Intel IPC: From Sandy Bridge to Broadwell

[tsk]

Anandtech has their full review of the Broadwell i7 5775C processor out, and for our reading pleasure they've tested the IPC increases, clock for clock, from Sandy Bridge to Broadwell.

 

Being able to do more with less, in the processor space, allows both the task to be completed quicker and often for less power. While the concept of having multiple cores has allowed many programs to be run at once, such as IM, web, compute and so forth, we are all still limited by the fact that a lot of software is still relying on one line of code after another, pegging each software package to once core unless it can exploit a mulithreaded list of operations. This is referred to as the serial part of the software, and is the basis for many early programming classes – getting the software to compile and complete is more important than speed. But the truth is that having a few fast cores helps more than several thousand super slow cores. This is where Instructions Per Clock (IPC) comes in to play.

TL;DR: 

Overall, a move from Sandy Bridge to Broadwell from an IPC perspective gives an average ~18.5% improvement.

Sandy Bridge to Ivy Bridge: Average ~4.8% Up

Ivy Bridge to Haswell: Average ~10.2% Up
Haswell to Broadwell: Average ~3.3% Up

NB! The averages exclude the WinRar benchmark as that test greatly benefited from Broadwells eDRAM cache.

So while the gains aren't huge there are improvements.
People usually complain that Intel haven't improved IPC enough from each processor generation, but I'd personally like to know what tasks these people want the IPC gains for. Very often that criticism seem to come from the gaming community.
The DX12/Vulkan gaming era is fast approaching and differences in gaming performance between processors will shrink.

On Wednesday a new generation processor from Intel will launch, this time it's a new architecture, which usually brings a bit more performance gains than a node shrink, but will Skylake make the people sitting on their trusty old Sandy Bridge platform want to upgrade? I don't think the processor itself will, but the Z170 chipset certainly look appealing.


Source


 

[Jlryan]

Not bad! my 3770k will last me long enough to save up for an X99 class board sometime in the next 2 - 3 years.

[i_build_nanosuits]

processor performance does increase over time...all the 5% and 10% from generations to generations does end up making everything better overtime.

i do wish skylake will be a 50% boost...that way i'll have a reason to upgrade from my 4770K...highly doubt it though !

[arronleeds]

Not bad! my 3770k will last me long enough to save up for an X99 class board sometime in the next 2 - 3 years.

i7 3770k is still a beastttt

[Fulgrim]

I don't care for benchmarks, unless they show real-world performance gains. I.e, gaming benchmarks. You'll probably only gain like ~5 - 7% cpu performance, at the same clock for gaming.

[tsk]

I don't care for benchmarks, unless they show real-world performance gains. I.e, gaming benchmarks. You'll probably only gain like ~5 - 7% cpu performance, at the same clock for gaming.

If you read the article they have gaming benchmarks as well.

[Fulgrim]

If you read the article they have gaming benchmarks as well.

And the only game which gets much of an improvement is GRID Autosport? Top kek, definatley not worth an upgrade.

[bomerr]

processor performance does increase over time...all the 5% and 10% from generations to generations does end up making everything better overtime.

i do wish skylake will be a 50% boost...that way i'll have a reason to upgrade from my 4770K...highly doubt it though !

even if it was a 50% boost there is still no reason to upgrade cuz GAMES ARE GPU BOUND.

 

 

And the only game which gets much of an improvement is GRID Autosport? Top kek, definatley not worth an upgrade.

 

That's cuz games are GPU BOUND. and as resolution goes up [i'm thinking 4k] the gpu bottleneck increases further. 

 

The one place where new CPUs have an advantage for games is emulators because those take advantage of newer instruction sets. 

[i_build_nanosuits]

even if it was a 50% boost there is still no reason to upgrade cuz GAMES ARE GPU BOUND.

That's cuz games are GPU BOUND. and as resolution goes up [i'm thinking 4k] the gpu bottleneck increases further. 

some people are not just gaming with their pc's you know?

[patrickjp93]

Ah Dolphin, an example of real programming at work. Being able to use SIMD really does help. There are some structural flaws holding it back, but for what it does, it's the best the world's currently got. Hooray for code multiversioning!

[bomerr]

some people are not just gaming with their pc's you know?

still no difference cuz if someone is doing rendering for instance than haswell-e > skylake cuz of the larger cache, quad channel ram and more cores. 

[i_build_nanosuits]

still no difference cuz if someone is doing rendering for instance than haswell-e > skylake cuz of the larger cache, quad channel ram and more cores. 

you're so funny...

[Curufinwe_wins]

Yea it's kinda amusing that dolphin is in many ways a gold standard. (Amusing from the perspective of someone that's used dolphin almost since its original release).

But it is....

[CTR640]

In the meanwhile I'll keep my i7-4770 for the next 5-10 years if this bullshit keeps going with those pathetic % increasements.

I can't believe this shit Intel is just doing nothing but controlling the CPU market. Wtf AMD, seriously, please do something to get Intel their ass in gear!

[patrickjp93]

you're so funny...

He's right actually. Unless Skylake S were to ship with AVX 512 and rendering software had new patches to take advantage of it. Then at that point the advantage would swing back to Skylake assuming you had no bandwidth bottleneck at dual channel 3000MHz or something else extreme like that.

[i_build_nanosuits]

He's right actually. Unless Skylake S were to ship with AVX 512 and rendering software had new patches to take advantage of it. Then at that point the advantage would swing back to Skylake assuming you had no bandwidth bottleneck at dual channel 3000MHz or something else extreme like that.

My point was a skylake hyper-threaded quad core setup will cost much less than an X99 one, and will perform better with games and stuff...and is alright for the rendering and workloads that i do...my 4770K is perfectly adequate for what i do with my system and so would an  i7-6770K...

[patrickjp93]

In the meanwhile I'll keep my i7-4770 for the next 5-10 years if this bullshit keeps going with those pathetic % increasements.

I can't believe this shit Intel is just doing nothing but controlling the CPU market. Wtf AMD, seriously, please do something to get Intel their ass in gear!

Do your research better. Intel has led the charge like no other chip designer in performance improvements and like no other singular group in pushing the evolution of computing. It was the first company or group to create a framework for multithreading (OpenMP in 2000), even before multicore chips existed. Intel basically invented SIMD to get around the fact serial processing, even with an Out of Order Superscalar engine, was choked by the statistical reality every 7-9 lines of code is an if-statement or entrance to a loop in most software. It was also the first company to include an integrated graphics chip both non-programmable and programmable into its CPUs to allow heterogeneous acceleration. Programmers in consumer space have not kept up.

 

The fault of small gains lies on the shoulders of software companies and microsoft. Intel has done 90% of the work and has worked on all facets of the problem. And people forget Intel has to drive its own sales even in the absence of AMD's competition in order to keep growth up, so it's not as though Intel isn't motivated to do better. The truth is Intel is a victim of its own success and Microsoft's monopolistic hold on desktop environments.

[patrickjp93]

My point was a skylake hyper-threaded quad core setup will cost much less than an X99 one, and will perform better with games and stuff...and is alright for the rendering and workloads that i do...my 4770K is perfectly adequate for what i do with my system and so would an  i7-6770K...

Except it wouldn't. The 5820K and 4790K have been evenly priced for quite some time in most parts of the world. And a 4x4 DDR4 RAM kit costs marginally more than a 2x8 kit, sometimes marginally less. The motherboards have an intersection space with Z97 as well where the low-mid X99 boards are at price parity with the mid-high end boards of Z97. You have no leg to stand on with that argument.

[i_build_nanosuits]

Except it wouldn't. The 5820K and 4790K have been evenly priced for quite some time in most parts of the world. And a 4x4 DDR4 RAM kit costs marginally more than a 2x8 kit, sometimes marginally less. The motherboards have an intersection space with Z97 as well where the low-mid X99 boards are at price parity with the mid-high end boards of Z97. You have no leg to stand on with that argument.

ok, so you're saying that skylake has no reasons to exist until skylake-E if i understand you correctly intel are morons for pushing out this new architecture?!

[patrickjp93]

ok, so you're saying that skylake has no reasons to exist until skylake-E if i understand you correctly intel are morons for pushing out this new architecture?!

Not at all. Skylake will catch those wanting to move up from Core 2 and Nehalem, maybe even some from Sandy Bridge. I'm just saying Skylake-S is not going to beat out Haswell-E for a number of workloads, but that it could if it came with AVX 512. At the same clocks, 6 cores x 256 bits per clock = 1536 bits churned per clock. 4 cores x 512 bits per clock = 2048 bits churned per clock, not to mention the quads tend to have the better clocks. And currently we don't know if the E3 Xeons will come with AVX 512 or not.

[i_build_nanosuits]

Not at all. Skylake will catch those wanting to move up from Core 2 and Nehalem, maybe even some from Sandy Bridge. I'm just saying Skylake-S is not going to beat out Haswell-E for a number of workloads, but that it could if it came with AVX 512. At the same clocks, 6 cores x 256 bits per clock = 1536 bits churned per clock. 4 cores x 512 bits per clock = 2048 bits churned per clock, not to mention the quads tend to have the better clocks. And currently we don't know if the E3 Xeons will come with AVX 512 or not.

too much maths for me son

anyways i had no plans to upgrade from my 4770K, it's still way fast and way strong for my needs anyways...i'm mostly gaming on this machine with the occasional video rendering and stuff but nothing too long to perform...i will throw in a next gen GPU in it and game at 1440p long before i change this CPU.

[patrickjp93]

too much maths for me son

anyways i had no plans to upgrade from my 4770K, it's still way fast and way strong for my needs anyways...i'm mostly gaming on this machine with the occasional video rendering and stuff but nothing too long to perform...i will throw in a next gen GPU in it and game at 1440p long before i change this CPU.

Cool. I don't see why people want to have to spend money to upgrade every 2 years. My dad made a Q6600 last nearly 8 years before moving up to the 4790K back in January. I moved up from a Q9550 to a 3820 when I took up programming to practice with multithreaded scaling at higher bandwidth. I expect that will tide me over until Kaby Lake-E or Cannonlake-E. I take pride in building a machine that will last me that long. Sure my Titan Blacks aren't great for AAA gaming with new tessellation standards, but I barely game anymore, and I can get 60+ fps on 1440P easily.