Hacker News Reader

Comment by alkonaut | Hacker News Reader

The reason there is diminishing returns is simply because intel doesn't have to try. Shrinking the process is terrifyingly expensive and difficult, yet they slowly are doing it.

If intel has no competition, they tweak the product enough to give it a 2017 sticker and sell it at the same price. Because why not?

Only more cores will save the performance trend. So if Intel had any competition in the enthusiast segment, the enthusiast i7's in the $350-500 range would have gone from 4C/8T to 6C/12T to 8C/16T in the years since Sandy Bridge.

View on HN

Intel has always been driven to push harder than anyone else, and they're not slacking off here. They're just out of easy gains.

In the early days of microprocessor design they could steal concepts from mainframe computers. Then later they could port features from ultra-high-end chips like those found in the Cray systems, then things like Alpha which they acquired from DEC. Lately they've packed everything they've mined from those systems and more into the current generation chip. There's not much left they haven't already done.

The only way out of this particular jam is to come up with entirely new ways of making CPUs. Maybe this is why Intel's acquisition of an FPGA manufacturer is relevant, that will open up entirely new opportunities. Maybe that's why they're exploring many-many-core designs. Maybe it's why they're investing more heavily into boosting GPU performance.

> So if Intel had any competition in the enthusiast segment...

The enthusiast segment is almost utterly irrelevant to Intel's bottom line. They're willing to pay a steep premium on certain chips, which can help when breaking in a new process, but in the long haul they're, at most, 5% of the market. Intel's missing out on way more money by not being a bigger player in the mobile space.

If Intel could've shipped a cost-effective 6C i7 design they would've done it by now. There's a version on the market for $600 which isn't too far out of your envelope, and we are talking about enthusiasts here so they're normally not as price sensitive as general consumers.

> they're not slacking off here. They're just out of easy gains.

They are slacking because they can. AMD hasn't been competitive in the desktop space for over a decade.

Okay tough guy. Show us your working 10nm process.

According on this: http://kyokojap.myweb.hinet.net/gpu_gflops/

There were multiple factors of improvements in speed in the GPU inside last few generation of Intel CPU.

Unfortunately, folks who care about GPU performance likely use real graph card in their system.

There is zero growth in GPU performance from the 520 to the 620.

Intel competes with itself. It needs to encourage people to upgrade as frequently as possible.

Also, for 99% of the users performance is good enough. People basically use PCs for browsing, email, word processing, spreadsheets, presentations, and some photo retouch.

I think the PC gamers are a bit more than 1% of the users.

Lets say what? 10% of PCs sold are for high performance gaming? Any other number?

I would be very surprised if it was that high.

Perf is good enough for gamers too (in the sense that the next dollar spent should always be on the graphics card).

The perf difference between my 2500K and my 6600K is not noticable.

Depends what your work-load is. For VR, CPU heft is becoming important again. I have an OC'd 2500K and that's the minimum for more advanced VR games.

CPUbench

2500k: 6400

6600k: 7921

That's a 20% improvement which doesn't sound like much on paper, but in real world testing that means about double the FPS for higher-end video cards at high res due to the older 2500k design lacking optimizations and other issues holding back those cards.

http://forums.joinsquad.com/topic/5973-my-2500k-vs-6700k-upg...

Lastly, gamers would most likely opt for the i7 6700k which benches at 11000, so almost double that of the 2500k.

Yeah the step from 2500k to today isn't nothing, but 2500K was also released in Januray 2011 making it six years, which is about a million years old in CPU terms.

At the same time, i7-2700k is around 8900: quicker than the 6600k. A generation later, and socket compatible, is the i7-3770k, around 9500.

On CPUbench, there's a big difference between i5 and i7 parts, and going to a i7 that's generations old gets you better scores than a current top-of-the-line i5.

Well, 2500k doesn't have AVX2 instructions (for 256-bit width), so integer stuff needs to use 128-bit wide SSE2+.

In some [1] cases that can make quite a bit of difference, up to almost 2x.

[1]: Real improvement is of course anywhere between none and 2x. A lot of software (most even?) is still not using AVX2 at all. Sorting, memory block operations (strlen, memcpy, etc.) and operations that need to do integer math on big chunks of data can benefit significantly. Some real world code I wrote saw up to ~1.9x improvement on systems with sufficient memory bandwidth.

what makes you think that?

also: my 2011 sandy bridge (2600k or 2700k, at work so can't check) does fine. these days the gpu is the major arbiter of gaming performance.

Actually....

While GPU is a MAJOR arbiter, CPU can bottleneck you too and it can hit you in the FPS. Years ago my gaming PC had inexplicably low FPS for the GPU it was using, and GPU utilization wasn't anywhere near 100% but CPU was... pop in a new chip+mobo and gaming was waaaay more enjoyable.

There are subtle differences in how bottlenecked GPU vs CPU acts in most games, but they're there.

CPU can be a bottle-neck, but only if the game happens to be very CPU-heavy (rare) and you happen to have a poor CPU.

In most modern games, a better CPU may yield a 10 FPS improvement, while a better GPU for the same price would yield 3x that.

Dwarf Fortress would like a word with you. :) Probably also Factorio.

But I suppose it's accurate to say, in general terms and given the kinds of games that are currently most popular, GPU bottlenecking is much more common.

Dwarf Fortress was discussed on HN a few times and even the author admitted that the code quality is bad and it's unoptimized.

I don't think it's a compelling argument for CPU being the bottleneck when your example is an unoptimized game with a sloppy codebase.

Why not? Out in the real world most users don't have the ability to peer into their code and recompile. Computing has a rich history of throwing hardware at problems when you can't fix em in software

Isn't this a contradiction in your own point?

> throwing hardware at problems when you can't fix em in software

DF's problems can be fixed in software, it's just that the creator chose not to. He likes spending his time adding more features instead, performance be damned.

Not that I blame him, he can do as he pleases with his own project.

I had similar experiences with ATB, Guild Wars 2, and Skyrim. Lots of games are heavy on the CPU

Honestly though, I think even in PC gaming, the processor is not where the bottleneck is at.

In the early 2000's, buying a new processor every year would yield noticeable returns in gaming performance. Once we entered the console port life cycle (circa 2008), where the performance of the dominant consoles defined what developers put in games, I found that from 2010-2014 I was able to use the same processor (and not a top of the line one) with no trouble, and only an intermediate graphics card upgrade.

I'm now on the core that I bought in 2014, and I honestly expect it to be good until 2020. We'll see a bit more get juiced out of graphics cards, but we're nearing a stage where the cost of making graphics good enough to overwhelm modern graphics cards will be so high that it won't make sense from a cost/benefit perspective. At that point, I expect CUDA applications to be the driving force behind graphics card R&D.

This bears out in benchmarks too - you can find sites that benchmark games against multiple generations of Intel CPUs and find that the performance difference is minimal to nonexistent.

I'm running a Haswell i7 in my gaming machine and looking at the numbers I'll maybe get a ~3% bump by "upgrading" to Skylake. No thanks.

considering the fact that most AAA titles are glorified visual experiences it is no surprise cpu is not the bottleneck, requiring probably very little logic and not having to render more than a 60dg first person FOV. However, when you try any game that has some flavor of RTS to it, they start trudging, I remember reading RTS games target 20fps as optimal performance in general, because of lots of polygons but also because they need to track a lot of stuff compared to your average scripted experience. Free roamers a la gta/arma/watch dogs also cpu bound, just like most MMOs which have to display lots of stuff and track of lots of things. The gaming segment is definitely something that would benefit from beefier cpus I would say.

Another factor to take into consideration is that people change gpu more often than cpu (I would blame intel and their sockets policies but I dont know) so it is obvious you target the lower common denominator which is gonna be the cpu therefore games that don't use lots of cpu to begin with.

Last argument, cpu usage in games can't really be scaled like gpu usage, you can turn down shadows at best but a game is gonna always use that much cpu, being again thelowest common denominator.

As much as I love CPU-heavy Paradox RTS games, I really doubt that they're on Intel's mind.

Performance is tolerable, not "good enough". Booting to desktop could be faster, rendering websites is too slow sometimes, and most applications still don't start instantaneously. The problem is not raw performance – it's that too many developers don't really give a shit about their user's experience as long as they can save even a couple of minutes of their oh so precious time. Computers got more powerful? Great, that means we can use an even slower dynamic programming language and add another layer of abstraction on top of our framework!

Faster CPUs won't change that, same as in the last 20 years.

Front-end frameworks are the bane of mobile web surfing.

> Faster CPUs won't change that, same as in the last 20 years.

In the last 20 years we've been throwing more and more hardware at the problem, eventually we'll hit a wall and start conserving those precious cycles.

>it's that too many developers don't really give a shit about their user's experience as long as they can save even a couple of minutes of their oh so precious time.

Really? The developers' fault? Whose time are we talking about here?

The company's time? In workplace development, it's managers putting down on the development effort that goes towards optimization / performant design - and not for crazy reasons, simply because in many market conditions and applications, it looks like throwing money away on the balance sheet.

If you're talking about open source projects, any of a developers' time we get is gratis; this would be a pretty good example of looking a gift horse in the mouth.

> [...] and not for crazy reasons, simply because in many market conditions and applications, it looks like throwing money away on the balance sheet.

Would you use this argument to excuse software bugs? What about security holes and data leaks? Besides, the 80/20 rule applies, i.e., a little more effort spent on performance goes a long way towards a better user experience.

> If you're talking about open source projects [...]

Interestingly, this problem is less prevalent in the OSS world, because C and C++ is still very widespread there.

I know about an order of magnitude more people who play games than people who retouch photos.

Reply on news.ycombinator.com