First "real" Haswell CPU preview, 4670K

[boxleitnerb]

http://www.pcgameshardware.de/Haswe...sts/Intel-Haswell-Core-i5-4670K-Test-1070973/

It's in German, but the numbers in the table are universal, I guess.
In CPU bottlenecks the IPC was a good 10% higher for the 4670K vs the 3570K. In Starcraft 2 the very good result for Haswell could be explained by increased cache bandwidth, but that probably is a special case.

 

[AtenRa]

Why reviewers bench the iGPU at 1080p when its clearly inadequate for that resolution and then they bench at 720p with a dGPU ???

 

[SPBHM]

Why reviewers bench the iGPU at 1080p when its clearly inadequate for that resolution and then they bench at 720p with a dGPU ???

I partially agree, but I think using 720p for CPU comparison (with a fast VGAs) is a good thing, it helps to reduce to GPU bottleneck, but using IGPs at 15FPS 1080p medium/high is kind of a waste of time... also test those games at 720p low please, it's probably going to be much closer to relevance (playable settings, 30FPS or something), but the 1080p data could be useful to compare the IGPs to faster cards and see the benefits, or reduce the CPU impact to a minimum.


anyway, decent gain.

 

[mikk]

Why reviewers bench the iGPU at 1080p when its clearly inadequate for that resolution and then they bench at 720p with a dGPU ???

CPU tests in 720p make sense to make sure it runs fully CPU limited. I agree for the iGPU Benchmarks, they are clearly unplayable.

 

[boxleitnerb]

Agreed with the iGPU thing. Unplayable. But they wanted to make sure to have a GPU bottleneck with the iGPU and a CPU bottleneck when testing the CPU part. So it makes sense, especially if you just want to find out the performance improvement over the predecessor.

I get 18.6 fps in the x264 benchmark with a 3930K@4.2 GHz, hyperthreading off.
Fits perfectly. If Haswell overclocks better than my Sandy-E, I will replace it.

 

[BallaTheFeared]

Looks like a decent upgrade for anyone still on SB or later hardware.

 

[aaksheytalwar]

Correction: SB or older hardware.

 

[Kallogan]

15% over Ivy and 30% over Sandy in x264 bench is not bad, guess it worth the upgrade for Sandy users. Depends if u want to pay a premium price for 30% and noticeable better gaming...

 

[AtenRa]

and noticeable better gaming...

At 720p ???

 

[boxleitnerb]

At 720p ???

Will you finally give it a rest? It get's really really old your constant 720p whining. That is how one benchmarks CPUs, period.
1080p or higher doesn't make your CPUs faster! If you need GPU benchmarks, look for separate ones. You cannot condense gaming performance into one single benchmark that tells you about both, CPU and GPU. Games can be CPU and GPU bottlenecked depending on your system, on the scene and your preferred fps (via your selected quality settings). You test at 720p to make sure to exclude a GPU bottleneck. With some games that isn't necessary, but with others it is. Since you cannot test one game at 1440p, the other at 1080p and a third one at 720p, you choose the smallest common denominator that makes sense, and that is 720p.

 

[BallaTheFeared]

At 720p ???

When CPU limited, min fps, RTS, MMO, anything with a hard single thread demand, 120Hz, MGPU, high end graphics performance.

Interestingly, there is more and more frame time variance creeping up into the results with the GTX 780 and the GTX Titan. This is due to the surge in CPU bottlenecks that occur when you add more and more GPU horsepower to the system. NVIDIA's frame metering technology can really only alleviate the issue when the GPU is the primary bottleneck, so when the other components cause the problem NVIDIA is mostly helpless.

 

[SPBHM]

At 720p ???

VGAs are getting faster all the time,
you are not locked to "Ultra 4K AA16x" on games.
tests like this can't show how CPU bound the game can get on its entirety, my money would be on the fastest CPU on this 720P test to perform better always (when needed, when CPU is the main limiting factor for one reason or another, which can happen you know...) testing CPUs on games without making an effort to avoid GPU bottleneck is not ideal.

 

[AtenRa]

Will you finally give it a rest? It get's really really old your constant 720p whining. That is how one benchmarks CPUs, period.

I dont have a problem benching at 720p, my problem is that they dont bench at 1080p as well. When people only see 720p they conclude that they will see the same gaming performance gains at 1080p (or above) as well.

So, instead of whining about me, start demanding from reviewers to also bench at 1080p.

 

[BallaTheFeared]

Is this really the thread for your pro AMD agenda?

Most of us are aware of how cpu's work, we don't need to be told higher gpu load reduces cpu demand.

 

[boxleitnerb]

I dont have a problem benching at 720p, my problem is that they dont bench at 1080p as well. When people only see 720p they conclude that they will see the same gaming performance gains at 1080p (or above) as well.

So, instead of whining about me, start demanding from reviewers to also bench at 1080p.

Then people are stupid, plain and simple. Because you cannot draw that conclusion since the gains are dependent on many things the reviewer has no influence over (your personal fps preference for example).
To bench at 1080p is a waste of time. It is 100x faster to check out some separate GPU benchmark and see for yourself what kind of fps you can expect with your graphics card(s). One doesn't need a reviewer for that. People should start to think for themselves and don't be so lazy. Two different types of benchmarks for two different components. Should be self-evident.

 

[AtenRa]

Most of us

Most of the people in the world doesnt know that. The reviews are not only meant for you and me but for every one with an access to World Wide Web.

It is highly misleading and they(reviewers) should also include 1080p benchmarks.

 

[BallaTheFeared]

Actually showcasing cpu performance when the bottleneck is the gpu is highly misleading.

 

[Kallogan]

At 720p ???

Ok barely noticeable better gaming then

A little boost for 768p notebooks with decent dGpu.

 

[Haserath]

Most of the people in the world doesnt know that. The reviews are not only meant for you and me but for every one with an access to World Wide Web.

It is highly misleading and they(reviewers) should also include 1080p benchmarks.

What GPU do you recommend for testing?

 

[omeds]

Looks promising coming from SB imo. Currently CPU bottlenecked in games like BF3 with a 2700k at 5GHz. Fingers crossed 4770k can oc to near the same levels.

 

[boxleitnerb]

For strategy games, simulations and high fps in many shooters (100+ fps in BF3 MP or Serious Sam 3 for example) a fast CPU is paramount. One can always reduce GPU load if needed, that's easy. CPU load on the other hand cannot be as easily influenced. That's why for many people a fast CPU is important, independent of the GPU(s) they have.

 

[inf64]

Haswell in impressive uarchitecture on paper, there is absolutely no doubt about it. Intel did everything they could within the TDP limits in order to improve the performance. The actual gains in legacy software are not that impressive though. Design itself has a lot of potential, it can be further optimized for power or performance and intel is not stopping there. The problem intel faces is diminishing gains in legacy softwer. AMD can attack this weakness in two ways: they can increase sheer performance of their own x86 cores since the room to grow there is bigger (vs intel which hit the wall more or less) and they have huge potential for significant performance jumps in their "HSA" initiative. Interesting times ahead.

 

[24601]

I'd wager most of the gains were from longer Turbo timers.

Sites need to do testing like this at fixed clock speeds so that overclockers have a frame of reference.

 

[Sweepr]

Very impressive gains vs Ivy Bridge, which is ~4% faster than SB at equal clocks. These benchmarks are showing the same kind of performance jump clock per clock vs SB that SB brought over Nehalem (if not more?). Cant wait to see Anand's reviews. Heres hoping that it overclocks at least ~5-10% higher than IB on air.
When you think that it takes a huge 4M/8C PD (~2x quad-core IB+GT2 die size; without an IGP) to match a quad-core IB with in MT tasks it makes you wonder how AMD is going to keep up with a ~10% faster per clock Haswell with future APUs topping out at 3M/6C and no successor to the FX8300 line. The magical 30% IPC jump claims better be true, otherwise... not very interesting times ahead, we'll be stuck with quad-cores + IGP in mainstream for years to come while servers offer 12+ core CPUs at <130W TDPs.

 

[ShintaiDK]

I'd wager most of the gains were from longer Turbo timers.

Sites need to do testing like this at fixed clock speeds so that overclockers have a frame of reference.

You never tried to see how turbo works on your PC I guess. Even linpack, the most power demanding thing you can run, gets full turbo all the time.