Intel no real CPU progress in last 4 years?

You oversimplify. Looking at raw clockspeed is not the only measurement. The architecture changed to allow more efficient processing of instructions and a larger range of instructions, allowing for higher throughput. In general, you can take that each release since Ivy Bridge added 7-15% improvement in similar benchmarks. At the same time, Intel shrink the litography to allow for more transistors (the base measurement of integrated circuits) - thus allowing for more cores and more cache on roughly the same die size, again increasing performance.

They achieved this while either reducing power sharply, or keeping it in line with previous generations.

That is just if you look at the processors similar to yours - the i5 range. In the higher bracket, there are also more cores, higher cache and also more integrated PCI lanes.

At the lower-end, there are now processors that consume as little as 4 - 7.5 Watt. THis area has probably received more attention than the mid-range and high-end desktops, since that is where the market is most active and competitive. At higher-end ranges, Moore's law is presenting difficulties and it seems to be getting really hard to shrink lithography below 7 nano-meter. So expect that at best clock speeds will stay around where they are, with only marginal upticks until we discover an entirely new way to etch silicon.

Intel hasn't been stagnating. Though the desktop enthusiast parts haven't really changed much in the last several generations they have managed a small increase in IPC each generation. But they have also been adding instructions sets and they have been getting more efficient. Performance per watt has seen a steady increase along with the IPC. I7-4790k is actually a little unique as it is clocked and voltaged higher then the other parts from this generation. 84W being the common value for Haswell chips. 3rd gen saw the lowest TDP at 77W. They have been hovering around there for a while now.

Each generation also, more or less, sees the CPU's themselves getting smaller and the on-die GPU getting larger and faster. It carries over the same performance/watt advantages. In the last several generations the onboard GPUs have about tripled in performance.

Where you see these improvements making a difference is in mobile devices, laptops and ultrabooks. That is the primary market and the underlying design for the desktop processors, so sadly we are not the target audience.

The PCI Express lanes in your Sandy Bridge i7-2600K are PCI Express Revision 2.0.

Intel implemented PCI Express Revision 3.0 in Ivy Bridge and later processors. That's double the PCIe data rate that is supported by your i7-2600K.

The clock speed wars are over because AMD isn't competing anymore.

For games, things havent moved on from fast quad cores, but other things absolutely can take advantage of the extra speed.

Also, you mention the 8 core 5960x being too expensive for the performance, but look at the 5820k, 6 cores and 12 threads with very high IPC for $380 is a good deal if you need the performance.

In reality your Athlon X2 probably didnt double your speed, back then very little was multithreaded, but I do see your point.

Like me, you timed it right buying Sandybridge which was a real high point for Intel. That said, from personal experience with the modern Intel high end 6/8/10 core cpus, they do bring significant performance benefits in multithreaded use.

Modern CPUs are getting more efficient with the allotted performance. IO and power savings have been two big improvements. CPU performance itself has no improved a lot for most work loads, but more advanced instructions, like AVX, are slowly getting traction.

I would think the next big thing will be memory/cache embed directly on the CPU, along with 3D stacked logic.

Intel's been making steady progress even if it doesn't seem mind blowing. They also haven't had much in the way of competition the past 4yrs either. If you're already ahead of your competition there's no incentive to widen the gap even further. Cpu's are also at a point where they greatly outpace the rest of the system in terms of performance. The cpu's always been the fastest component, if ram, data lanes for interfaces like pcie, sata etc, the storage whether hdd or ssd are multitudes slower then making the cpu twice as fast won't make the pc twice as fast. We were already seeing that until ssd's hit the market, whether a cpu runs at 2ghz, 4ghz or even 5ghz - if the data storage bandwidth feeding the ram which then feeds the cpu is the bottleneck it won't matter much. It makes cpu speed improvements seem less fantastic.

This can be seen in things like gaming where if the system is stuck using an r7 250, it matters little in terms of fps at higher resolutions. You can't magically keep gaining substantial fps increases by just making the cpu faster, everything works together as a system held up by the slowest part in the chain (which can vary from task to task). Balance always makes for a more efficient smooth machine than pairing extremely fast parts with extremely slow ones. In the case of burning a disc like a cd or dvd, it doesn't matter if you have an i5 and a 7200rpm hdd or an i7 and a top of the line ssd or even raided ssd's. The burner is still limited to xyz burning speed and as the slowest part will burn discs at the same rate regardless. The same holds true for other components. Moving from 4200rpm to 5400 rpm to 7200rpm hard drives eventually storage stagnated. The best anyone could do was run raided 10k rpm drives like the raptors or expensive 15k rpm scsi drives. Ssd have changed that some for sustained data throughput but even so they've only made incremental speed improvements since their release. Many are capable of around 500 (give or take) mbps read/write. They didn't improve to the point of 800mbps read/write or 1000mbps read/write. The incremental speed improvements on storage (which feeds the ram and in turn cpu) from early to current ssd is more along the lines of rpm improvements seen in hdd's.

> Intel has no competition so no incentives
Well... I have no incentive to upgrade and buy their stuff. I would think this would be an incentive for Intel.

> performance by watt has increased
Really?
i7-860: TDP: 95W (45nm)
i7-6700K: TDP: 95W (14nm) seriously where is the improvement?

i7-860 vs i7-4790: http://cpuboss.com/cpus/Intel-Core-i7-860-vs-Intel-Core-i7-4790
single core perf: 7.5 vs 9.5
about EXACTLY explained by the jump from 2.8Ghz to 3.6Ghz.

For all I (and people using a graphic card) know they are shipping the exact same part at a slightly improved frequency.

Oh and if shrinking allows to add a lot of transistors, and if you are not redesigning the entire core, then at least you can add some cache memory to speed ios, right?
Nope:

i7-860
-------
L2 cache 4 × 256 KB
L3 cache 8 MB

i7-4790
-------
L2 cache 4 × 256 KB
L3 cache 8 MB

0 progress in 5 years.

As a comparison, I would say the performance in the previous 5 years improved probably by a factor 10.

Bit of an old post.

Maintaining a wattage envelope while increasing performance is the same as improved performance per watt. If you were to benchmark your chosen comparison choices the improvement would be obvious. Keep in mind that newer CPU is a lot smaller then the one you are comparing it to. So dissipating the same wattage in smaller area is actually a fairly impressive feat.

They have chips that have increased levels of cache, along with more cores. When you are dealing with only 4 cores it doesn't make sense to increase resources, expensive in real-estate on the wafers, that would go un-used. They would have to make architectural changes to accommodate a use for the extra memory.

CPU Boss and GPU Boss aren't exactly a good metric to use. Something that is more quantifiable is a lot better, like PassMark which shows about a doubling in speed between Lynnfield and Haswell.

I don't mean to diminish the technical feat of shrinking to 14nm. The same processor would be 10 times smaller at 14nm than at 45nm.
But the point is: what does it do for me?
- Shrinking, by itself, doesn't help me.
- Dissipating heat, by itself, doesn't help me.
- Adding a GPU in the processor (which I guess absorbed most of the progress) doesn't help me as I have a graphic card.
- Reducing energy consumption modestly from 95W on my desktop doesn't help me,
- Increasing frequency helps a bit.

What we see is first: Intel has abandoned desktops to focus on laptops.
And then we see processors progress running into walls: frequency, heat, core design, and soon further die shrinkage will hit a wall too.
A doubling of speed in 5 years is a rather mediocre achievement by comparison to the previous 5 years.

They blame tablets for the PC sales shrinkage, but I think the reason is simply this: the progress is not great.
People have been trained historically to upgrade regularly, enthusiasts especially. But today for people with a 3-4 year old PCs, there is very little reason to upgrade.

Tablets and smartphones had a very large impact on the PC industry indeed. Consumption of web services is a large part of the internet. E-mail, social media, file sharing, entertainment, and general browsing are all fully capable on mobile devices completely bypassing the PC. Toss in free operating systems and access to extremely cheap software and you can see they weren't making that up.

There is a limit to what hardware can do vs. software. We went from uni-processor computers to multiple cores, to uni-processors with hyperthreading, etc relatively quickly. Software still hasn't quite caught up with this. During the era of single core chips that made for a dramatic increase in performance by being able to bounce processes from core to core seamlessly and handle main threads while sub threads hang out on other cores.

Yes replacing PCs themselves becomes less important for the average user. Computers have been able to do spreadsheets and accounting for decades. Basic web browsing works on the cheapest computer from many years ago. Here again we have software to blame. I haven't "Needed" to upgrade my machine except for gaming performance, and full HD playback in one case. Hardware failure still drives replacement at a reduced rate.

There is a reason you don't see more then 4 cores on the consumer class chips, beyond a certain point it just makes the chip more expensive and not perform any better unless used in very specific ways that consumers generally don't.

Certainly doesn't quite fit into Moore's law in terms of costs as we should be about 4 times where we are now. Though the 'law' has been adjusted to make reference to process nodes and IPC vs raw performance.

Honestly Skylake doesn't excite me I see no reason to upgrade for only a slight improvement. My 3770k@4.7ghz is still going strong has no problems with any games I play or programs I use. Hopefully Zen will put some pressure on Intel to make more advancement right now there just isn't any competition and no reason for Intel really push there CPU's.

wrg

This thread is over a year old now. It was partially revived after 3mo of inactivity and now necroposted a year later.

Closed.