TieSKey:

Sounds reasonable but we need to switch from silicon in the next 5 years or we are doomed. We know clocks won't get over 5Ghz on current silicon tech, 10-15 years to double single thread performance is pathetic.

Oh so you develop processors do you? didnt think so....

Passus:

Oh so you develop processors do you? didnt think so....

I don't agree with him but this argument is weak. I can say that about anyone's opinion on anything on this forum.

nevcairiel:

It should be noted that its easier to do that when you are still on lower levels of performance, while eventually it'll slow down once you have exhausted all the low-hanging fruits for improving performance, and the complexity ramps up. The next decade will be truely interesting for ARM to see where they potentially top out and need to slow down - approaching desktop levels of performance, maybe. As did AMD by not producing worthwhile processors for a decade or so.

That's not true at all especially once you consider multicore. Intel could have been ramping up the core counts even if they somehow couldn't push IPC more (which again is BS considering AMD already beats Intel on IPC). But no, they kept quad core as the premium consumer cpu for like a decade. Intel milked it and they lost their talented employees in the process and it shows. Now they are reduced to hoaxes like Ryzenfall and spreading FUD and hiring paid shills. And blaming AMD? Intel did illegal bs against AMD to hold them back when they briefly held the lead before with Athlon 64. AMD has a TINY FRACTION of the revenue that Intel has, which means they can't spend as much on R&D. AMD never purposely held back like Intel did, they were just so small it took a long time to catch up. And sadly not only did they catch up to Intel they are now pulling far ahead.

Denial:

I don't agree with him but this argument is weak. I can say that about anyone's opinion on anything on this forum.

Yeah true enough

slyphnier:

so amd most likely will also depends on tsmc new processing, rather on their chip-design alone the bad things that can happen is if the there are any difficulties (delay) in tsmc roadmap, this can cause issue with amd proggression too reading at news so far tsmc confident with their roadmap up to 3nm... after that might be more challenge?

Die shrinks aren't going to do much for IPC anymore, and they ostensibly are the reason for clock speeds being limited. We emphasize die shrinks with performance a lot more than we really should. At this rate, there are 3 primary benefits to die shrinks: 1. Better efficiency 2. More products that can be fit on each wafer (and since AMD uses a chiplet design, that allows even more products to be fit per wafer) 3. More cores can be fit on a single package. And yes, this does offer more total performance, but not single-threaded performance. So - AMD can't depend on TSMC. As discussed in another thread, they can't depend on adding instruction sets either since that requires developers to adopt their change (and they're definitely not going to do that if Intel doesn't also support it). They need to make optimizations elsewhere, and there is only so much they can do without making sacrifices. The CCX and chiplet designs were a necessary sacrifice, where they allowed for more+cheaper cores at the cost of higher latency.

This was a really good interview by Ian over at Anandtech. I'm glad AMD is going to keep pushing while Intel is down. I do hope AMD can do a summer release for Ryzen 4000 instead of 4th quarter. Moving to 7nm + using EUV should get the clock speeds up(less jagged lines than quad patterning, oversimplification) and should also make much more consistent cores vs what is going on now where we usually only have one core that can hit rated boosts. The move to TSMC's 5nm will be iterative but the move to 3nm is going to be a full node move and will be huge as that is when they move off of FinFET's.

Passus:

Oh so you develop processors do you? didnt think so....

angelgraves13:

We haven't even started 3D stacking silicon. That'll be another 20 to 30 years of innovation. There's a lot of gas left in the silicon tank. Zen 3 will be mine! Zen 4 as well. Intel's history.

The speed limit for silicon based electronics is in a big part, a "problem/property" of silicon itself. I'm just stating the obvious, unless we start demanding more investment in the new materials and thinking out of this box we will continue to be stuck with frequencies like we are since almost 15 years. You can find this data almost everywhere but let me link u one: https://www.karlrupp.net/2018/02/42-years-of-microprocessor-trend-data/ If we had advanced at this pace from the very beginning we would be still rocking 600Mhz cpus.... Ofc it's not easy, but u know, companies like to make money and avoid taking risks, so if people buys a 7% faster processor like hot cakes, why invest billions in trying to bring out something completely new? I know there's research going on but I don't think it gets all the funding it should. Worse still is DRAM speeds... on multithreaded systems it becomes a bottleneck really fast, we need some new whole system architectures too.

H83:

Right now, i would prefer higher clocks over a small IPC increase.

Why so?

mackintosh:

Good, I postponed my main PC upgrade to later in the year and it sounds like this year might bring lots of cool toys for me to assemble an entirely new system with. Sadly, it seems this year will also see my venerable i7-980X put out to pasture as well. Not worth the power consumption to keep it around as a backup system any more. Still, I got a decade of use out of the X58 platform. Just goes to show how little things have been improving until recently.

I have been waiting for a true upgrade to my i7-980 (initially i7-920) for a long time, maybe this is finally the year to upgrade, or might have to wait until 2021-2022, I'm still planning to stick with Intel.

I retired my 980X from my main system four years ago, but it still doesn't feel like it's been adequately replaced. I do think this year will be the year though.

H83:

Right now, i would prefer higher clocks over a small IPC increase.

??? higher clock doesn't mean "more fast" exept on the same model of CPU. 4.0 Ghz CPU can be pounded by 2.0 Ghz in any way. Clock = compare CPU on the same gen or technology equivalent. otherwise it is useless.

mackintosh:

I retired my 980X from my main system four years ago, but it still doesn't feel like it's been adequately replaced. I do think this year will be the year though.

Still have a backup machine with a 980x and 24gb ram, runs everything. Between my 3930k and ryzen 1600 isnt much of a difference compared to the old beast.