Intel's strategy

[I spell gud]

Hi everyone,

This is my first post so I apologize in advance if I do anything wrong.

I've been following Intel and AMD closely in probably the past few years and it looks like Intel is going to have some real competition with the release of Ryzen (FINALLY!). I'm curious as to why Intel seems to be flooding the market with generation after generation of CPUs while only making slightly incremental advances in performance. If leaked benchmarks are correct, then Ryzen would be the equivalent of a 6900K and MUCH cheaper. Why then, would Intel spend money developing products like the 7700k which bring almost nothing to the table? What is their strategy in doing this?

 

 

[tlink]

money extortion is their stratagy /s they mainly focus on improving their chipset i believe and everything that isn't their single threaded performance. also you posted this in the wrong section, you should've read the posting guidelines for this section before posting.

[Nfan]

Because a lot of people already have Intel CPUs and IoT is whole new market: mobile devices, sensors, dashboards, routers with DPDK+OvSwitch/PfSense etc.

New Desktop CPUs appear mostly because their sellers can sell them.

[WkdPaul]

Thread moved to General discussion, @I spell gud make sure you post threads in the appropriate section, thanks

[paddy-stone]

Yes, it sucks for those of us that are looking for bigger performance boost over previous generations. But don't forget the new PC buyers/system builders, for them they're getting the 10% or whatever better performance for around the same price (or a little bit more expensive) than the previous gen, so they're getting a good deal IMO.

To us it doesn't really matter, if performance isn't enough to make us switch then just wait a few years, no worries.

[Oshino Shinobu]

They're making profit off minimal effort and improvements. It's a solid and completely understandable business strategy, even if it does suck for the consumer. 

 

The more appropriate question to ask is: What reason would Intel have to make a big performance improvement?

 

They've settled into making 5-10% performance improvements with each new generation. If they made a big leap in performance, it would just make it harder for them to improve on the next generation. Intel is complacent, they've barely had any competition for desktop processors for years. I'm betting that Intel has something ready to release if Ryzen turns out to be very good, but at this point there's no reason for them to release anything like that. 

[masterkickass7]

Because while zen. Isn't out still there is still no competition. And when AMD drops zen intel will  probably have something in there pocket that would blow zen out of the water 

[LordXenu]

Call me an intel fanboy, but I have been using intel cpus since the 80's I've owned the occasional AMD here and there too, I'm not racist. But I still depend on intel primarily. I am really dissapointed by them not keeping PC cpu performance as the highest priority.  

[Sniperfox47]

All of you are seeking to ignore one important aspect... That there's not much more they can do to make their CPUs faster at this point.

 

I hear lots of people saying to make them perform better but how do you suggest they do that?

 

Architecturally? Their architecture is already executing near 1 instruction per cycle... What are they going to do other than maybe add a little extra cache?

 

Core count? The majority of software other than benchmarks shows vastly diminishing returns after 4 cores... Even 8 cores are very rarely fully utilized.

 

Process? They need to wait for 10mm to mature otherwise they'll have abyssmal production rates. That's not really a realistic thing to push. And optimizing 14nm is what Kaby lake is all about.

 

Clocks? That's what Kaby Lake improved with the more mature process. But there's only so far you can push clocks before fighting with physics.

 

Are there ways to make the CPUs faster? Sure. Are we ever going to see big performance jumps like we have in the past? It's not likely with traditional serial CPUs.

[spellmanuk]

the main changes come in the laptop cpus more than ther desktop cpus as each generation gets a lower TDP meaning significant battery life increase from say a dell xps 13 9434 with 5th gen and a xps 13 9360 with a 7th gen cpu

 

they just release the desktop processors more to make an extra buck apart from things like skylake that brought official support for DDR4 

[Dash Lambda]

Another way to look at it is that Intel is trying to expand their feature set rather than push performance. Sure, Zen may be able to outperform Intel CPU's (hypothetically), but what if you want TB3?

They're adding a lot of exclusive stuff to their CPU's, which can carry them even if they get beaten in performance.

 

24 minutes ago, Sniperfox47 said:

-snip-

I don't know quite enough about computer microarchitecture to speak with certainty, but I at least know that there are a lot of factors that can be tweaked and changed to improve performance. We'll need to develop different technologies to actually see huge gains, but that doesn't mean the 5% improvements we're seeing from Intel are entirely due to physics.

[JCBiggs]

i don't tend to think that intel's strategy is nefarious.  I think they have a plan, and a schedule and they stick to it the best they can.  for all we know they have people in the closet writing completely new chip architectures, microcode, execution code..etc.  do they make profit ..sure... but you have to understand that business the use huge amounts of data and power will pay a certain amount for improvements.. and thats usually percentage base. intel has a clear line into what the improvements will save the company, and charge accordingly.  (thought thats going to bite them in the ass in a few more generations)

 

having amd compete with them MIGHT lead to  bigger increments of improvements.. but i doubt it.  AMD made a ton of money on the PS4 and they used that money to push out the best chip they could...Zen.. if intel was not at the edge of current technology, then zen would be huge percentages of performance better.. and its not.. its right on par...

 

i tend to think the performance increases we all want, are no longer going to come from new hardware.. its going to take getting away from x64/x86 and serious code optimizations to make things happen faster with less overhead.   AVX instructions are a great example of this.   they just need to be taken further.

[Sniperfox47]

3 minutes ago, JCBiggs said:

i tend to think the performance increases we all want, are no longer going to come from new hardware.. its going to take getting away from x64/x86 and serious code optimizations to make things happen faster with less overhead.   AVX instructions are a great example of this.   they just need to be taken further.

This. So much this. If we're going to see more "performance" gains moving forwards it's going to require software developers supporting things like AMD's implementation of HSA. I feel like future SoCs are going to take on dedicated coprocessors for pattern recognition/deep learning (IBM TrueNorth) and computer vision(Google's Tango with a DSP-GPU combo). Performance is going to have to be more about the things supporting the CPU than about the CPU itself.

 

I think that may be where Intel is trying to go, making sure their chipset/platform is as versatile as it can be with the silicon it has to work with so that all these addons have a place to function.

[Mira Yurizaki]

45 minutes ago, Sniperfox47 said:

All of you are seeking to ignore one important aspect... That there's not much more they can do to make their CPUs faster at this point.

 

I hear lots of people saying to make them perform better but how do you suggest they do that?

 

Architecturally? Their architecture is already executing near 1 instruction per cycle... What are they going to do other than maybe add a little extra cache?

 

Core count? The majority of software other than benchmarks shows vastly diminishing returns after 4 cores... Even 8 cores are very rarely fully utilized.

 

Process? They need to wait for 10mm to mature otherwise they'll have abyssmal production rates. That's not really a realistic thing to push. And optimizing 14nm is what Kaby lake is all about.

 

Clocks? That's what Kaby Lake improved with the more mature process. But there's only so far you can push clocks before fighting with physics.

 

Are there ways to make the CPUs faster? Sure. Are we ever going to see big performance jumps like we have in the past? It's not likely with traditional serial CPUs.

I would give you 10 thumbs up if I could, because I walked into this topic thinking this. People want improvements, but do they even know where to start looking at? Do they think improvements just come out of thin air? That and along with the software support for newer hardware accelerated instructions as pointed out before. Maybe Cinebench doesn't use AVX2 instructions (I dunno if it does or not). There's also the point that more compute work is being shifted onto the GPU.

 

But even then, a lot of the programs consumers use are interactive. They are waiting on something at some point. All Cinebench wants is a processor and some RAM. A game wants more than that.

[JCBiggs]

moores law isnt dead... its just taking a break..

 

its well known that adding a third state will increase cpu power exponentionally... thats what im really waiting for..  i think were going to end up with some sort of "pre quantum" cpu that doesnt just check if a gate is open or closed, but also how MUCH power is stored inside the gate... .or something of the sort... thats a very generalized example... this would almost make computers head in an analog direction... with a digital master... its an interesting concept to me. 

[Sniperfox47]

8 minutes ago, JCBiggs said:

moores law isnt dead... its just taking a break..

 

its well known that adding a third state will increase cpu power exponentionally... thats what im really waiting for..  i think were going to end up with some sort of "pre quantum" cpu that doesnt just check if a gate is open or closed, but also how MUCH power is stored inside the gate... .or something of the sort... thats a very generalized example... this would almost make computers head in an analog direction... with a digital master... its an interesting concept to me. 

Trinary computers have been a thing for a long time. They're nothing new. They haven't been used for any high performance applications though due to stability problems with voltage levels and troubles differentiating them.

 

Even if the problems with them are solved though they likely wouldn't see widespread adoption any time in the near future because they'd necessarily be an entirely different architecture than x86.

[JCBiggs]

12 minutes ago, Sniperfox47 said:

Trinary computers have been a thing for a long time. They're nothing new. They haven't been used for any high performance applications though due to stability problems with voltage levels and troubles differentiating them.

 

Even if the problems with them are solved though they likely wouldn't see widespread adoption any time in the near future because they'd necessarily be an entirely different architecture than x86.

I would imagine that these types of cpus would be great in machine learning environments... like anything else, its just a matter of someone doing the leg work.. building the instruction sets, compilers, etc... well get there.. it just not needed yet.  eventually we will hit the wall and have no option but to change. 

[Mira Yurizaki]

30 minutes ago, JCBiggs said:

moores law isnt dead... its just taking a break..

 

its well known that adding a third state will increase cpu power exponentionally... thats what im really waiting for..  i think were going to end up with some sort of "pre quantum" cpu that doesnt just check if a gate is open or closed, but also how MUCH power is stored inside the gate... .or something of the sort... thats a very generalized example... this would almost make computers head in an analog direction... with a digital master... its an interesting concept to me. 

Hybrid computers were used mainly during the 1950s-1980s, but they were mainly used in control systems and simulations.

 

Also gates don't store "power". Gates are insulators. They don't conduct any current. If your gate conducts current, you have a very bad problem.

 

EDIT: Ah wait, sorry, I wasn't thinking right about MOSFETs.

[AresKrieger]

57 minutes ago, Dash Lambda said:

I don't know quite enough about computer microarchitecture to speak with certainty, but I at least know that there are a lot of factors that can be tweaked and changed to improve performance. We'll need to develop different technologies to actually see huge gains, but that doesn't mean the 5% improvements we're seeing from Intel are entirely due to physics.

Needs to be done from the software end rather than hardware at this point, if code monkeys keep using inefficient codecs performance on the hardware side will not compensate for the massive loss on the software end. Now could we see a good jump after ditching silicon, yes but even in those cases we'll still never see moores law level improvements again at least not with electricity.

[JCBiggs]

15 minutes ago, AresKrieger said:

Needs to be done from the software end rather than hardware at this point, if code monkeys keep using inefficient codecs performance on the hardware side will not compensate for the massive loss on the software end. Now could we see a good jump after ditching silicon, yes but even in those cases we'll still never see moores law level improvements again at least not with electricity.

i agree software is a huge problem.   ASICS demostrate what can be done with efficient code on optimized hardware.  they havent optimized software for a cpu in forever.  its now all just optimized to run on windows, and then only marginally.. its all microsoft's fault..

i actually cant remember the last time i used a beautifully optimized fully bug free peice of software.  

[Mira Yurizaki]

32 minutes ago, AresKrieger said:

Needs to be done from the software end rather than hardware at this point, if code monkeys keep using inefficient codecs performance on the hardware side will not compensate for the massive loss on the software end. Now could we see a good jump after ditching silicon, yes but even in those cases we'll still never see moores law level improvements again at least not with electricity.

 

26 minutes ago, JCBiggs said:

i agree software is a huge problem.   ASICS demostrate what can be done with efficient code on optimized hardware.  they havent optimized software for a cpu in forever.  its now all just optimized to run on windows, and then only marginally.. its all microsoft's fault..

i actually cant remember the last time i used a beautifully optimized fully bug free peice of software.  

While we can easily blame software for its shortcomings, there also lies a problem of determinism. A lot of software you use is not deterministic or is not built to be deterministic. One example of software that will never be deterministic is your web browser. And I can comfortably say "never" because your web browser's performance ultimately depends on so many factors outside your control.

 

Anyway, there have been attempts made to go beyond CISC and RISC instruction sets like VLIW and NISC which are highly efficient, but these require the software to be highly deterministic. That or you must build a translation engine that can turn code chunks into deterministic ones your processor likes. So far Elbrus and NVIDIA have made promising results in this field, but nothing really seems concrete because it's not a thoroughly developed concept yet.

 

On another note, I was randomly poking at Wikipedia on transistors and came across the tunnel field effect transistor (TFET). Apparently it's new and it has promise in low power applications: https://en.wikipedia.org/wiki/Tunnel_field-effect_transistor. While it's Wikipedia and should probably be taken with a grain of salt in the end, it had this to say about TFETs:

Quote

The advent of a mass-producible TFET device with a slope far below 63 mV/decade will enable the industry to continue the scaling trends from the 1990s, where processor frequency doubled each 3 years.

 

[JCBiggs]

i think a ternery computer could be the next step IF if was built in such a way that it worked underneath already existing technologies.

 

 

 for example (and this is completely generic and may not even work)   lets say  that the micro code in a cpu has a built in instruction that generates 100 1s and 0s  or 100 bits, and then that 100 bits has to be transfered over to another part of the cpu.  well what during the original generation of the 100 bits, the ternary bit switch (whose results would be specifically pre-addressed into the memory register)  sent a particular and pre determined series of 1s or 0s over to the memory  register...lets say that this particular ternary bit triggers the write of  1111 to a particular memory address, and lets say the series 1111 exist 6 times in the 100 bits of data...   now instead of transferring 100 bits to another section of the cpu, you have only sent 82.. 76bits, and 6 triggers from the ternary switch... 

 

 

so now.. you havent changed the base programming architecture at all, (since the programmer is still programming with x86) but the microcode has interpreted the "to be transfered data"  and reduced it by 18%.  the second cpu sees the trigger at a particular address, and automatically inserts the 1111 in the correct place... if nothing is in the address, it would just over look it.  as this is scaled,  you should be able to send more data through the chip in less clock cycles. and that is ultimately how a cpu gains performanace.. ie "how fast can you move ones and zeros around.   grouping them together could be a huge increase.

 

 

i know this is full of techincal holes. lol  jsut illustrating a point