10+ years ago, performance was more than doubling every two years through a combination of higher clocks, die shrinks, extra transistors, fundamental breakthroughs in logic circuit designs, etc. Right now, mainstream CPUs are only ~60% faster than mainstream CPUs from four years ago because clocks are stuck near the 4GHz mark, die shrinks are becoming much slower in coming, nearly all fundamental breakthroughs have been discovered and modern hardware is already more powerful than what most people can be bothered with so there is a general lack of demand for significantly faster low-mid-range CPUs to make things worse.

Progress is slowing down and I can only imagine it getting worse in the future.

Maybe he was talking FADD.

In a float addition, you need to denormalize the inputs, do the actual addition and then normalize the output. Three well-defined pipelining steps, each embodying one distinct step of the process.

But as you said yourself, CPUs (and GPUs) generate a lot more heat. They are already challenging enough on their own, imagine how hot the CPU or GPU at the middle of the stack would get with all that extra thermal resistance and heat added above and below it. As it is now, CPU manufacturers already have to inflate their die area just to fit all the micro-BGAs under the die and get the heat out.

Unless you find a way to teleport heat out from the middle and possibly bottom of the stack, stacking high-power chips will not work.

At best, you could stack memory and CPU/GPU for faster, wider and lower-power interconnects.

For most of those issues, the solution is simple: if you forget cables and adapters so often that it is a major hassle, you might want to buy some spare cables and adapters to suit most scenarios. Type-A plugs are not going to disappear overnight (USB 3.0 Type-A maps directly to Type-C so Type-A on PCs, power adapters and anywhere else where shaving cubic millimeters does not matter is not going anywhere) so an A-to-C cable should have you covered in most cases where you cannot do C-to-C... assuming Type-C devices even give up Type-A power adapters.

My guess is the transition will be mostly from A-to-microB to A-to-C. Most people are not going to bother with microB-to-C adapters; they will just get a straight A-to-C cable.

The whole point of Type-C is to address the ugly kludge that is the current micro-USB3 connector that almost no phone or tablet adopted because the connector is huge - over twice as wide as micro-USB.

As for the EU and others with mandated micro-USB charging, I bet they will include Type-C as an acceptable or even preferred alternative in short enough order.

Broadwell-H might be Intel's shipping name but the roadmap name has been Broadwell-K for about a year. That's why you see Broadwell-K used everywhere.

The fact that K-series chips (the enthusiast unlocked chips) will be from the Broadwell-K lineup likely contributed to most computer enthusiast sites choosing to stick with the old roadmap name instead of adopting Intel's new production codenames.

The CPU side might be different but the GPU side remains the same and in GFXBench, the results will likely end up similar, give or take whatever they gain/lose on the CPU.

If Nvidia wanted to go all-out with this Transmetaism, the logical thing to do would be to put together a custom ART runtime that merges with their online recompiler/optimizer.

Looking at Shield Tab reviews, the K1 certainly appears to have the processing power but actually putting it to use takes a heavy toll on the battery with the SoC alone drawing over 6W under full-load: in Anandtech's review, battery life drops from 4.3h to 2.2h when they disable the 30fps cap in GFXBench.

The K1's processing power looks nice in theory but once combined with its power cost, it does not sound that good anymore.

But the comment I was replying to was about Broadwell-K which is the desktop variant. Shaving a few watts on a desktop CPU is not going to get you much battery life even if you have an UPS. Most people who will buy Broadwell-K will be using it with a discrete GPU too.

While Iris Pro performs quite well when you turn down graphics low enough to fit most of the resources in the 128MB Crystalwell L4 cache, nobody interested in mid-range graphics would be willing to give up this much quality for decent frame rates. Once you exceed that 128MB, even low-end discrete GPUs with GDDR5 take the lead. Broadwell's four extra units are not going to change this by much.

If Intel released chips with an upgraded 512MB Crystalwell and twice the L4 bandwidth, then that would nuke low-end GPUs and possibly start hurting mid-range.

The P4 was getting destroyed by AMD in benchmarks, the 65nm die shrink failed to translate into significant clock gains and interest in power-efficient desktop CPUs was starting to soar so Intel had little choice but to execute their backup plan to save face: bring their newer and better-performing next-gen Core2 mobile CPU design to the desktop.

Broadwell only brings minor performance improvements to desktops and shaves a few watts along the way. If Intel decided to scrap Broadwell-K, or perhaps produce them in limited quantities due to launch dates getting too close to Skylake for full-scale production, few tears will be shed.

Since Broadwell-K is not going to launch until half-way through 2015 and Skylake was still on the 2015 roadmap last time I remember seeing one, I would not be surprised if Intel canned Broadwell-K altogether - no point in flooding the market with parts that only have a few months of marketable life in front of them. If Broadwell-K does launch beyond OEMs, it may end up being one of Intel's shortest-lived retail CPUs ever.

In the first Broadwell roadmaps, there were no plans for socketed desktop parts; all mobile and embedded.

Many analog scopes had many more trigger options than that.

But with modern low-end scopes like Rigol's DS1xxxZ-series featuring relatively deep memory, 20k waveforms per second trigger rates, intensity grading, up to 1GSPS sampling rate (single channel), relatively easy hacks to enable all the options, segmented memory to record events, pass/fail mask, etc., the 10-20 second startup time on an instrument most people will usually use for hours at a time is well worth it.

Nowhere near as bad as Agilent's Windows-based bench multimeters that take nearly two minutes to boot... but even that is fine since they need ~10 minutes of warm-up time to fully stabilize before you can get the full 6.5-digits precision.

Boot time in lab instruments is a silly thing to worry/bitch about when most instruments have long warm-up times and should ideally be powered up 10-30 minutes before use anyway.

L3 cannot deliver the data to Verizon since there is not enough connectivity between L3 and Verizon to hand the data over at the interfaces where L3 is attempting to do so.

Verizon does not want to put all their bandwidth eggs in L3's basket just to accommodate Netflix so they want Netflix to either peer directly or force L3 and its other CDNs to re-route traffic through other Verizon peers.

Depending too heavily on a single upstream provider is not sound business practice and Verizon wants to avoid getting tied up in that sort of relationship with L3 mostly due to Netflix.

It would also add ~2mm to thickness and 10-20 grams for the sliding mechanism, the keyboard, stiffening structures and bottom cover.

And there is the sliding mechanism as an additional mechanical and electrical point of failure.

I prefer physical keyboards over on-screen as far as typing goes but the design and cost compromises, not so much.