Is mainstream desktop CPU development "completed"?

[mikk]

I'm expecting a 5% IPC gain from Broadwell to Skylake. I'd be surprised if it's 10% and shocked if it's greater than that.

You will be shocked then. 5% might be the difference from HSW to BDW.

We haven't seen more than 10% since Prescott to Conroe and that was 10 years ago.

Sandy Bridge improved IPC by 10-15%. Also Nehalem was over 10% afaik.

 

[Ramses]

http://www.techradar.com/us/news/computing/beyond-silicon-the-processor-tech-of-2035-1039114

 

[Fjodor2001]

Do I need to remind you the IPC progression over the years?

But again, what you ask for is 500W huge CPUs. Something there isnt buyers for.

Do you also remember CPU prices back then? 1000$ in todays money didnt get you much. A P3 1Ghz would cost you 1400$ today. A P2 400 1150$ and so on.

Would you also be willing to pay the price? Or is it more like 200$?

Lots of statements, none of which I have said. Strawman rampage. I see you are up to your usual standards.

 

[Fjodor2001]

One could also say you use old obsolete code.

What matters is the code in SW that people actually use, like it or not. And you won't be seeing any 80% performance increase on that going from Ivy Bridge->Haswell like you claimed. Even if the SW developers designed new SW taking full advantage of the new instructions you'd not see that kind of performance improvement for general SW. Only for very specific use cases, like encryption or media encoding.

 

[witeken]

I'm not a huge fan of those futurist articles. The most interesting thing is:

Intel has a well-established compound semiconductor programme, and in 2005 it announced an InSb (indium and antimony) transistor that was a remarkable five times faster than its silicon equivalent, with a 10th of the power consumption.

Intel is already working for more than a decade on III-V transistors, much longer than I thought.

 

[Fjodor2001]

Welcome in reality.

I already know that's reality. As I said, I don't expect to see the performance improvements I listed as necessary to be considered "major" in the near future.

That's why I have arrived at the conclusion that mainstream desktop CPU performance increase has ground to a halt, and that desktop CPU development is more or less "completed" as far as performance goes.

A combination of market economics, consumer demand, but most importantly the laws of physics, is what killed it. At least for now. Let's just hope there is some disruptive discovery in the future that resurrects it so we'll see some noteworthy performance improvements again.

 

[Ramses]

I'm not a huge fan of those futurist articles. The most interesting thing is:

Intel is already working for more than a decade on III-V transistors, much longer than I thought.

I'm more interested in them than the SOS we have now lol..

 

[Hulk]

You will be shocked then. 5% might be the difference from HSW to BDW.




Sandy Bridge improved IPC by 10-15%. Also Nehalem was over 10% afaik.

Ivy to Haswell was about 5 to 8% on legacy apps not using new instructions, which is what I based my assumption on.
http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6


http://www.anandtech.com/show/2658/18

Yeah, Penryn to Nehalem was pretty damn significant. Can't argue with you there!

The memory subsystem overhaul made a big difference on real world apps.

 

[creativedotit]

Rumors tells that SAMSUNG wants to acquire AMD.

http://www.bidnessetc.com/37987-sam...ed-micro-devices-inc-and-battle-industry-gia/

A leader on mobile chips joins the CPU/GPU PRICE/PERFORMANCE LEADER.

AMD will have enough founds to revive its research&development department.

Competing AGAIN with a very advanced Intel one

Can you imagine what will happen?

SoC based computers are more close to reality than you think.

Also because, the new Macbook Air LAPTOP it's already out with 14nm mobile processor


Did you remember what happened when Apple did the iPod? CD died.
Did you remember what happened when Apple did the iPhone? The revolution on mobile.
Did you remember what happened when Apple did the iPad? Tablets from every vendor, including microsoft.

What will happen now that it put a 14nm processor on a LAPTOP?

Do you think that this trend will not reach Desktop and Server Market?

(and 14nm is not the end, 10nm is planned for 2016 and 5nm for 2020, and some scientists in 2012 demonstrated that is possible also to have a 1nm, single atom, transistor:
http://en.wikipedia.org/wiki/5_nanometer )

So forget your liquid cooling. You'll soon have an actual Destkop power in a 2,5inch box, just connect your monitor and keyboard to it.

Larger cases will be needed just to have multiple CPUs, FINALLY multi-CPU and parallel GPUs will come to your Desktop.

You think you don't need HUGE multicore/parallelism but I think this is the most big bottleneck of ANY current computer. When you browse the internet with chrome did you checked task manager to check how many processes/trheads it spans when you have lot of tabs open? Apps and OS(s) are already ready for huge multi core. For now those process are using a crappy time-division round-robin on our very limited 4C+8T CPUs, but when we'll have a 80core@5GHz SoC we'll be fine to enjoy your MULTI-TASKING computer.

What about GAMES?
In the same way, i don't think that SINGLE CORE @12GHz can do a GAME better than 80cores@5GHz. Also GAME 3D rendering, and game logic, can be processed in parallel, and it's already this way in many games I think.

And 14nm processors can reach 5GHz they just need to leave the "underclocking" they put for saving battery life on mobile devices and optimize things on the chips (it's like it's done with new DDR-X RAM is out, the frequency gets improved some year later after engineers work on optimizing the chips and always a lower nm architecture will go more fast than the previous one)

Anyway the ONLY way to dream something like 10GHz+ is just the new mobile-driven nanotech development, because the smaller it is the less you need to cool it down (so you can make near-to-unlimited power happen on the silicon, because nowadays the issue it's like the nuclear fusion issue, you can't bring to more frequency because you'll melt the container) .

http://www.extremetech.com/computin...that-will-take-us-to-the-limits-of-moores-law

http://www.extremetech.com/extreme/...scaling-exploring-options-on-the-cutting-edge

So I think, "desktop CPU development" is not completed, au contrarie, we're facing a new BIG revolution. Give just some year or two: just the time for them to sell out all the current 32/22nm stock and they will launch awesome things.

You'll have fun in new way, liquid cooling fan? Yes if AMD/Samsung/Intel will release some Overclockable SoCs/3D chips etc. you will have fun the same way, but trying to bring them at 20GHz!

 

[cbn]

The premise of your argument in the OP appears to be that you believe the pace of advancement stagnated, and thus the consumer had no choice but to migrate to other compute platforms and spend their money on mobile products. I don't buy that, the development money follows where decision makers think the markets are headed.

In the past, I have seen written that every node shrink brought good reduction in voltage and power all while using planar xtors.

Then this slowed down and progressively more exotic materials started to be needed just to approach the gains the old nodes had. Then at 22nm, Intel started using exotic transistor designs in addition to exotic materials.

So in some ways isn't the move to mobile also influenced by increasing R & D costs. If node development didn't get more expensive it probably would have been a lot easier to design wider CPU cores that still met various performance per watt targets.

 

[shady28]

You will be shocked then. 5% might be the difference from HSW to BDW.




Sandy Bridge improved IPC by 10-15%. Also Nehalem was over 10% afaik.

IPC is an extremely poor way to measure performance changes from one arch / generation to another. It completely ignores that one arch may clock up very high, while another clocks very low.

Ignoring that, it's conceivable that a CPU could be made that ran at 1Mhz and had an IPC 100x greater than a competitors 1Ghz CPU. Yet the 1Ghz CPU would still be 10x faster. And it may well run cooler, and take less power.

Mhz still matters :

 

[TidusZ]

Funny how 5Ghz seems to be some kind of OC standard now for SB. Just a shame it wasnt so, it was very rare.

I o/c'd two 2500k with safe voltages, one hit 5 ghz other hit 4.8. I imagine a 2600k would hit 5 ghz more often, but judging from o/c's I saw on forums and my own experience I doubt it was that rare

 

[mikk]

Ivy to Haswell was about 5 to 8% on legacy apps not using new instructions, which is what I based my assumption on.
http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6

I was talking about Sand Bridge. Go to hardware.fr, computerbase, ht4u, ixbt. IPC increase was in a range of ~12-13% averaged. Ananadtech is a bad source for CPU tests by the way.

 

[Hulk]

IPC is an extremely poor way to measure performance changes from one arch / generation to another. It completely ignores that one arch may clock up very high, while another clocks very low.

Ignoring that, it's conceivable that a CPU could be made that ran at 1Mhz and had an IPC 100x greater than a competitors 1Ghz CPU. Yet the 1Ghz CPU would still be 10x faster. And it may well run cooler, and take less power.

Mhz still matters :

Yeah but you're missing the point completely. We're comparing architectures, ticks and tocks, that are all based on the same basic architecture going back to Conroe. Intel has been improving the same core. In addition, clockspeeds have stalled over the last 7 or 8 years. Sandy, Ivy, Haswell have all clocked within a few hundred MHz of one another.

We're not talking about hypothetical processors running at 1MHz and 100MHz. We're discussing the real thing.

 

[shady28]

Yeah but you're missing the point completely. We're comparing architectures, ticks and tocks, that are all based on the same basic architecture going back to Conroe. Intel has been improving the same core. In addition, clockspeeds have stalled over the last 7 or 8 years. Sandy, Ivy, Haswell have all clocked within a few hundred MHz of one another.

We're not talking about hypothetical processors running at 1MHz and 100MHz. We're discussing the real thing.

In the "real world" people don't care about IPC.

If you are talking about desktop performance, compare the fastest 4-core desktop class CPU available in say April 2010 vs April 2011 vs April 2015.

That would be the i7-930 (2.8 Ghz) released in Feb 2010, the i7-2600k (3.4 / 3.8 Ghz) released in Jan 2011, and the i7-4790K (4.0 / 4.4Ghz) released in June 2014.

If you look at the clock speeds I just listed for those 3 years, it pretty much tears up your argument that only IPC matters.

By that I mean - in the "real world" - a consumer looking for a fast desktop PC would be looking at one that has a 42% higher clock rate today than what was available in 2010.

Comparison here between the i7-930, i7-2600k, and i7-4790k :

http://www.tomshardware.com/charts/cpu-charts-2015/compare,3712.html?prod[7418]=on&prod[7440]=on


The performance difference is a lot more than 20%.

This is i7-930 vs i7-2600k vs i7-4790k :

Total Time
Of all time based benchmarks
(time in seconds)
Score in Time Go 1224.00 1678.00 2238.00

 

[ShintaiDK]

The i7 930 wasnt the fastest quadcore.

i7 880 was 3.06Ghz in 2010. In 2008 it was funny enough 3.2Ghz with C2Q.

IPC certainly matters.

The frequency barrier was hit in 2004 at 3.8Ghz. 11 years after and we got +600Mhz.

 

[Dave2150]

In the "real world" people don't care about IPC.

If you are talking about desktop performance, compare the fastest 4-core desktop class CPU available in say April 2010 vs April 2011 vs April 2015.

That would be the i7-930 (2.8 Ghz) released in Feb 2010, the i7-2600k (3.4 / 3.8 Ghz) released in Jan 2011, and the i7-4790K (4.0 / 4.4Ghz) released in June 2014.

If you look at the clock speeds I just listed for those 3 years, it pretty much tears up your argument that only IPC matters.

By that I mean - in the "real world" - a consumer looking for a fast desktop PC would be looking at one that has a 42% higher clock rate today than what was available in 2010.

Comparison here between the i7-930, i7-2600k, and i7-4790k :

http://www.tomshardware.com/charts/cpu-charts-2015/compare,3712.html?prod[7418]=on&prod[7440]=on


The performance difference is a lot more than 20%.

This is i7-930 vs i7-2600k vs i7-4790k :

Total Time
Of all time based benchmarks
(time in seconds)
Score in Time Go 1224.00 1678.00 2238.00

I suggest you read up more. I7 930 wasn't the fastest quad core at that time.

The fastest quad-core as of April 2010 was the I7 960 @ 3.2Ghz.

Not that it matters, since every I7 920 (released 2008) could clock to 3.4Ghz+ easily on air.

 

[Fjodor2001]

The i7 930 wasnt the fastest quadcore.

i7 880 was 3.06Ghz in 2010. In 2008 it was funny enough 3.2Ghz with C2Q.

Both were HEDT CPUs in todays terms though, not mainstream desktop. The i7 880+ was $583.

The frequency barrier was hit in 2004 at 3.8Ghz. 11 years after and we got +600Mhz.

Yeah, agree, and this is significant. 3.8 GHz + 600 MHz => 15.8% frequency increase. In 10 years(!).

15.8% is almost the percentage frequency increase we used to see per 2-3 months before desktop stagnated!

 

[ShintaiDK]

Power consumption also went up 13x in 10 years for that timeframe.

1700W CPUs if they continued? And I think the current ones use too much...

 

[Fjodor2001]

Power consumption also went up 13x in 10 years in that timeframe.

1700W CPUs? And I think the current ones use too much...

You're looping. See previous post about this.

At current performance/watt increase rate, we'd see 17000 W CPUs if the previous performance increase rate was to be kept up. That's how much crappier desktop CPU performance increase per watt is these days.

 

[shady28]

I suggest you read up more. I7 930 wasn't the fastest quad core at that time.

The fastest quad-core as of April 2010 was the I7 960 @ 3.2Ghz.

Not that it matters, since every I7 920 (released 2008) could clock to 3.4Ghz+ easily on air.

Right, I looked at the Q1 2010 intro for the i7-930 and assumed it was faster. It was actually a binning with tie i7-960 from Q4 2009 being the fastest.

So adjusting the numbers 4790k (4Ghz) vs 2600k (3.4Ghz) vs 960 (3.2Ghz) :

[36] Total Time
Of all time based benchmarks
(time in seconds)
Score in Time Go 1224.00 1678.00 1988.00

TDP is 88W / 95W / 130W.

So that's 62% performance increase and a 33% power decrease in 5 years. Core clock increase is 800Mhz or 25% (4Ghz vs 3.2Ghz).

 

[Innokentij]

Havent seen any improvement worth talking about for gaming since sandy bridge on max OC. Kind of OK with it to, havent upgraded the cpu/mobo/ram since either but changed gpu 4 times.

 

[Hulk]

In the "real world" people don't care about IPC.

If you are talking about desktop performance, compare the fastest 4-core desktop class CPU available in say April 2010 vs April 2011 vs April 2015.

That would be the i7-930 (2.8 Ghz) released in Feb 2010, the i7-2600k (3.4 / 3.8 Ghz) released in Jan 2011, and the i7-4790K (4.0 / 4.4Ghz) released in June 2014.

If you look at the clock speeds I just listed for those 3 years, it pretty much tears up your argument that only IPC matters.

By that I mean - in the "real world" - a consumer looking for a fast desktop PC would be looking at one that has a 42% higher clock rate today than what was available in 2010.

Comparison here between the i7-930, i7-2600k, and i7-4790k :

http://www.tomshardware.com/charts/cpu-charts-2015/compare,3712.html?prod[7418]=on&prod[7440]=on


The performance difference is a lot more than 20%.

This is i7-930 vs i7-2600k vs i7-4790k :

Total Time
Of all time based benchmarks
(time in seconds)
Score in Time Go 1224.00 1678.00 2238.00

If you don't like to discuss IPC that's fine. Besides looking at frequency, number of cores, and special instruction sets, I also like to investigate IPC. Especially with legacy code using a benchmark which has actually been applied to processors going back to 1989. It's just another metric to analyze Intel's progress. This one isolates one particular aspect of performance. And using this metric IPC advances over the last few generations has been less than 10%.

When you move outside of this outdated bench which fits in the cache you do get a better picture of overall core and memory subsystem performance. The link you provided doesn't equalize clocks, which as I stated is fine if you don't care for that data.

A while back I did a pretty extensive analysis of Anand's data when he tested Haswell against many older cores.

http://forums.anandtech.com/showthread.php?t=2324672&highlight=haswell

The summary is here:
Increase from Sandy Bridge to Haswell – 16.4%
Increase from Nehalem to Sandy Bridge – 14.6%
Increase from Conroe to Nehalem – 22.3%
Increase from Pentium 4 to Conroe – 50.7%

As you can see even going tock-to-tock only Conroe to Nehalem got past 20%. Definitely higher than 10% but for that number I was specifically referring to CPUmark99.

Keep in mind these results are with clocks equalized.

We shall see if Haswell to Skylake can crack 10% IPC improvement with CPUmark99. I'll be shocked if it does. And I'll be just as shocked if it can crack 20% IPC increase on actual apps not using instructions only in Skylake (cherry picked apps).

 

[Dave2150]

If you don't like to discuss IPC that's fine. Besides looking at frequency, number of cores, and special instruction sets, I also like to investigate IPC. Especially with legacy code using a benchmark which has actually been applied to processors going back to 1989. It's just another metric to analyze Intel's progress. This one isolates one particular aspect of performance. And using this metric IPC advances over the last few generations has been less than 10%.

When you move outside of this outdated bench which fits in the cache you do get a better picture of overall core and memory subsystem performance. The link you provided doesn't equalize clocks, which as I stated is fine if you don't care for that data.

A while back I did a pretty extensive analysis of Anand's data when he tested Haswell against many older cores.

http://forums.anandtech.com/showthread.php?t=2324672&highlight=haswell

The summary is here:
Increase from Sandy Bridge to Haswell – 16.4%
Increase from Nehalem to Sandy Bridge – 14.6%
Increase from Conroe to Nehalem – 22.3%
Increase from Pentium 4 to Conroe – 50.7%

As you can see even going tock-to-tock only Conroe to Nehalem got past 20%. Definitely higher than 10% but for that number I was specifically referring to CPUmark99.

Keep in mind these results are with clocks equalized.

We shall see if Haswell to Skylake can crack 10% IPC improvement with CPUmark99. I'll be shocked if it does. And I'll be just as shocked if it can crack 20% IPC increase on actual apps not using instructions only in Skylake (cherry picked apps).

I feel we'll be lucky to see Skylake this year for desktop. It will probably be delayed to 2016, 14nm process being crappy and all.

Remember, still no desktop 14nm CPU's have shipped yet.

Haswell launch = June 2013, we're coming up to 2 years on that now, with the same CPU's available. You'd almost think this was AMD I was talking about, with their ever shrinking budgets and inability to launch new desktop chips.

 

[ShintaiDK]

It shouldnt be any news that desktop is the last to a node. After server, mobile, tablet etc.

Dont expect to see Cannonlake for desktop either, but rather a direct go to Icelake.