The disappointing slowdown CPU progress in last 6 years vs 4 years before (10 yrs) ago

[unseengundam101]

I have been comparing my speed increase from the Core2 Duo E8400 in 2008 to fasted CPU I got, a 3770K from 2012.

In UserBenchmark, I see a 440% increase in Mixed Multi Core, 110% in Single Core, and 275% in average user bench, 209% Effective.

How compare 6 years ahead (to CPUs in 2018) my 3770K vs i9-9000KF, I see a 182% MC increase, only 27% Single Core, 92% Average User, and 62% Effective

3770k vs 2018's Ryzen 2700x is even worse with 141% MC, measly 12% Single Core, 60% user, and 33% effective.

Ryzen Zen 2 (3950x) just might be able to get to 300% MC, but I doubt it will hit my 440%. Also, that Single core I am excepting only a 30% bump, no where close to jaw dropping 110% increase from 08 to 12, in just 4 years.

I am expecting such a slowdown in CPU progress that after my next CPU upgrade in 1-2 years I bet I could stick with for 1 or 2 decades (10-20 yrs)...

 

[Markfw]

Of course it will slow down, but progress is being made. Is this a troll post or what ?

 

[cellarnoise]

I do miss the old days of cpu advancement which was single core heavy. Wish speed of electrons and light would change for single core to have a chance to advance. 2 bit / on / off switches are also holding us back. Good to have cpu industry going wider and software trying to go multi-thread after all these years. Tom's had the scoop on the dual socket pentium 133mhz years ago though really did not have software to use it as I remember. Imagine a narrow gpu today. Like watching paint dry on the screen.

 

[IntelUser2000]

I have been comparing my speed increase from the Core2 Duo E8400 in 2008 to fasted CPU I got, a 3770K from 2012.

In UserBenchmark, I see a 440% increase in Mixed Multi Core, 110% in Single Core, and 275% in average user bench, 209% Effective.

There's no way a 3770K is 110% faster in single thread over E8400. At best, maybe 50% faster in average.

I am expecting such a slowdown in CPU progress that after my next CPU upgrade in 1-2 years I bet I could stick with for 1 or 2 decades (10-20 yrs)...

This is natural. Exponential increases plateau after some time. Yes, it can be seen as disappointing, but I believe we are actually in the Golden Age of computing. Powerful computers are now in the palm of your hand. People in poor African countries in documentaries about their living conditions have cheap Smartphones. Computers of all sizes and form factors are proliferating.

 

[epsilon84]

If you look at overall MT performance then I don't think things have plateaued at all, especially with Ryzen pushing the boundaries.

IPC and clockspeeds have really hit a wall though, so I'll give you that.

 

[linkgoron]

This is only true if you look at Intel.

AMD have made huge performance improvements across the board - both in ST and MT. ARM/Qualcomm/Samsung and especially Apple have also made very impressive gains in the last few years.

 

[Nothingness]

This is only true if you look at Intel.

AMD have made huge performance improvements across the board - both in ST and MT. ARM/Qualcomm/Samsung and especially Apple have also made very impressive gains in the last few years.

Starting from lower perf made their task easier

Intel had such a lead... I wonder if they can still find a lot of perf without killing efficiency. I guess conventional out of order CPU are reaching a plateau; most of them now have very similar micro-architectures.

We can only hope there will be a breakthrough but I'd be surprised if it didn't require heavy changes in software and programming languages.

 

[epsilon84]

This is only true if you look at Intel.

AMD have made huge performance improvements across the board - both in ST and MT. ARM/Qualcomm/Samsung and especially Apple have also made very impressive gains in the last few years.

It's all relative I guess. Ryzens gains vs Bulldozer are as much a reflection on how bad BD was just as much as how good Ryzen is. Intels IPC gains vs Core 2 doesn't seem all that impressive but if you compare it to a P4 it basically mirrors Ryzen vs BD

 

[PotatoWithEarsOnSide]

Going from 4c mainstream to 16c mainstream in around 30 months is a big deal IMO.

 

[PotatoWithEarsOnSide]

Starting from lower perf made their task easier

Surely that applies to older CPUs too, otherwise you're asking for the future to defy your own logic...?

 

[ondma]

This is only true if you look at Intel.

AMD have made huge performance improvements across the board - both in ST and MT. ARM/Qualcomm/Samsung and especially Apple have also made very impressive gains in the last few years.

Well, for AMD, that is easy to do because they were so far behind. If you have a heavy multi-threaded workload, they yes, AMD has made progress. However, in single threaded, or even up to 6 core workloads, they are now just catching up to what Intel had two years ago (8700k). The real progress has been in Apple and ARM chips, but likely at some point that will slow down as well.

 

[ondma]

Going from 4c mainstream to 16c mainstream in around 30 months is a big deal IMO.

In what world is the FX 8 "core" not mainstream. The FX 8xxx has been available for many years. More correctly, they have gone from 8 cores to 16.

 

[A///]

What were your opinions on Clarksdale and Arrandale? Intel brushed those under the rug a long time ago in your timeline. I'd argue that there's a lot of untapped ability in processors today because Windows and Linux simply don't focus on optimization as much as they could.

Of course it will slow down, but progress is being made. Is this a troll post or what ?

Has it truly slowed down or does it seem like it slowed down to Intel becoming comfortable?

 

[A///]

The real progress has been in Apple and ARM chips, but likely at some point that will slow down as well.

Are these even relative to a traditional desktop processor? They're faster in certain benchmarks, but could you do AAA desktop gaming at 1440p or 4K for hours? 1080p to 4K video editing?

 

[IntelUser2000]

Starting from lower perf made their task easier

Even on PCs, the single thread performance of 15W Amberlake Whiskeylake chip is very close to the 9900K.

9900K/8565U

Cinebench R15 ST: 213/179(19%)
Geekbench 4.3 ST: 6339/5447(16.3%)

Their Goldmont Plus core is 30-40% faster per clock compared to Goldmont, making it roughly equal to Core 2. Say Tremont brings another 30-40% leap(and I have reasons for believing this). That may bring their "little" cores to Haswell territory! Tremont on Haswell level means those vaunted much larger Sunny Cove cores are going to be maybe another 40% better.

Sure, the vector performance is much weaker, but that's much easier to improve. And they are going to focus on this with Gracemont next gen cores.

Then more than ever, cores are going to be separated mostly because of thermal/power limitations. That'll lead in more specialization with specific instructions, and accelerators that sacrifice flexibility for much more performance.

This may not be a popular opinion, but expect consoles to differentiate themselves even more as a gaming system. From what I'm reading, the implementation of the storage architecture in PS5 and Xbox "Scarlett" should show what SSDs can truly do, not the meagre 2-3x real-world gains we see on PCs even with NVMe interface. It won't easily be replicated on PC because of the truly vast ecosystem with its dizzying array of hardware configurations.

 

[Nothingness]

Surely that applies to older CPUs too, otherwise you're asking for the future to defy your own logic...?

I'm not sure to understand what you mean. Do you deny AMD and ARM cores did get more performance increase each last generations than Intel cores? Do you also deny at the starting point a few years ago, they were significantly less performant than Intel cores?

 

[A///]

Isn't Amberlake in those low end MacBooks?

 

[IntelUser2000]

Do you deny AMD and ARM cores did get more performance increase each last generations than Intel cores?

Yea, you can just look at Intel side and see this. 18% with Sunny Cove? What about 30-40% with Goldmont Plus? And 30-40% with Goldmont?

Isn't Amberlake in those low end MacBooks?

Corrected. I meant Whiskey Lake.

 

[PotatoWithEarsOnSide]

I'm not sure to understand what you mean. Do you deny AMD and ARM cores did get more performance increase each last generations than Intel cores? Do you also deny at the starting point a few years ago, they were significantly less performant than Intel cores?

I meant, the CPUs from 10 years ago were also of lower performance ergo the performance gains were easier to make.
Asking for exponential growth is simply unrealistic. Going from 1 to 2 is a 100% improvement, yet going from 2 to 3 is only a 50% improvement despite the fact that the gain is exactly the same amount in real terms.

 

[Flayed]

I had a thought, hypothetically speaking if AMD or Intel designed a cpu for a TDP power budget of say 300W would it have more performance than heavily overclocking a current chip with an incresed voltage such that it comsumes 300W?

 

[amrnuke]

I'm not sure to understand what you mean. Do you deny AMD and ARM cores did get more performance increase each last generations than Intel cores? Do you also deny at the starting point a few years ago, they were significantly less performant than Intel cores?

What's happening now - why returns are diminishing - is because there is a wall with silicon. A silicon atom is 0.2nm but the behavior of circuits breaks down far before that theoretical limit. Silicon's electron mobility degrades severely with heat, so while transistor density continues to rise, cooling has become a major concern. One cannot simply clock higher, because with higher clocks and denser transistors comes more heat, which degrades performance. Further, electron hole mobility is poor in silicon. They're working on feasibility of semiconductors like GaN or SiC but these are not nearly as cheap as plain silicon. Though, in critical applications, GaN and SiC have taken on big roles.

So with all the limitations of silicon, there is a threshold for performance. Intel have already hit the flattening curve of diminishing returns, now AMD is likely there, and ARM and Apple are coming soon.

Three things are already happening:

1) they focus on optimizing software for multithreading, and continue to refine single-core efficiency with small tweaks (similar to how commercial jetliners have an optimal cruising speed of 550mph - and now they're working on efficiency for less fuel consumption by adding scimitar tips to wings, high-bypass turbofans, etc)

2) concurrently, engineers continue investigation into new materials for traditional electronics, which is a huge undertaking, e.g. moving back to germanium instead of seeding silicon with germanium, or utilizing indium arsenide with germanium. The electron / electron hole mobility in germanium and indium arsenide are great, but the technological and manufacturing issues with them are high and very costly to develop. In the end, though, it's likely that silicon, germanium, and indium arsenide will end up having a huge role in future speed increases from a semiconductor standpoint

3) concurrently, engineers will work on replacements or supplements to electronics -- photonics and quantum computing. We probably won't see this in our lifetimes at a mass scale.

So it is ENTIRELY unsurprising that speed increases from a single-core perspective are stagnating. It's because silicon has diminishing returns with respect to transistor density - hard limits that are bound to the silicon atom's size and physical properties. We are waiting on the technology to grow up to move on from silicon-based electronics to other semiconductors as well as to photonics and quantum computing.

 

[dainthomas]

There's no way a 3770K is 110% faster in single thread over E8400. At best, maybe 50% faster in average.



This is natural. Exponential increases plateau after some time. Yes, it can be seen as disappointing, but I believe we are actually in the Golden Age of computing. Powerful computers are now in the palm of your hand. People in poor African countries in documentaries about their living conditions have cheap Smartphones. Computers of all sizes and form factors are proliferating.

Raspberry Pi is probably the best example of extremely affordable with good performance.

 

[moinmoin]

My first thought was this thread must be a necro bump of a topic before Ryzen was launched.

We essentially reached the top end of classical ST performance a decade ago. Since then there have been more (more and more SIMD extensions, how about the mythical multi-cores for single thread execution eventually?) or less (brute force frequency increase) creative approaches to improve it further.

 

[Thala]

Are these even relative to a traditional desktop processor? They're faster in certain benchmarks, but could you do AAA desktop gaming at 1440p or 4K for hours? 1080p to 4K video editing?

Yeah sure, why not?. All the mainstream ARM architectures (ARM Cortex A77, Exynos 9820?, Apple A12) have surpassed Intel and AMD with respect to IPC. The only thing holding actual implementation of these architectures back in terms of desktop computing performance are the moderate clock frequencies as direct consequence of low TDP and hence low Voltage and high-density/high-Vth cell library mix.

 

[ondma]

Yeah sure, why not?. All the mainstream ARM architectures (ARM Cortex A77, Exynos 9820?, Apple A12) have surpassed Intel and AMD with respect to IPC. The only thing holding actual implementation of these architectures back in terms of desktop computing performance are the moderate clock frequencies as direct consequence of low TDP and hence low Voltage and high-density/high-Vth cell library mix.

That seems like a gross oversimplification. Would be easy to change the TDP and take over the desktop market as well as the mobile one. As another poster said, it is basically an apples to oranges comparison, since they dont run x86.