How much has CPU performance increased the last 4 years?

[ForkIt]

I recently had an interesting conversation with an IT professor who claimed that CPU performance barely increased the last 3-4 years.

So I was wondering: Is that actually true? How much better (in percent) did performance get since 09/20/10 if we just compared the most powerful CPU from there and from today?

And how about the average performance increase per day considering only the most high end CPU that was available per date, by how much did it decelerate over the years? 

 

I know that factors like hyperthreading/more cores/core clock speed/Lx caches etc are factors that are not easily compared in benchmarking, but maybe you have a nice idea on how to still compare old technologies with new ones?

[79wjd]

About 5-10% with each generation since the first gen Intel core series. 

[WereCat]

Core to core performance... not by much

[harrynowl]

Nehalem-> Sandy was ~15%

 

Sandy -> Ivy was ~8%

 

Ivy -> Haswell was ~10%

 

Keep in mind there's other things, such as a generational jump in iGPU performance and AVX2 in Haswell along with doubled floating point bitwidth etc.

[Sonefiler]

About 36.95 %

[AlwaysFSX]

Over the period of time, roughly 35%.

[Si3Rra_7]

Nehalem-> Sandy was ~15%

 

Sandy -> Ivy was ~8%

 

Ivy -> Haswell was ~10%

Right answer here

[Manny Calavera]

About 15.5125 %

nah thats way too conservative

 

Over the period of time, roughly 35%.

not looking at numbers, that should be about right

[AlwaysFSX]

nah thats way too conservative

 

not looking at numbers, that should be about right

"Best case scenario"

[Sonefiler]

nah thats way too

 

not looking at numbers, that should be about right

conservative

I fixed it right before you posted

[AnnoyedShelf]

I recently had an interesting conversation with an IT professor who claimed that CPU performance barely increased the last 3-4 years.

So I was wondering: Is that actually true? How much better (in percent) did performance get since 09/20/10 if we just compared the most powerful CPU from there and from today?

And how about the average performance increase per day considering only the most high end CPU that was available per date, by how much did it decelerate over the years? 

 

I know that factors like hyperthreading/more cores/core clock speed/Lx caches etc are factors that are not easily compared in benchmarking, but maybe you have a nice idea on how to still compare old technologies with new ones?

I'd say quite a bit, considering my laptop outpaces my 5-year old desktop.

[Manny Calavera]

I fixed it right before you posted

geez, and on two decimal places! we are talking now! anyone can get the third decimal right?

[ForkIt]

Nehalem-> Sandy was ~15%

 

Sandy -> Ivy was ~8%

 

Ivy -> Haswell was ~10%

When was each of these technologies first available for consumers?

[79wjd]

When was each of these technologies first available for consumers?

Nehalem was 2010~ 

Sandy 2011~ 

Ivy 2012~ 

Haswell 2013~ 

Broadwell 2015~ 

[xNeon]

No, it's not true. For productivity, quite a bit. For gaming, not a lot but I'm pretty positive your professor isn't a gamer. If you look at benchmarks, you can see a finite improvement, but if you USE CPUs that you've upgraded to, sometimes it feels an order of magnitude better. I think it's cool that you aren't blindly accepting your professors words by the way.

[Glenwing]

Since Sandy Bridge (2011) there's been barely any change.

[Burdock]

Comparing my old 17-920 to the i7-4770..... I had a 30-80% performance improvement o.0 

I guess 35% sounds about right, but there are architecture, software, and driver, improvements to take into account.

[That Norwegian Guy]

Nehalem-> Sandy was ~15%

 

Sandy -> Ivy was ~8%

 

Ivy -> Haswell was ~10%

 

Keep in mind there's other things, such as a generational jump in iGPU performance and AVX2 in Haswell along with doubled floating point bitwidth etc.

 

IPC advantage of Haswell over Ivy was about 7.2%, (according to passmark aggregate results since then) and the Ivy overclocks on average at least 10% better, so you can say the improvement in Haswell was -3%.

Sandy -> Ivy it was 8-9% IPC difference, and Sandy overclocked on average 5-7% better, so that was the only real improvement that happened.

 

#IvyBridgeMasterRace

[harrynowl]

IPC advantage of Haswell over Ivy was about 7.2%, (according to passmark aggregate results since then) and the Ivy overclocks on average at least 10% better, so you can say the improvement in Haswell was -3%.

Sandy -> Ivy it was 8-9% IPC difference, and Sandy overclocked on average 5-7% better, so that was the only real improvement that happened.

 

#IvyBridgeMasterRace

measuring generational cpu performance with overclocked numbers is a flawed logic

[HippY]

take a look at it this way: back in 2008 (ish) I had an E4600 2 core at 2.4 GHz. Now, my phone in my pocket is 4 cores at 2. something GHz.

Yes, I know different architecture and stuff, but try to concentrate on anything else than the stagnating consumer market, thanks that there are no real high end competition. 

 

http://www.evga.com/articles/00537/ 4 years ago it took as 2 high end server grade, $1666 per piece CPU to get 24 threads

Now you can buy for $1500 a motherboard and a cpu, that has 16 core, overclockable and consumer friendly

 

CPU-s in general did go a large way, but the consumer versions are stagnating for a reason. C`mon AMD, show us something!

[Jey R.]

Core to core performance... not by much

Correct. My i7 920 is still giving it's all in all today's application. Only thing is I have to overclock to compensate. But in terms of use performance it's still up there.

The thing is, most developers are still coding and building by the least common denominator which is two cores.

[harrynowl]

When was each of these technologies first available for consumers?

Nehalem was introduced in 2008 on the 45nm process. But it was used until 2010 on the 32nm Westmere core.

 

Sandy 2011, Ivy 2012, Haswell 2013, Broadwell/Skylake 2015, Cannonlake 2016

[Manny Calavera]

 the Ivy overclocks on average at least 10% better, so you can say the improvement in Haswell was -3%.

yes, haswell is not even an upgrade. not even from sandy.

and haswell refresh is like lowest form of upgrade ever +5% better overclock

[That Norwegian Guy]

measuring generational cpu performance with overclocked numbers is a flawed logic

 

I wasn't. Comparing just 3570K to 4670K both at 3.4Ghz shows a difference of 7.2% of performance. That is due to IPC only.

 

But you can't ignore the fact that the Ivy Bridge chip will beat the Haswell when both are overclocked. You can't exactly ignore overclocking when these chips are MEANT to be overclocked, and marketed as unlocked chips. THAT would be flawed logic.

 

TLDR Ivy Bridge unlocked chips are better, fact.

[harrynowl]

I wasn't. Comparing just 3570K to 4670K both at 3.4Ghz shows a difference of 7.2% of performance. That is due to IPC only.

 

But you can't ignore the fact that the Ivy Bridge chip will beat the Haswell when both are overclocked. You can't exactly ignore overclocking when these chips are MEANT to be overclocked, and marketed as unlocked chips. THAT would be flawed logic.

 

TLDR Ivy Bridge unlocked chips are better, fact.

You said you could say Haswell was a -3% improvement. I said comparing overclocked results is flawed logic. It is. It's like comparing the power/weight ratio of cars and throwing in a load of modified car results.

 

I can ignore overclocking as the OP said "How much has CPU performance increased in the past 4 years" not "How much has overclocking performance impacted IPC gap over the last 4 years".

 

OK so lets compare overclocked numbers. I'm going to use HWBot averages as my weapon of choice as that's the biggest OC database I'm aware of.

 

3770k: 4719MHz

4770k: 4470MHz

4790k: 4613MHz

 

So with an IPC gap on average of 7.2%, let's compare the 3770k and the 4770k to start with. We have the OC of 4470MHz, so lets times that by 1.072 and we have 4791.84, so the IPC gap between the two makes up the ~7% Ivy oc potential (so not "at least 10%"). So Ivy doesn't win in this scenario.

 

Now we can compare the Haswell-Refresh 4790K as that's the Haswell-K most people will be buying today, 4613*1.072 is 4945MHz. So we see again that Ivys average overclock does not make up the IPC gap by the 3% number you stated.

 

Now let's compare some results from the LTT community itself. The highest scoring 3770K is 918 at 5.1GHz, this is below about 15 Haswell i7 results, we have a 4770K @ 4.633 beating the 5.1GHz 3770K by a whopping 1.3%, however 4.6GHz whilst uncommon on a 4770K is much more common than getting a 5.1GHz 3770k. With Devil's Canyon i7 average OC being 4.6 that's fairly conclusive.

 

TLDR Check numbers before you say something is fact.