Is this a serious question? Yes many do
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Have CPUs reached the limit?
- Thread starter fraps
- Start date
Is this a serious question? Yes many do
A lot of programs don't.
Encoding and decoding does, because it's easy to split that work up into N-chunks for each core.
Games are starting to get there, but they don't really need that much CPU for most stuff.
What is potentially more problematic is the lack of competent optimisation in applications. The amount of work you can do these days on moderns cpu's is staggering. But insistence on high level inefficient languages, and lack of understand of hardware means a lot of programs run terribly.
For games to better utilise cores means either needing very good low level programmers given enough time in the development process to optimise things - slim chance!
Or better middleware or frameworks which "attempt" to balance things - maybe by having their own schedulers.
A tall order, but if max clocks and IPC become stagnant there may be no other options.
Increasingly larger CPUs will actually have to dial down the clocks. Maybe more like big.LITTLE like but with even larger differences between the mainstream little cores and 2 to 4 very fast monster cores.
Or better middleware or frameworks which "attempt" to balance things - maybe by having their own schedulers.
A tall order, but if max clocks and IPC become stagnant there may be no other options.
Increasingly larger CPUs will actually have to dial down the clocks. Maybe more like big.LITTLE like but with even larger differences between the mainstream little cores and 2 to 4 very fast monster cores.
Last edited:
Soldato
- Joined
- 22 Nov 2018
- Posts
- 2,839
An example of "reaching the limit" was when Intel were releasing tiny 4% improvements for 10 years. Although that was caused by the limit of their competence.
We're not even close to that at the moment.
We're not even close to that at the moment.
Last edited:
Soldato
- Joined
- 15 Oct 2003
- Posts
- 15,634
- Location
- Chengdu
Wasn't lack of competence in that case, it was lack of competition.
Directors and shareholders might want us to believe we're near the limit, but that's just because of the price bump this would dictate.
Directors and shareholders might want us to believe we're near the limit, but that's just because of the price bump this would dictate.
Or even performance regression, which seems could soon be the case again for Intel.
Soldato
- Joined
- 21 Sep 2009
- Posts
- 5,266
- Location
- London
There doesn't seem to of been any repeats of the boost that was Sandybridge in 2011
Soldato
- Joined
- 19 Oct 2008
- Posts
- 6,029
Jensen @ Nvidia predicted eventually dedicated CPU's will be made obsolete. Was said during one of his presentations but also covered here:
Link
Link
so why are so many tech companies making their own CPUs then...
dood should stop sniffing his leather jackets
Last edited:
Soldato
- Joined
- 6 Sep 2016
- Posts
- 11,227
Sounds like he has silicon poisoning.
Says the man who wanted desperately to get into bed with Intel, head AMD and then buy ARM. Graphics cards will be replaced by APUs ever more rapidly over time.
Associate
- Joined
- 30 Jul 2007
- Posts
- 1,265
Do others get the impression that the writing is on the wall for x86...Why is it in a world where compatibility is king, Windows on ARM has such strong support?...is WINTEL going to be a historical footnote soon. Surely efficiency doesn't trump other benefits in the desktop space.
No I don't think so.
AMD and Intel are teaming up to make sure they make future x86 instructions compatible and more easily adopted.
Intel and AMD Launch x86 Ecosystem Advisory Group to Standardize Architecture
Intel and AMD have established the x86 Ecosystem Advisory Group, aiming to harmonize the x86 instruction set architecture (ISA). This collaboration involves several prominent technology companies, including Broadcom, Dell, Google, HP, Lenovo, Meta, Microsoft, Epic Games CEO Tim Sweeney, and...
www.guru3d.com
There's also some proposals to drop 32bit support and a load of other cruft that would perhaps give up to 5% per core improvement. Though it's mostly to reduce the amount of backwards compatible validation testing they need to do.
Last edited:
Soldato
- Joined
- 16 Aug 2009
- Posts
- 8,068
They've been saying "The End is Nigh!" for x86 for the last 40 years. x86 was obsolete the day they stopped making the original IBM PC. Everything since then is all about compatibility efficiency be damned.
Soldato
- Joined
- 25 Sep 2009
- Posts
- 10,006
- Location
- Billericay, UK
I'm no expert but from what I've read and learnt on this issue the main driver that's holding back CPU performance is available memory/cache.
Memory doesn't get the same benefit from a node shrink as processing cores do so going forwards the focus will be on packaging and designing ways to bring memory chips closer to the.CPU die like AMD have done with X3D.
Memory doesn't get the same benefit from a node shrink as processing cores do so going forwards the focus will be on packaging and designing ways to bring memory chips closer to the.CPU die like AMD have done with X3D.
Or make CPUs bigger - why be constrained by old ATX standards, make CPUs are big as GPU and change the motherboard design, then you'll have plenty of space for memory, maybe the CPU can slot in like a PCIE card? Or maybe this isn't the right solution but the point is to always ask "why" - if we say the issue is too little memory on CPUs, then ask why can't we have more memory, and if the answer is because it takes up too much space, then ask why are CPus so small compared to other PC parts and why can't we make them bigger, oh because of ATX there is limited space for socket, so why are we using 30 year old ATX then etc
Last edited:
Now they're out of their contract with glofo hopefully they'll do just that
With gaming it's always been an issue that you'll typically have a small number of very hungry threads that it is hard to balance and parallelise. This has obviously improved in recent years (because that's how CPUs have scaled out) but I think the effort required to fully optimise for multithreading may not be cost effective in some cases.
One thing I found fascinating was how incredibly well optimised Assassin's Creed Unity seems to be for multi-core. I have a 12c24t cpu and it basically spreads the load across 24 threads. This is a 10 year old game that got slated for its performance on release, yet has the best threading of any game I've seen. In terms of peak cpu utilisation it is probably the highest, it's very rare in other games that i'd see 20+ threads all ticking along at over 50% utilisation
There are some older games especially Hitman and Assassins Creed titles which will spread over a lot of cores/threads but they seem to have just "threaded all the things" for the sake of threading and it doesn't often gain much - in fact in some cases makes the game feel a bit unresponsive to play.
As you mentioned though some of the core heavy lifting of game engines is very serial in nature and very hard to thread effectively - where games can utilise additional cores often comes from additional and/or more complex features like physics.