be more and more cell like. Off load heavy floating point operations to the Larrabee esk cores freeing up the "smart core" to handle other operations. That's waht i'd like to see.
 
Duke3d87 said:
Power doesn't always scale linearly with clock speed; i believe it's more of an exponential increase
Click to expand...
Actually, power scales linearly with frequency and exponentially with voltage (in an idealized model of a CPU, that is).
 
yeah you're probably right. But couldn't it be said that to hit a certain clock speed, you need to increase voltages which would cause the exponential increase in power?

When overclocking my quad, I could hit say 3.4 on a pretty reasonable voltage, but to hit 3.6 i had to increase the voltage much higher relative to the amount going from 2.4 to 3.4.
 
Duke3d87 said:
yeah you're probably right.
Click to expand...
Feel free to look it up if you don't believe me. The equation for power consumption of an idealized microprocessor is P = C*(V^2)*f.
Duke3d87 said:
But couldn't it be said that to hit a certain clock speed, you need to increase voltages which would cause the exponential increase in power?
Click to expand...
Well, not necessarily. That depends on the design of the CPU and how much voltage it requires. For example, my Q9550 can run comfortably at 3.4GHz with stock voltage.
 
You must log in or register to reply here.
Back
Top