My last round of upgrades for the wife's and my desktops was entirely motivated by upgrading to lower power parts that provided the same (or better in the case of video cards) performance.

The office became noticeably cooler, it was a good upgrade.

This is entirely the reason I chose Intel and Nvidia over AMD/AMD. I run linux, and the open-source nature of AMD appeals to me, but I was so sick of the heat and the fan noise. Plus my old computer only ran at it's best during the winter.

It costs around 500$ to get similar performance. So ~500$ / (0.20$/kwh * 0.102kw * 2/3) that's ~37,000 hours or 24/7 for 4 years.

So, no lower power is not a solid reason to upgrade.

On a desktop. I know you don't care, I said so. But Kaby Lake is a silicon product for a bunch of markets (of which "desktop" is a rapidly shrinking one), and in those markets it represents solid improvement over its predecessors.

I don't think the desktop is shrinking as much as sales suggest. I would happily drop 2k on a new desktop if there where any reason to do so. Intel chose to waste 5 years of performance gain on a low end graphics capability's instead of bumping cheap multi core performance.

That use of "cheap" is doing too much work. Obviously Intel hasn't squandered many-core scalability either, c.f. Broadwell-E which is available with as many as 10 cores and targetted at exactly your market.

It's just not cheap, being a 240mm2 die (almost twice what Kaby Lake is, with attendant yield penalties) targetted at a niche market with low volume.

So your point isn't that Intel isn't innovating, it's that you want them to do it in a way that violates straightforward economics and hand it to you for less than it takes to produce.

Basically: you got spoiled by the early days of the VLSI revolution. Silicon scaling isn't doing what you expect any more.

At Intel's scale having 4 core chips with integrated graphics or 6 core chips without integrated graphics and slightly smaller transistor count is not a significant cost, meaning they could both sell for similar prices.

They don't do this because they have zero real completion and can milk the high end market. Further, by holding back performance gains they can extend the upgrade cycle as long as possible.

PS: Cellphones are approaching Desktop chip performance.

> At Intel's scale having 4 core chips with integrated graphics or 6 core chips without integrated graphics and slightly smaller transistor count is not a significant cost, meaning they could both sell for similar prices.

Yeah, that's not how silicon manufacturing works. Tooling up for a new part of this size and ramping production gets into the tens to hundreds of millions of dollars. No one does niche-market (and yes, gaming desktops are niche) parts at the prices you expect.

Low-volume big parts cost a thousand dollars or more, that's no different at Intel than it is anywhere else.

It's going to be ~1:1 with the number of discrete nvidia + AMD graphics cards sold which is not exactly niche.

Further, Intel already has a range of chips in production making that 100's of millions per chip vastly overblown. You can measure these numbers in several ways, but ramping up 100 million of chip X or 80 million of chip X and 20 million of chip version Y is nowhere near 200+ million.

Ed: For clarity the cost delta is not 200+ million, the price is vastly more than 200+ million to make 100 million of chip X.

While I certainly wouldn't turn away multi-core performance improvements, the onus isn't entirely on Intel. A large portion of the software market still makes no or very poor use of multiple cores. Cranking that performance up for many apps that won't use it really isn't a selling point to the general market. It would be a boon for specialized workflows, certainly not denying that, but it's not nearly as attention grabbing or experience improving as solid integrated graphics are. Just a few years ago integrated graphics were garbage, and now they're reasonably formidable for most tasks, even some light gaming.

Why, about any resource intensive line of application out there does use all of CPU & GPU gains you throw at it. Adobe Premiere does, Autodesk Revit, Inventor etc.

Just finished a CAD workstation build for my wife with state of the art components. The performance gain over a couple years old "pro laptop" she used before is enormous. And not just on raytraced views and renders, the whole UX is much much smoother.

Is desktop a shrinking market or have we reach peak saturation? Desktops are very affordable (relatively speaking). The people who want one will get them. Processors are some of the least upgraded [citation needed] components of a desktop system, because there isn't a large incentive to get one. It's rarely the bottleneck of the system. As someone else has said, it's a mature market.

I recently upgraded from an '09 processor because my motherboard died. Trying to find a new one to fit the old chipset was rough, so I got a processor as well. I didn't need to get a new processor, because it wasn't the bottleneck. There are many instances where the new one will give me better results, but it wasn't so bad that I was hurting. Unless you have a specific need, upgrading for the sake of upgrading is largely wasted money.

Lower power means less demand for quickly spinning fans. Fan noise is quite a huge purchase decision factor for some people looking to buy a home pc (like me, I'm one of those people).

On a desktop you will get much farther with bigger fans than with low power chips. That is, unless you want to be fanless, that is a expensive extreme.

> that's ~37,000 hours or 24/7 for 4 years.

That's 4 years running the CPU at full tilt. The 2600K will idle at a lot less than 102W

102W is a /lot/ of power and heat...

> 102W is a /lot/ of power

Yeah, and OP was saying that was only the socket consumption, you've also got to add in chipset, memory, discrete GPU, hard drives, NIC, sound, etc... and power supply overhead as well.

I bet that system was pulling 400-500W total under the load he described.

A single old style light-bulb is not what most people think of as a lot of power and heat. https://www.1000bulbs.com/category/100-watt-standard-shape-l...

Further the delta is only like 60w under full load.

If they touched the light bulb, they might think otherwise.

> Desktop applications tend to be either not performance limited or limited by something other than the CPU.

The browser is a desktop application and, once the page is loaded off the network, is certainly CPU bound.

Hmm... I still find browsers more memory blind than anything, any machine with less than 16GB of RAM I find really struggles with JavaScript heavy sites, god damn JavaScript.

Your JS heaps are not approaching 16GB unless you have thousands of tabs loaded. You can verify this in about:memory.

That would be more relevant id the choice was 15GB vs 16GB. However it's normally 8 vs 16 and that 8 is used by many processes. So, it can easily be comparing 0.5 of free memory vs 8.5 GB of free memory.

_something_ about the browser (either chrome or firefox, fresh profile, no extensions) is eating ungodly amounts of ram however, and it tends to be at its worst with JS heavy sites.

EDIT: no addons either

Maybe you have some random addons(not extensions) installed?

It's probably a combination of all the caching browsers do to improve observed performance and sloppy memory management in the browser software, not javascript.

I wouldn't care unless I lived somewhere warmer. Energy can not disappear from a closed system and that inefficiency will keep you feet toasty. :)

I care more about fan noise than I do about heat.

Yep, efficiency is key here. That is where most improvement has done in terms of cpu design in recent years.