Nehalem?

Status
Not open for further replies.
S

spock

Gawd
Joined
Apr 5, 2000
Messages
521
So Intel is launching Nehalem late this year, and it's looking to have more cores, higher IPC, lower latency/higher bandwidth memory access, and faster clocks (I've seen 3.2GHz on an ES). Fine, but what is AMD planning to do about it? Surely not just a faster clocked 45nm Phenom?

I'm just thinking to myself: forget this "round," it's clear who the winner is already. nVidia and ATI have both essentially "sat out" of a GPU refresh cycle and came back with a hugely successful product for the next one (I'm thinking Geforce 5800FX). AMD has enough depth in it to continue for another 12 months on what they already have, but I see it as absolutely essential to compete in 2009. Maybe it's unannounced but I'm wondering if they have mentioned any products that will be out in the Nehalem timeframe?
 
Not really more cores for enthusiasts. Desktop market is still going to be limited to quad core, though each core will have the next generation of hyper-threading... The Octo-core is going to be for xeon server systems only from what I've read.

Nahalem is likely going to be launched by Christmas, if the rumors hold up, but I would guess only in $1000-$1200 Extreme Edition format. Mainstream, affordable parts of Nahalem will not likely be ready for cost conscious enthusiasts and consumers until this time in 2009 (late Q1 or Q2) would be my guess.

The next thing up for AMD is Shanghai. But it's still based on the K10 architecture and will be another refresh on a shrunk down 45nm process. Rumors I've seen say it's going to be "sometime" in 2009, and may also have 8core by linking a pair of quad core together in the same package as Intel has been doing. They may need 8core in order to keep up with 4core+hyperthreading of Nahalem's. Who knows *shrug*, too early to tell.

Everything is still largely rumor.

I have my doubts that Shanghai is the product AMD will take back the performance lead crown with.
 
AMD's only answer to the Nahalem for the consumer market in 2008 will be the Deneb core, which is slated for H2 of 2008, but probably overlapping into 2009. That is the 45nm die shrink, clockspeed ramp (finally!!) and possibly some other improvements to their SSE logic, not really sure about that last bit.

The interesting part is that AMD seems to be going gung-ho for the low-cost market and may be set up to do quite well there. The enthusiast class Deneb, likely the successor to the FX line, will be loaded up with 6MB of L3 cache. However, there is also a rumored L3 cache-less design in the works, the "Propus". Those 6MB of cache take up quite a bit of die space and also increase the thermal spec of the core. Without the cache you wind up with a chip using far fewer materials (= cheaper) that does not create as much heat.

Similar to the X3 variants of the current Barcelona dies, those cache-less designs seem poised to enter into a market segment that Intel doesn't really have a product for besides the Celeron and older Core2 CPUs. OEMs love those X3 chips because the average joe walks into CompuShop and sees that 3 > 2. If they can match the price of the X3 with these Propus cores then there's an even bigger advantage, 4 is twice as good as 2! How much of a performance impact is incurred going without L3 has yet to be seen, but for most consumer applications I would guess very minimal, the L2 cache available would suffice.
 
AMD is probably betting on fighting with Shanghai until Montreal/Fusion/Bulldozer.
They are betting that Nehalem will be tough to produce (big die, Intel's first IMC, more expensive to make). WIth the increased cost, Intel can only sell these at the high end, and will leave only Penryn to compete with Shanghai in the desktop market which should be a close fight.
 
Deliximus said:
AMD is probably betting on fighting with Shanghai until Montreal/Fusion/Bulldozer.
They are betting that Nehalem will be tough to produce (big die, Intel's first IMC, more expensive to make). WIth the increased cost, Intel can only sell these at the high end, and will leave only Penryn to compete with Shanghai in the desktop market which should be a close fight.
Click to expand...
Penryn is Intel's test bed for their 45nm process. Now that that is done, and their upcoming Dunnington variant with 6 cores, Intel is going to be well practiced at producing high yield 45nm parts.

I think AMD would be a foolish to put much faith in that line of thinking.

Core2, conroe and penryn recycled the old lga775, so the new tooling and mechanicals, and getting all the third party taiwanese board makers up to speed may be a legitimate bottleneck moving to a new socket... but I wouldn't count on that either.

They just need to focus on getting Shanghai out in time, and marketting it in a good cost/performance bracket like they are doing with Phenom right now. And then strike back hard at the high end later with Bulldozer or their next major micro-architecure change.
 
Menelmarar said:
Penryn is Intel's test bed for their 45nm process. Now that that is done, and their upcoming Dunnington variant with 6 cores, Intel is going to be well practiced at producing high yield 45nm parts.

I think AMD would be a foolish to put much faith in that line of thinking.

Core2, conroe and penryn recycled the old lga775, so the new tooling and mechanicals, and getting all the third party taiwanese board makers up to speed may be a legitimate bottleneck moving to a new socket... but I wouldn't count on that either.

They just need to focus on getting Shanghai out in time, and marketting it in a good cost/performance bracket like they are doing with Phenom right now. And then strike back hard at the high end later with Bulldozer or their next major micro-architecure change.
Click to expand...

u are bang on with that.
there's nothing much AMD can do about nehalem at this point except to tweak out Shanghai. K10 itself right now does perform very well in server 4p and above, so Shanghai will make that attractive still by being socket compatible.
 
Deliximus said:
AMD is probably betting on fighting with Shanghai until Montreal/Fusion/Bulldozer.
They are betting that Nehalem will be tough to produce (big die, Intel's first IMC, more expensive to make). WIth the increased cost, Intel can only sell these at the high end, and will leave only Penryn to compete with Shanghai in the desktop market which should be a close fight.
Click to expand...

Not true at all, mainstream Nehalem chips will arrive in Q1 2009. Nehalem's die size (246mm^2) is almost identical to Shanghai (243mm^2).
 
spock said:
(I've seen 3.2GHz on an ES)
Click to expand...

AMD showed off a 3.0Ghz ES for K10 B2 stepping. What is your point?

I've heard rumors that the IMC is going to severely limit intel's clocks, and they're banking more on IPC improvements and cores over frequency. Plus, it's supposedly going to have a pathetically low multiplier.
 
sdlvx said:
AMD showed off a 3.0Ghz ES for K10 B2 stepping. What is your point?

I've heard rumors that the IMC is going to severely limit intel's clocks, and they're banking more on IPC improvements and cores over frequency. Plus, it's supposedly going to have a pathetically low multiplier.
Click to expand...

The difference is that Intel generally doesn't show off stuff it won't release in future, or is at least confident of doing so. ;)

Think back to IDF 2006, they demoed a Conroe @ 2.67GHz, and the top bin at release 3 months later was 2.93GHz.

I'd like to see where those rumours came from, would it happen to be AMDZone? :p:D

General consensus is that Nehalem will clock similarly to Penryn, perhaps a touch higher. It will run hotter clock for clock because of the IMC, which will account for 15 - 20W. However, Nehalem is built on a 2nd generation HK/MG process, which improves upon the current HK/MG process on Penryn chips.

David Kanter (respected CPU architecture expert) has an excellent writeup on Nehalem at realworldtech, and he doesn't mention any of those points from your rumours, but hey, fanboys probably know more about Nehalem than he does, right? ;)
 
Of course, anything that's not explicitly said in one review must not be true. David Kanter's review is the bible of CPU Archs, and if it's not in there, it's not true. Silly me!
 
harpoon said:
I'd like to see where those rumours came from, would it happen to be AMDZone? :p:D
Click to expand...
So low, but so true. The last 22 months have been a terrible nightmare to some of the posters at AMDZone. :p
 
sdlvx said:
Of course, anything that's not explicitly said in one review must not be true. David Kanter's review is the bible of CPU Archs, and if it's not in there, it's not true. Silly me!
Click to expand...

Please enlighten us why would the IMC restrict Intel's clocks ??

Due any of you geniuses happen to know how the IMC is build , its trade offs and so on ?

IIRC IMC is asynchronous with the cores , that means the cores can use a different frequency.Secondly , Intel has showed time and time again it knows how to build state of the art MC.
 
harpoon said:
The difference is that Intel generally doesn't show off stuff it won't release in future, or is at least confident of doing so. ;)

Think back to IDF 2006, they demoed a Conroe @ 2.67GHz, and the top bin at release 3 months later was 2.93GHz.

I'd like to see where those rumours came from, would it happen to be AMDZone? :p:D
Click to expand...

No, the rumors came because the uarch's stated that the silicon type used for the MCH is much different than the type used for CPU logic. That would prevent high clock speeds. But the term "high" is wholly subjective - especially when we compare to Penryn. Penryn's higher quad core bins can easily be released commercially at 4Ghz *right now*.
 
dexvx said:
Penryn's higher quad core bins can easily be released commercially at 4Ghz *right now*.
Click to expand...
Says who? There's far more to a commercial release and Intel meeting processor specifications than enthusiasts on a forum stress testing with Prime95/Orthos.
 
Wait, wait. So there are people here who actually think that because Intel can get a little above 3.0GHz on two separate 107 mm^2 dies (QC Penryn), it means that Intel will be able to hit around the same clocks on one single die at 246 mm^2 (Nehalem)?

These people should really hit the web for some yield estimators, binning estimators, and defect density modeling. While one could never find real values for the actual Nehalem (yields, defect densities, and bins are closely guarded secrets), one will see, after a small amount of effort, that one single die with size X is not going to be able to be produced with the same bins as two dies which sum to the same size X.

My personal view is that Intel will probably be able to hit around 2.8GHz to possibly 3.0GHz on the top on the non-XE list. There may be one XE model a bump over 3GHz, however.

Will Nehalem be a monster chip that gives AMD some really rough competition? Sure. Will Intel still hold the performance crown at the very high end? Almost certainly. However, the logic used here by quite a few people is just silly. Just to barely scratch the surface, read: http://www.siliconfareast.com/test-yield-models.htm . Doubling the effective die size is going to make a huge difference in yields and binning.
 
visaris said:
Wait, wait. So there are people here who actually think that because Intel can get a little above 3.0GHz on two separate 107 mm^2 dies (QC Penryn), it means that Intel will be able to hit around the same clocks on one single die at 246 mm^2 (Nehalem)?
Click to expand...
And you're saying Intel can't, even if > 3GHz is a design target frequency? Bookmarked.
 
Intel can hit 3 GHz with 65 nm. Who knows what 45 nm could hit if Intel had a reason to ramp up the clockspeed (and adopt the original 65 nm TDP specs).
Let's not forget that Intel also built the Pentium D 800-series, which was a big single-die (!!!! only 900-series was two dies) 90 nm dualcore (about 215 mm^2) that clocked at pretty impressive clockspeeds (up to 3.2 GHz).
So Intel already has plenty of experience with building large dies and running them at high clockspeeds.

Also, these Pentiums had big caches and superfast and long pipelines. Nehalem will have relatively small caches, no superlong or double-pumped pipeline, and instead part of the diesize is taken up by a memory controller, which doesn't have to run at such high clockspeeds anyway.
So in a way, Nehalem isn't really that much of a risk compared to what Intel has already achieved in the past. Considering that the engineering samples are looking very good about half a year before launch, it's not very likely that Intel is going to screw up this round.
 
dexvx said:
No, the rumors came because the uarch's stated that the silicon type used for the MCH is much different than the type used for CPU logic. That would prevent high clock speeds. ...
Click to expand...

In what way different ? Chipsets are made in n-1 or n-2 CPU process.While the transistor characteristics may be different , the difference can't be great.

If anything , the MCH should clock higher on state of the art transistors.

You may be onto something , but that's like shooting blindfolded : the main problem with IMC is the fact that on die IMCs operate at far lower voltages than the DRAM links.With traditional chipsets , you could use higher voltages and circumvent this problem.
I'm pretty sure though that Intel had the manpower and the resources to overcome this.
 
visaris said:
Wait, wait. So there are people here who actually think that because Intel can get a little above 3.0GHz on two separate 107 mm^2 dies (QC Penryn), it means that Intel will be able to hit around the same clocks on one single die at 246 mm^2 (Nehalem)?

These people should really hit the web for some yield estimators, binning estimators, and defect density modeling. While one could never find real values for the actual Nehalem (yields, defect densities, and bins are closely guarded secrets), one will see, after a small amount of effort, that one single die with size X is not going to be able to be produced with the same bins as two dies which sum to the same size X.

..
Click to expand...

Dies close to each other have the same characteristics.Did you take this into account as well ?

Secondly , Intel claims world class yields and defects.Manufacturing prowess is their core asset and I'm pretty sure Intel did its job on Nehalem.
 
savantu said:
You may be onto something , but that's like shooting blindfolded : the main problem with IMC is the fact that on die IMCs operate at far lower voltages than the DRAM links.With traditional chipsets , you could use higher voltages and circumvent this problem.
Click to expand...

Well, my 965 chipset also runs at pretty low voltages. I think it's 1.025v or something, I'd have to check the BIOS settings, but it's far lower than the 1.8v of standard DDR2 modules anyway, and also lower than the 1.3v that the CPU itself runs at.
 
Scali2 said:
Well, my 965 chipset also runs at pretty low voltages. I think it's 1.025v or something, I'd have to check the BIOS settings, but it's far lower than the 1.8v of standard DDR2 modules anyway, and also lower than the 1.3v that the CPU itself runs at.
Click to expand...
Yes, and Nahalem is also going to be DDR3 exclusive, correct? which is 1.5V memory... So... low voltage all around.
 
Here's some rumor mill for you,
6 and 12core server CPU based on 45nm Shanghai core:
http://techreport.com/discussions.x/14592
Seems like that would pull an insane amount of power... but also considering AMD hasn't gotten even dual and quad cores shipping on the 45nm process... I'm a bit skeptical, best case it's still probably quite a ways off. Still... 12 cores sound impressive. 4 of those in a quad socket motherboard? wow...
 
All I know is that I'm hoping there will be a Dunnington variation that is compatible with Socket 775.
Because then I can upgrade my current system to 6 cores and skip Nehalem/DDR3 for now, and wait for the next tick-phase, or perhaps even more.
If not, I'll just have to get a 45nm quadcore and see how far that takes me.
 
gee scali2 thanks for discussing your intel CPU plans in the AMD CPU section!
 
http://www.hardforum.com/showthread.php?t=1298946

Thanks, I'm here all week.

And scali2, it's obvious you come here to troll and shit all over the AMD forum. Go back to your little intel fanboy corner and leave us AMD guys alone.

What was that little remark about the whole "rumor has it nehalem isn't going to be overclocking friendly"? Something about a low blow to amdzone?

At least you're the type of guy to open your mouth and remove all doubt of you being an idiot.
 
sdlvx said:
And scali2, it's obvious you come here to troll and shit all over the AMD forum. Go back to your little intel fanboy corner and leave us AMD guys alone.

What was that little remark about the whole "rumor has it nehalem isn't going to be overclocking friendly"? Something about a low blow to amdzone?

At least you're the type of guy to open your mouth and remove all doubt of you being an idiot.
Click to expand...

Are you referring to me? I don't recall being involved in any discussion about Nehalem overclocking rumours or AMDZone?
And I'm an AMD/ATi guy aswell, I have various AMD CPUs and Radeons over here. Been using AMD since the Am486, and ATi cards since the 8500.
 
harpoon said:
This whole thread is off topic. WTF is a Nehalem thread doing in the AMD forum anyway?! :rolleyes:
Click to expand...
The OP brought up the upcoming release of Nehalem, and asked if AMD would have anything to counter it by then. Thus the speculation about Nehalem is completely on topic, both for this thread and this subforum, as long as it's discussed in the context of how it will compare to AMD offerings.
sdlvx said:
http://www.hardforum.com/showthread.php?t=1298946

Thanks, I'm here all week.

And scali2, it's obvious you come here to troll and shit all over the AMD forum. Go back to your little intel fanboy corner and leave us AMD guys alone.

What was that little remark about the whole "rumor has it nehalem isn't going to be overclocking friendly"? Something about a low blow to amdzone?

At least you're the type of guy to open your mouth and remove all doubt of you being an idiot.
Click to expand...
Off-topic. You're the one bringing Nehalem overclocking into this thread, and the rest is nothing but insults and flame-bait.
Megalomaniac said:
Nehalem, Deneb - why do i get flashbacks of FreeSpace 2 and other space sims...?
Click to expand...
I'd be hesitant to buy any AMD - or Intel - CPU named Capella.
Menelmarar said:
Here's some rumor mill for you,
6 and 12core server CPU based on 45nm Shanghai core:
http://techreport.com/discussions.x/14592
Seems like that would pull an insane amount of power... but also considering AMD hasn't gotten even dual and quad cores shipping on the 45nm process... I'm a bit skeptical, best case it's still probably quite a ways off. Still... 12 cores sound impressive. 4 of those in a quad socket motherboard? wow...
Click to expand...
That would be an interesting move on AMD's part if they were able to pull it off. These rumors are probably in response to rumors of Intel's own upcoming 6-core CPU.
 
LstOfTheBrunnenG said:
I'd be hesitant to buy any AMD - or Intel - CPU named Capella.
Click to expand...

Shiva? Deimos? Ravana?

/yes, a thread hijacking is in progress :cool:

12 cores and beyond in a few years. AMD is steaming ahead. intel is behind in the technology curve (again) but they seem to be much better in the performance curve.

Amd 1st to have on board memory controllers (though intel might be the first one to have tri level onboard memory controller)
Amd 1st with 64 bit chips

both huge advances in the past few years.

Amd 1st with multi-core chips - FTW!
 
Megalomaniac said:
Amd 1st to have on board memory controllers (though intel might be the first one to have tri level onboard memory controller)
Click to expand...

AMD certainly isn't the first. Onboard memory controllers are quite common on microcontrollers and embedded CPUs, both of which Intel also produces.
AMD was the first for the desktop/x86 market, but nothing more.

Megalomaniac said:
Amd 1st with 64 bit chips
Click to expand...

Certainly not the first. Technically the first 64-bit machines were around in the 1960s.
In the mid-to-late 90s, 64-bit CPUs were already pretty common in big workstations/servers (DEC Alpha, HP PA-RISC, IBM POWER, Sun Sparc etc).
Intel was producing 64-bit chips years before AMD was (Itanium architecture).

Megalomaniac said:
Amd 1st with multi-core chips - FTW!
Click to expand...

Again AMD was not the first, IBM had a dualcore POWER4 in 2001: http://en.wikipedia.org/wiki/POWER4
 
ooook.
For the x86 architecture.. as in, where it matters to us here, within this thread, and within the realms of this PC based world. Not with in the universe.

http://en.wikipedia.org/wiki/X86-64
there is also a note about your itanium there, read up

all of us here run x86, thats what matters here.

ps, unless AMD licensed the technology, then they had to put in some R&D into it.
thats some effort. I applaud AMD for moving ahead like this.
If there are 2 64-bit manufacturers of chips out there, great. If both use completely different approaches, even better. This is innovation, without it we would be using Pentium 5s with the old architecture and the old Intel approach of 1% gain for 2% power increase.
 
Megalomaniac said:
Amd 1st with 64 bit chips
Click to expand...
Intel had Itanium before Athlon 64, and it was a pure 64bit architecture, not the hybrid architecture x86-64 is.

Intel could have done x64 earlier, they were adamant that consumers didn't need 64bit processors yet. And they were right! It's only been within the last 6-8 months that even enthusiasts are moving up to 4GB of RAM and to 64bit variants of Windows. It has taken another 3-4 years since AMD launched the Athlon 64 for even PC enthusiasts to begin switching over to utilizing the 64bit capability.

Megalomaniac said:
Amd 1st with multi-core chips - FTW!
Click to expand...
Wiki says Athlon X2: August 2005.
http://en.wikipedia.org/wiki/Athlon_64_X2
Pentium D: May 2005.
http://en.wikipedia.org/wiki/Pentium_D

Intel first for x86 desktop/consumer CPU's.
 
Megalomaniac said:
ooook.
For the x86 architecture.. as in, where it matters to us here, within this thread, and within the realms of this PC based world. Not with in the universe.
Click to expand...

I just don't see why one should applaud AMD for implementing concepts that have been around 'in the universe' for ages.
I'm more inclined to say "It's about time!".

Megalomaniac said:
all of us here run x86, thats what matters here.
Click to expand...

We wouldn't have to, if it wasn't for AMD and its x86-64. We could have been running Itaniums now, or some other modern architecture. Intel deliberately chose not to extend x86 yet again.
 
Megalomaniac said:
Nehalem, Deneb - why do i get flashbacks of FreeSpace 2 and other space sims...?
Click to expand...

EnB FTW!


but yeah.. to menelmarar, x64 bit processes have been going on for quest a few years, not just after vista / leopard / whatever the linux kernel is was released. just because enthusiasts didnt use it on thier overkill rigs until now, doesnt mean it wasnt used in other places.
 
Status
Not open for further replies.
Back
Top