"What Nehalem is really about"
- Thread starter fellix
- Start date
Saw this at aces:
http://intel.wingateweb.com/US08/published/sessions/TCHS001/SF08_TCHS001_100t.pdf
Nehalem will be able to go a step beyond clock gating and almost completely shut cores off from the power plane.
In other news, it switches to 8T SRAM. More impressive, earlier comparisons of cache density between AMD and Intel's L3s showed that while AMD still lagged, it wasn't lagging nearly as much as AMD lagged for the L2.
Apparently, that was because Intel threw in an extra 33% in transistor count for those cache arrays.
That means Intel's cache will be more reliable, yield better, and operate at a lower voltage.
AMD's SOI cache cells will remain larger than they need to be and will run at higher voltages.
On a power note, Nehalem switched back to static logic, even though it is not as fast as the domino logic or LVS used in other x86s.
With Intel's engineering resources and the power wall, losing the benefits of the faster logic was apparently compensated for.
Might be not so good news for overclockers, though.
http://intel.wingateweb.com/US08/published/sessions/TCHS001/SF08_TCHS001_100t.pdf
Nehalem will be able to go a step beyond clock gating and almost completely shut cores off from the power plane.
In other news, it switches to 8T SRAM. More impressive, earlier comparisons of cache density between AMD and Intel's L3s showed that while AMD still lagged, it wasn't lagging nearly as much as AMD lagged for the L2.
Apparently, that was because Intel threw in an extra 33% in transistor count for those cache arrays.
That means Intel's cache will be more reliable, yield better, and operate at a lower voltage.
AMD's SOI cache cells will remain larger than they need to be and will run at higher voltages.
On a power note, Nehalem switched back to static logic, even though it is not as fast as the domino logic or LVS used in other x86s.
With Intel's engineering resources and the power wall, losing the benefits of the faster logic was apparently compensated for.
Might be not so good news for overclockers, though.
Nehalem is and always has been about modularity, multi-threaded performance enhancements, and bringing a new platform infrastructure. Single-thread performance was never its focus.
IOW: DUH, Anand...
IOW: DUH, Anand...
The author isn't Anand, it's Johan DeGelas.
I am not too worried about the overclocking bit. But they seriously gotta release CPUs that are out of the box @ 4ghz and beyond. For some reason i am convinced that is a possibility with Nehalem if not Sandy Bridge which if I am not mistaken is going to be 32 nm? Have you noticed that all the CPUs released are just in 2-3 ghz and some at 3.2 ghz range and thats it?
It may very well be a lack of knowledge on my part but if these chips can be overclocked to 4 ghz without much tweakge, why does not Intel just release a 4 ghz part?! I think 4 ghz has such a nice ring to it.
It may very well be a lack of knowledge on my part but if these chips can be overclocked to 4 ghz without much tweakge, why does not Intel just release a 4 ghz part?! I think 4 ghz has such a nice ring to it.
The wattage numbers explode past a certain point, because of the higher clock and the voltage bumps to get the higher clock. In general, a given chip at stock voltages with a retail box cooler is not going to hit 4 GHz.
Reliability is also not too high on overclockers' lists.
Even a low rate instability or silent data corruption would not meet the standards of a mass-market product, and Intel already had one bad experience with the 1.13 GHz PIII.
Reliability is also not too high on overclockers' lists.
Even a low rate instability or silent data corruption would not meet the standards of a mass-market product, and Intel already had one bad experience with the 1.13 GHz PIII.
I guess going over to AT has caused Johan to dumb down his writing quite a bit. I remember his various Xeon and Opteron platform explorations in the early days, apparently it wasn't popular enough to continue. That's too bad.
I've stopped reading AT since their R700 review. I can't stand Derek Wilson's constant pot-shots at ATi, nor Anand's pandering to Intel. I just naturally assumed any "der" comments would come from Anand or his crony Derek Wilson (although he only writes about graphics so I assumed he had nothing to do with this piece).
Because there is no pressure on Intel to do so. Welcome back to the Pentium 2 and early Pentium 3 days (up until the release of the Athlon). Thanks for the 1 or 2 clockspeed bumps per year, Intel. Real innovation there.
Although I will admit that pricing is much better nowadays. Which is what the Extreme Edition line is for
True. The amount of chips that can run truly reliably @ 4GHz on air cooling without serious volts (and subsequently serious heat) is pretty small, I'd imagine.
Spaceman-Spiff
Regular
They ditched the whole SSEx base, already?
Maybe cause there's too many now to keep track: SSE, SSE2, SSE3, SSSE3, SSE4, SSE4.1 (Intel), SSE4a (AMD), SSE4.2 (Intel), SSE5 (AMD), AVX, 3DNow!
Just read Anand's article and it says only the L1 and L2 uses 8T SRAM
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3382&p=10
Much of the blame lies at AMD's feet for not producing a competitive archtecture though. You can't really blame a company for milking its market when that option is available (thats not to say i'm happy with the situation - i'm seriously routing for AMD here!)
Intel. Real innovation there.
Yeah until they manage to kill off AMD (which is exactly what the low pricing is aimed at doing). Then we can expect the $200 CPU of today to cost $500.
In fact Nehalem would seem to be a pretty decent opportunity for AMD. Gaming performance seems to drive the desktop market (A64 only seriously beat P4 when it came to gaming) and if Nehalem isn't going to improve it then AMD has a chance to catch up. OK they are going to lose the server market but thats pretty much unavoidable at this point IMO.
Concentrate on winning back the heart if the consumer. At the end of the day, even IT directors are consumers in their spare time
Concentrate on winning back the heart if the consumer. At the end of the day, even IT directors are consumers in their spare time
Yeah, sounds good but you know with chip design you can't just change everything on the fly. And I'm afraid Shanghai/Deneb won't also deliver better gaming/single threaded performance as it builds upon Barcelona.. Though if they somehow manage to clock beyound 3.2 (whitch seams extremly unlikely to me), it will be something.
And I don't know what's Bouldozer aimed at, but if it's not at consumers it will still miss this would-be marketshare regaining oprortunity.
True, but Intel is ultimately responsible for their own actions.
Screwing over the customer (or milking them, if you prefer) is never the right answer.
Intel needs AMD to avoid being investigated and charged under anti-trust laws. If AMD goes out of business, Intel as we know it will cease to exist (meaning it will be split up into separate entities).
Then we can expect the $200 CPU of today to cost $500.
Indeed. It is only AMD's desperate need to survive and the subsequent bargain-basement pricing structure which has forced Intel to compete on price.
Catch up as in achieve performance parity or catch up as in split overall marketshare of x86 MPUs with Intel? I find the likeliness of the former to be pretty slim, at least until BullDozer. As for the latter, I would be absolutely shocked if AMD ever plays anything but second fiddle.
They already have lost the server market. They still have a lot going for them here, though. B3 rev. Barcy's aren't half-bad, particularly the energy efficient models. Oh, and AMD still wears the FP performance crown, and may even still when Nehalem launches. Granted, integer and SSE performance are much more important in just about every segment of computing outside of HPC, though.
Concentrate on winning back the heart if the consumer. At the end of the day, even IT directors are consumers in their spare time
Absolutely. This is what brought them to prominence in the first place. No reason to think it wouldn't work again. Trouble is, they need another K7 or K8 (i.e. back-handed slap to Intel) in order to pull that off again. I don't know that AMD has another K7 or K8 in them. I only hope I'm wrong, however.
Similar threads
