First "real" Haswell CPU preview, 4670K

[24601]

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.

Reviews mention that turbo timers are different between Sandy Bridge, Ivy Bridge, and Haswell.

I'm just taking their word for it.

 

[Yuriman]

So if we're looking at 30% over SB (in some games), and 2011 chips have 50% more cores, you would need a 15% clock advantage to match a Sandy Bridge hex with a Haswell quad.

That is to say, a 4.2ghz 3930k would be perfectly matched in things that can fully utilize all 6 cores by a 4.8ghz Haswell quad, and would be 50% faster in things that can only utilize 4 cores or less.

Let's hope Haswell brings in a bit of overclocking headroom.

 

[Mallibu]

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

Because you see the real differences in performance for GPUs, in GPU bottlenecked scenarios, and the same for CPUs.
You would test a car in an open road, wouldn't you?
Testing CPUs in GPU bottlenecked scenarios, hides CPU weakness, and would result in an i3 3220 beeing the same as an i5 in some scenarios.
Same with AMD cpus, look in how much games the FX 4** * is the same as 8***.
CPUs provide consistent fps despite resolution change, so if an i7 provides ~80 fps at 720p, I know what I'm getting in every resolution from my CPU. And that's what I want, when I'm reading a CPU review.

 

[frozentundra123456]

While I totally disagree with using gpu limited benchmarks to evaluate CPU performance, I would have liked to see a wider range of tests. That is use a mid range and a very high end card to test a series of settings from very CPU bound to gpu bound. Then the user could decide which settings are most relavent to his use. Hopefully we will be seeing these complete tests soon. The results do look somewhat hopeful though.

 

[ilogik]

So would it be worth for me to upgrade from a i7-940 to this? PC is used mainly for gaming with GTX 680.

 

[frozentundra123456]

So would it be worth for me to upgrade from a i7-940 to this? PC is used mainly for gaming with GTX 680.

I would wait for more complete tests, including overclocking results. Is your current cpu overclocked? Are you cpu limited in the games you are playing now? Do you have money to burn??

Kind of joking about the last question, but overall I am sure there would be improvement, how much and if it is worth it depends on a lot of factors, including your own price/performance metric, what kind of overclock you have on the 940, what games you play.

 

[AtenRa]

What GPU do you recommend for testing?

Any High-End from HD7950/GTX670 and up will do.

 

[BallaTheFeared]

And knowing there will be little change how does that help me extrapolate the benefit of a newer, faster cpu in situations where my current one is limiting me?

If I don't know how much faster a cpu is via testing which creates the greatest cpu limitation, how am I ever going to apply that to my needs?

 

[AtenRa]

Testing CPUs in GPU bottlenecked scenarios, hides CPU weakness, and would result in an i3 3220 beeing the same as an i5 in some scenarios.
Same with AMD cpus, look in how much games the FX 4** * is the same as 8***.

And thats exactly what i would like to know, if the CPU will grand me any gains at the resolution I and most of the people play and thats at 1080p. Bench the games at 720p but also show us if we will get any gains at 1080p as well(obviously not with low settings).

Would you consider upgrading a CPU+Motherboard if you would know that you will only gain a 2% at 1080p ??

 

[AtenRa]

And knowing there will be little change how does that help me extrapolate the benefit of a newer, faster cpu in situations where my current one is limiting me?

If I don't know how much faster a cpu is via testing which creates the greatest cpu limitation, how am I ever going to apply that to my needs?

If your needs are more FPS at 1080p and a faster CPU at 720p doesn't provide you with any gains at 1080p then you would know that you will need a better GPU and not a faster CPU. If the game will show any significant gains at 1080p then you and I and everyone else will know that for that game a CPU upgrade will be beneficial.
But if you only bench at 720p, you will not have any data to know what happens at 1080p.

 

[Khato]

Huh, somewhat disappointing for there to be a 'preview' today as I was hoping that Intel would do a repeat of Ivy Bridge and allow reviews to be posted a week before availability. (Ivy Bridge reviews went live on April 23rd with availability on the 29th.) Granted that could still be the case, just would be somewhat odd to have a preview so close to an actual review.

 

[frozentundra123456]

Here we go again, the endless debate about cpu gaming benchmarks. I agree that both benchmarks are needed, as they tell you different things. Problem is when one benchmark is used to show something that it doesnt.

GPU limited benchmarks show only one thing. At the settings tested all the cpus that have the same framerate are "good enough" to use the gpu in the test to the max at the settings being tested in that particular game. If you used a more powerful gpu or a more cpu limited game, the results most likely would be different. The problem comes when one tries to extrapolate that because cpu A is equal to cpu B in one game at one gpu limited setting, that they are equal as gaming cpus.

Low resolution benchmarks show more clearly the difference between cpu performance, but the same relative differences obviously will not be maintained as the gpu becomes the limiting factor. I would say however, that if one cpu shows higher performance at low resolution than a different cpu of the same architecture it can pretty safely be extrapolated that it will ultimately allow higher performance with a powerful graphics card. This is the usefulness of low resolution tests. The user must determine if his ultimate use will be gpu or cpu limited.

If only one test system is available, I would prefer to see a very high end card, with tests done at lower settings to evaluate cpus and with increasingly more demanding settings to see where the gpu becomes limiting. Then one could extrapolate that data to get some idea of gpu vs cpu limitations with other cards.

 

[exar333]

I really want to see what the TDP is when the CPU is at 100% compared to SB and IB. This is not including the iGPU, because I don't really care about that. The overall TDP is up, but I really want to see what the cores in HW can do vs. SB and IB. That will hopefully tell us more about what performance to expect/watt.

 

[Abwx]

I really want to see what the TDP is when the CPU is at 100% compared to SB and IB.

15-25% less ; depending of the loading , iddle power
being reduced even more in %age.

 

[frozentundra123456]

I really want to see what the TDP is when the CPU is at 100% compared to SB and IB. This is not including the iGPU, because I don't really care about that. The overall TDP is up, but I really want to see what the cores in HW can do vs. SB and IB. That will hopefully tell us more about what performance to expect/watt.

On the desktop, I would rather see increased clocks than power savings, within reason of course. Granted this is more meaningful in mobile cpus.

 

[WhoBeDaPlaya]

On the desktop, I would rather see increased clocks than power savings, within reason of course. Granted this is more meaningful in mobile cpus.

Not a chance. Intel has caught tablet/smarphone fever.
I've consigned myself to IPC stagnation and will just bite the bullet on HW since I need a new rig this summer, but won't bother upgrading for a good long while.

 

[escrow4]

Spent $2K on a 3930K/Titan box a few days back. Still not interested. Haswell is much more a tock for mobile than performance.

 

[piasabird]

What happened to playing games at max resolution? May as well have a TV for a monitor.

 

[BTRY B 529th FA BN]

Glad my only upgrade will be the CPU & Motherboard.

 

[StinkyPinky]

Not bad but I'l probably wait until broadwell to replace my 2500k

 

[galego]

Conclusion: wait about a 1-5% gain in gaming performance under more realistic setups.

 

[BallaTheFeared]

Conclusion: When you're cpu limited expect up to 20% more performance.

 

[NTMBK]

I'm waiting for a proper 4770k review. I'm very curious about how hyperthreading does in Haswell- it has a wider core, so in theory Hyperthreading should scale better than it does in IB. If multithreaded performance is up, it would bode well for the Haswell i3.

 

[frozentundra123456]

Conclusion: wait about a 1-5% gain in gaming performance under more realistic setups.

Maybe you should upgrade. What are your specs again?

 

[frozentundra123456]

Conclusion: When you're cpu limited expect up to 20% more performance.

The Tom's article looked hopeful, while that chinese article seemed to show no gain. Could be they were gpu limited. Clearly we need more reviews with a wide variety of games and GPUs to really know what kind of improvements are coming.