Intel's 6-core Dunnington CPU coming this year, Nehalem gets official
Quad-core shmod-core Intel, we need 6 cores or more to keep our uh, web browsers snappy. While you're at it, how about tossing in some Simultaneous Multithreading (SMT) so that each core can process two threads at a time -- 16 simultaneous threads per 8-core processor or 32 for dual-processor, 8-core rigs. If that sounds good then you're in luck; Intel just went official with its near-term architecture plans which include the 2008 launch of a 6-core Dunnington-class server CPU platform based on Intel's 45-nm Penryn "tick" architecture. On deck is Intel's second generation Nehalem "tock" architecture with SMT and scalable from 2- to 8-cores. We're talking "dramatic" performance and energy improvements, according to Intel, from a microarchitecture bent on delivering an 8 MB level-3 cache, DDR3-800 memory support, 25.6GB per second Quickpath interconnects (so long Front Side Bus!), an integrated memory controller and optional integrated graphics to high-end servers and eventually laptops. Hear that AMD? Tick, tock goes the clock.
P.S. That's Nehalem pictured. What, can't you tell?
[Via BetaNews, thanks Mike O.]
P.S. That's Nehalem pictured. What, can't you tell?
[Via BetaNews, thanks Mike O.]



















I'm gonna wait till they bring out the OVER 9000!!! core chips before upgrading from mi G3. lolz
wow nice, AMD is seriously in trouble. they are nowhere near 8 cores, aren't they?
Actually AMD is the only one currently with native quadcores on the market whereas Intel will have natives with the nehalem architecture. Both will need to glue them together for an octo core.
Sure, AMD does have native quad four already but only on 65-nm, right? I believe 45-nm is impressive because its virtually impossible to do using SiO2... so Intel have changed the process to use Hafnium instead.
AMD is actually very close to 6 and 8 cores. Their architecture is robust even if it isn't as efficient clock-for-clock as Intel's current stuff. AMD needs 45nm manufacturing to shrink the chips to a point where adding more cores makes sense.
Robust architecture is nice and all, but they're not delivering.
Intel applies their tick-tock approach to core count increases too. Tick and they take two single-core dies and glue them together. Tock and they properly mate them in a true dual-core solution on a single die. Tick and they glue two dual-core dies together. Tock and they make a proper quad-core solution.
AMD's approach is to only ever put out a "proper" solution. This has advantages and disadvantages. The big disadvantage is cost. They're building one single die that is double the size. This double their failure rate; they're twice as likely to produce a bad die because there is twice as much surface area. The advantage, however, is that it's a more efficient solution, using less power (and producing less heat).
Intel's approach has the advantage of cost. The dual-core dies are produced separately and working ones are glued together. Their yields are the same as before, and they waste less silicon. It's especially easier to produce higher performing parts, something AMD has had serious issues with when it comes to the Phenom (and their only-true-multicore approach is probably part of the problem). So Intel can push their parts faster (and cheaper) due to better yields. They can also get their parts out far sooner, as much less R&D is required for the initial product. The disadvantage is that they're less efficient, producing more heat and drawing more power.
Anybody who simply dismisses Intel's approach just because they're gluing two dies together obviously hasn't evaluated the advantages and disadvantages of each approach. I, for one, prefer Intel's approach; at the moment, it's producing faster cheaper chips sooner at the expense of heat/power, which eventually gets corrected.
And Intel widens the gap.
And Intel widens AMD's asshole. C'mon AMD let's see some competition!
I want one!
so as everybody else here, just my guess
I want many things but a 6-core CPU is not one of them.
@Paris, I can understand not saying that you "need" something but what do you currently have that would make you not even "want" a 6-core CPU rig?
@Jonathon
I would have no use for a 6 core CPU therefore I don't want one. (If someone offered to give one free to me I would only want it to sell it to someone else).
Whatever I'm still rocking my Socket A Athlon!...
...I wan't one! (or two)
Socket?
I'm using an Athlon 650. (that's SLOT A for all you young uns out there)
This is AMD calling .... WE NEED CASH FAST!
Google, Microsoft, ... All are welcome!!!
"Quad-core shmod-core Intel, we need 6 cores or more to keep our uh, web browsers snappy. While you're at it, how about tossing in some Simultaneous Multithreading (SMT) so that each core can process two threads at a time -- 16 simultaneous threads per 8-core processor or 32 for dual-processor, 8-core rigs. If that sounds good then you're in luck; Intel just went official with its near-term architecture plans which include the 2008 launch of a 6-core Dunnington-class server CPU platform based on Intel's 45-nm Penryn "tick" architecture. On deck is Intel's second generation Nehalem "tock" architecture with SMT and scalable from 2- to 8-cores. We're talking "dramatic" performance and energy improvements, according to Intel, from a microarchitecture bent on delivering an 8 MB level-3 cache, DDR3-800 memory support, 25.6GB per second Quickpath interconnects (so long Front Side Bus!), an integrated memory controller and optional integrated graphics to high-end servers and eventually laptops. Hear that AMD? Tick, tock goes the clock."
... Huh?
No more front side bus?
Please expand :)
They're replacing it with something analogous with Hyperthreading on AMD's.
Can't remember what it's called, but it's essentially the same thing. A really ,really fast bus for everything.
ahh.
sweet. :)
there's no front side bus in many of the new CPU's and probably won't ever be. Busses take up too much space within the CPU and now that we can afford it, having a direction connection to each component within the CPU allows for much higher performance CPU's (as it reduces the load on each wire). Therefore each component (i.e. the ALU) has the minimum number of connections it needs and this takes away the need to have a bus connecting everything together.
It's not hyperthreading, its hypertransport. They are replacing the front side bus with a much wider bus for communication between the cores (if they still decide to the that whole cpu talking over the bus instead of communicating directly) and for RAM access.
Oh shit, hyperthreaded multicore processors?
And I was all about to buy a Core2Quad.
Wait till the hyperthreaded quad core processors come out.
Remember to buy a bib while you're at it as well. Drooling is bad.
WEEEEEEEEEEE stuff cores insideeeeeeeeee
keep stuffinggg
till the end of year we will have about 16 core CPU :D
May i just ask... HOW necessary is this????!!?!?!?!?!?!?
Very!
Apart from the speed gains there are also efficiency gains and a reduction in power usage. http://www.reghardware.co.uk/2008/03/18/intel_sse_4_text_tweaks/
obviously, you have never built a flux capacitor.
He/she has obviously never done any CGI or folding/number crunching.
@ producerism
That is the funniest thing i have read today. thanks for the chuckle
it plays crysis at max setting with no videocard /sarcasm
I love these multicore chips. I have a Core 2 and watching the CPU usage gives a real insight into how they work. Something like ray tracing in Catia uses 100%, each core renders a different section of the screen and it halves the time taken. However, dynamic modelling in ADAMS uses only 1 core so the CPU never goes above 50% and there is no advantage except you can keep iTunes running with no penalty for your actual work!
So now it's the turn of the programmers to improve algorithms and make use of what is available. Quite a turnaround since it always used to be the chip manufactures who were playing catch up.
Some types of processing are serial, and just won't benefit much from parallelization as much as others. that's a limitation we can't get around easily without bumping clock speed.
hickory dickory dock, your wife went on my ...
this is going to be great for my virtualzation project next year...were talking some serious amounts of virtual machines to one piece of hardware....
"P.S. That's Nehalem pictured. What, can't you tell?"
Yes, and the cores seem to be color-coded :)
I can't wait to get one of these cranking out folding@home units!
Anyone else notice that along with QPI, Nehalem is eating up the memory controller? What will become of our GMCH?? Will it just be a dedicated PCIe hub?
People need to realize that our current software can barely utilize 2 cores, let alone 6. The real bottleneck of performance is the ability for code to utilize parallel processors/cores.
Until we can significantly change our programming mindset away from a serial mindset, or companies like RapidMind are able to save us, then these processors will be severely under-utilized. (http://csclub.uwaterloo.ca/media/Riding%20The%20Multi-core%20Revolution.html)
I agree, but for me it would be great. 3DS Max rendering can use all of them, I have 4 cores at the moment, but seeing 32 render buckets on 1 frame would probably make me faint with excitement. not to mention that one machine like this would double the amount of cores on my render farm. :-D
Fair enough... until I look at my work CPU running, three different multi-threaded apps at once, Outlook, iTunes, virus scanner, work required hours tracking software, and Firefox...
Even if you run all single-threaded apps, most people do plenty of multi-tasking to take up the extra cores' provided powers. I'm a Graphic Designer and while most CS3 apps have multithreaded code, things like Photoshop are still single-processor-centric, BUT, most of the time I'll be running Photoshop, Illustrator and InDesign, or PS, Ill, Flash, Fireworks and Dreamweaver... So even if each only can use one of the cores, I'm still able to more-efficiently manage my workflow.
Gimme 32 cores and all your RAM, I'd be plenty happy. Right now.
Somebody tell Anon it isn't 2001 anymore. Ever heard of multitasking? Or are you one of those people that start a task, and sit there and watch/wait while it works? Of course you are, you're still on single core!
My video conversion/editing applications sure like the 4 cores. And as pointed out, you CAN run more than one app at a time. Try it, it'll work even with a single core - albeit it slowly.
They should have skipped 4 core and went straight to 6 or 8.
And of course, today is also the day all the parts for my quad core, 8 GB ram, 2.3 TB hdd, 8800GT workstation beast machine came in :[
Damn you Intel!!!
Why damn Intel? Quad core and 8GB memory.... Dude, (for now) thing uses more than 4GB of ram unless you run a server!
Also, wanting always the latest things is useless, you'd had to upgrade every 2 weeks.
Couldn't run a holodeck without one... or several... thousand.
CSI, Common System Interconnect is what is replacing the Front Side Bus
Someone pointed out that this is useful for CGI or number crunching and protein folding. How often are normal users going to use those apps?
There are not even quad threaded applications yet, sans Flight Simulator 10, and the usefulness of quad threading for games is not fully explored yet. And they want to push out 6 core processors?
This is just a way to get consumers to waste their money. I hope at least retailers are kind and tell consumers the truth but I doubt it.
Also, adding more cores actually makes each individual core less efficient and less capable. Right now, thats not really an issue, but the effect builds as you add more cores, so eventually it could get to the point where adding more cores actually lowers the performance. Not to mention, this is just drawing more and more power....
How would Intel name this new line of processors???
Lets see: Core 2 Duo, Core 2 Quad, Core 2 Se..
I am not sure if AMD is really even going after Intel right now. With the blind side release of the ATI HD 3850 completely taking over the sub $200 GPU market with athority, and the near silent release of the 780G intergrated GPU, AMD seems to be spending its days plotting the death of Nvidia. Granted that ATI is still its own company, AMD seems to be hell bent on using GPU's to take near complete control over the low and mid-range markets. Even the new Nvidia 9600 GT can't kill the HD 3850 in price/performance, leaving the HD 3850 as the still number one choice for mid-range dedicated GPU power. The 780G is just plan evil. Once a better cooling solution is finished for laptop use, the 780G will virtually be the only choice for intergrated GPU's. When it takes a dedicated 8600M GT GPU to out perform an intergrate SOLUTION (I mean that one), that also boasts Blu-ray/HD DVD capabilities, over clocking to 1.1 Ghz, a 4 digit score in 3D-benchmark06, and a price point below $100 in motherboards with those chips, you are in trouble.
To me it seems AMD is more focused on the GPU market then trying to compete with Intel, at least during this Quarter. If AMD follows the same plan as the "ATI department" we should see something completely out of the blue from them in the next 3 months that no one knew about until a week before the release date.
wow "so long front side bus" yes! finally we can get rid of that performance crippling fsb!
Honestly, this sounds like a great development. I used both the current Intel Quad core and AMD Phenom CPU's, and Intel is beating the AMD in almost everything. I am disapointed in the AMD's performance, guess it's due to the huge difference in cache that Intel is offering on their product. Once the front side bus is gone AMD will go MAD because the only advantage that they are offering now to gamers is the higher front side bus...
My favorite is all the "There aren't any multithreaded apps to take advantage of it" people. I live in a large city, and would not want you planning my next freeway project. One lane freeways everywhere right? There simply aren't enough cars to take advantage of anything more than one lane right?
The reality is, whether the code is there to take advantage of it right now or not, the old saying still holds true. If you build it they will come. Its really just common sense if you apply some logical thought to it. While developers have been slow to catch on they certainly won't write multithreaded code for non-existent multi-core processors.
I have a q6600 at home and its a nice feeling to finally have a machine that I can't slow down no matter what I throw at it. Like many have stated, multithreaded apps are preferred, but until they are more mainstream I'm very content with encoding video, surfing the web, toggling in and out of Supreme Commander, and using Photoshop simultaneously without my system slowing down one bit (no pun intended).