They also mention at 4.8-5ghz, 3770k burn temperature was very high, 4770k is also very high... I guess maybe Intel did not solder the die to the IHS this time as well...
They also mention at 4.8-5ghz, 3770k burn temperature was very high, 4770k is also very high... I guess maybe Intel did not solder the die to the IHS this time as well...
Cant wait for the Intel Israeli Haifa team with their 14nm Skylake.
Cant wait for the Intel Israeli Haifa team with their 14nm Skylake.
Boring.
At least nvidia and AMD are still delivering new products that are exciting and raise the performance bar significantly year in and out.
Grooveriding, I understand your comment about "BORING", but seriously you have such a cutting edge machine (3930K OC'd to 4.8, Titans in SLI etc) it's going to have to be an enormous jump in performance to impress you. I'm not trying to insult you but really to you I could understand BORING. How about to someone coming from an older Intel dual core or an AMD X series?Boring.
At least nvidia and AMD are still delivering new products that are exciting and raise the performance bar significantly year in and out.
Grooveriding, I understand your comment about "BORING", but seriously you have such a cutting edge machine (3930K OC'd to 4.8, Titans in SLI etc) it's going to have to be an enormous jump in performance to impress you. I'm not trying to insult you but really to you I could understand BORING. How about to someone coming from an older Intel dual core or an AMD X series?
At least Haswell improves on it's predecessor in standard benchmarks. I have a feeling it has some major improvements that have yet to be exploited.
Sandy Bridge was such a huge jump that I doubt we will see such a jump for awhile, just incremental improvement.
I read the other day that intel uses a very poor 22nm process.
They claimed that Intels 22nm process didnt scale very well with voltage. 32nm voltage scaled up with performance & heat at a reasonable curve. The Intel 22nm heat goes off the scale very quickly and performance doesnt increase anywhere near the same way as 32nm. It was somthing to do with the type of process used. They said there are different processes which have shown to be far better for high mhz clocking.
This was actually discovered when they were trying to make certain A9 ARM cores run faster and a few companies hit nearly 3ghz with A9 cores where as the other process type would be limited at 1.7-1.9ghz
Intels 22nm process type is better for lower voltages at the same mhz as 32nm. But the problems come when you try and ramp up clock speed with voltage.
Boring.
At least nvidia and AMD are still delivering new products that are exciting and raise the performance bar significantly year in and out.
Every year GPUs go up massively while CPUs are pathetic. 10% if you're lucky and 'better power consumption'. Pretty boring.
Is anything actually known about Skylake at this point other than it's the tock after Broadwell?
Not comparable. On GPUs you can add more and more shaders every year and we get better performance in games which doesn't work on CPUs. The current Core architecture already is on a high level IPC and high frequency too. Big gains are unrealistic. For Skylake I expect 6 cores for the mainstream. But you need the software, not everything gets a speedbump from more cores.
I would not conclude that Intel's 22n is "very poor".
i7-3770K vs. i7-2600K: Temperature, Voltage, GHz and Power-Consumption Analysis
Pushing a 3.5GHz product to 5GHz is always going to generate extreme conditions for which the process node was not intended to operate.
32nm was developed to enable <= 4GHz clockspeeds, as was 22nm. When you compare in that regime you see 22nm does well in lowering power consumption, operating voltage, and die-size while operating at nearly identical temperatures to 32nm.
All those advantages continue as you push clockspeeds up above the 4GHz envelope, towards 5GHz, but the lead is not maintained (reflecting the fact you are pushing the ICs into a voltage/clockspeed regime for which they were simply not designed to operate).
To say that reflects poorly on the underlying process node itself is silly. The sort of thing someone would say if they didn't really understand how ICs are designed and iteratively optimized before tapeout (and after, with respins) to maximize internal design targets for the project.
To put it into an analogy, consider the core clockspeeds (and overclockability) of Thuban vs Llano vs Piledriver.
If the Thuban to Llano (45nm -> 32nm) were any indication of limitations in the underlying process node then you would have falsely lead yourself to conclude that neither bulldozer nor piledriver would be overclockable to 5GHz on the same 32nm node. But they can, because they were optimized with different goals in mind during design, layout, and respins before the final production stepping was settled on.
That doesn't mean a given node cannot be considered "very poor". It just means you need the right kind of data, information regarding the electrical parameters themselves (length-scale normalized to make them comparable to other nodes). Parameters like NMOS Idrive current per micron at 1V and 90C, or IDDQ (one type of leakage).
And don't forget reliability, having stellar electrical specs but degrading so fast the chip can't reliably function for more than a few months would make for a "very poor" node in comparison to say less spectacular starting parameters but having a rate of degradation that enables the IC to reliably function for 20yrs.
Note:
i7-3770k had DDR3 OCed to 2800+ mhz
i7-4770k had DDR3 OCed to 3000+ mhz
^ haswell can handle faster memory speeds.
CPUmark99 = 10% gain (oooold benchmark).
Spi1M = ~1.3% gain
Spi32M = ~4.5% gain
Cinebench 11.5 (multi-threaded) = ~8.6% gain
The "avg" gain is less than 10% in terms of IPC.
*edit:
That chinese site took the info from a ENGLISH page.
Here:
http://www.futuredrift.com/showthre...-DDR3-3000-era-!-Check-it-out&p=7418#post7418
No need to go through google-translate now
I think the point was more about 22nm should be clocking past 5ghz but it doesnt.
Its not even comparative to 32nm it actually clocks lower when it should clock higher.
Its not poor in the sense it doesnt perform at all but it clearly wasnt better than 32nm when it should have been.
22nm was clearly focused on low power at sub 4ghz speeds. There is a process technology which favours clock speeds over power saving but intel hasnt used this. Its said to be the technology that will be used by TSMC and GF.