First complete review of Haswell i7-4770K

[CHADBOGA]

They can but will they, talk realistically here because we're not talking about the enthusiast market anymore !

If AMD manages to get within 85% of Intel's single core performance, then it would seem to be a trivial matter to go from 4 cores to 6 cores, to ensure that AMD don't take substantial market share off Intel.

But it depends on whether a supposed catch up chip from AMD has an IGP in it, if it doesn't, then Intel could always abandon the high pricing of their current 6 core, IGP-less CPU's and keep AMD well contained.

How would AMD's CPU's with IGP's in them(APU's), catch up to Intel on CPU performance, any time soon? They are way behind, and it is not feasible to expect them to catch up any time soon.

Regardless of the above, with you quoting Idon'tcare, he seemed to be suggesting that if Intel were going to battle AMD's IGP-less FX range with Haswell and its successors, then and only then might AMD catch up to Intel, but that would be because they have twice the cores.

Thus the obvious response from Intel is to either increase core count in their CPU's that have IGP's, or lower the prices of their 6 core IGP-less CPU's.

Both options are easy to imagine.

 

[mikk]

They can but will they, talk realistically here because we're not talking about the enthusiast market anymore !

They will definitely do that at some point. You don't know when. My bet is Skylake. (Haswell-E 8 cores for High-End, Skylake-DT 6 cores fo the upper midrange).

 

[Cerb]

I know what you are all saying, that TDP is thermal power and not electrical power. But via the law of conservation of energy, there is no way a CPU consuming 200 W only puts out 125 W of heat. Where would you propose the rest of that energy went?

It's not like that. The CPU may use 10W, then get a higher load and use 30W, then a higher load and use 80W. Mobile CPUs have enough power conservation going on that they will be doing this sort of thing tens of times a second, visibly, and internally clock-gating at granularities of tens of cycles.

Now, if you define the thermal load by typical consumption, it may be possible to exceed the rating, as happened with some P4 CPUs (this is not new territory for Intel). So now, your cooling design, made for 100W, may not be good enough, long-term, for a 115W CPU. It may only increase failures by 1-2%, but that matters. More likely, it would decrease the lifespan of nearby capacitors, thus increasing the chances of death by blown cap, ripple damaging components, or a small surge that would have otherwise been dampened.

So, you can't be doing that. TDP needs to be the maximum draw that you can make happen, with the worst chip bin you will sell as that model, over a length of time that may matter to the cooling system (Watts are always an average over time). The cooling system must be able to handle the TDP, though actual power consumption during use will typically be much less than that rating (of course, many notebook makers ignore TDP, to save money on the coolers, leading to thermal throttling and shutdown if you try to game on a cheap 'gaming' laptop ). A CPU with a TDP of 125W may only use 40W for you to browse Youtube, FI.

 

[R0H1T]

If AMD manages to get within 85% of Intel's single core performance, then it would seem to be a trivial matter to go from 4 cores to 6 cores, to ensure that AMD don't take substantial market share off Intel.

But it depends on whether a supposed catch up chip from AMD has an IGP in it, if it doesn't, then Intel could always abandon the high pricing of their current 6 core, IGP-less CPU's and keep AMD well contained.

How would AMD's CPU's with IGP's in them(APU's), catch up to Intel on CPU performance, any time soon? They are way behind, and it is not feasible to expect them to catch up any time soon.

Regardless of the above, with you quoting Idon'tcare, he seemed to be suggesting that if Intel were going to battle AMD's IGP-less FX range with Haswell and its successors, then and only then might AMD catch up to Intel, but that would be because they have twice the cores.

Thus the obvious response from Intel is to either increase core count in their CPU's that have IGP's, or lower the prices of their 6 core IGP-less CPU's.

Both options are easy to imagine.

Yes but only one of them will see the light of day & I suspect it to be the latter one, in which case AMD might release their own true hexa/octa core variants provided they don't exceed the current 3970X TDP levels.

 

[CHADBOGA]

AMD might release their own true hexa/octa core variants provided they don't exceed the current 3970X TDP levels.

Not sure what you mean by true hexa/octa core variants?

Even if you don't regard AMD's Piledriver as a true hexa/octa core, changing to a "true" version, is surely too big an architectural change, and one that would have to wait till their next generation architectural base comes around and that probably won't be before 2016.

 

[Idontcare]

Used in switching the transistors

The switching just generates heat (power-consumption) as well.

Thermodynamically it is a closed system, power in equals power out, no different than your resistive space heater plugged into an outlet.

^ that Pdynamic you see above captures the power consumed in the switching of the transistors.

It is, thermodynamically speaking, no different than you switching on and off a light switch a few billion times per second. Sure some of the energy is momentarily converted into photons, the rest being immediate waste heat (phonon) generation, but the photons also nearly immediately (the speed of light and the distance across the room) become absorbed and converted into phonons (heat) themselves.

The only time this doesn't happen is when some of the energy is temporarily captured and stored as chemical potential energy or kinetic energy. But even then it is just a matter of time, usually on the order of nano to microseconds, before that potential energy then finds its way into becoming heat (phonons or IR photons).

I understand that probably doesn't sit well with some people, but it is the reality as far as science has been able to determine for the past hundred years or so.

 

[R0H1T]

Not sure what you mean by true hexa/octa core variants?

Even if you don't regard AMD's Piledriver as a true hexa/octa core, changing to a "true" version, is surely too big an architectural change, and one that would have to wait till their next generation architectural base comes around and that probably won't be before 2016.

By true I meant six/eight modules & twelve/sixteen cores, going by their marketing team's nomenclature

 

[Face2Face]

If anyone is interested -

 

[zebrax2]

The switching just generates heat (power-consumption) as well.

Thermodynamically it is a closed system, power in equals power out, no different than your resistive space heater plugged into an outlet.

The proc send data to other parts though which means energy leaving out of the system

 

[CHADBOGA]

By true I meant six/eight modules & twelve/sixteen cores, going by their marketing team's nomenclature

How big is this CPU going to be and will there be enough memory bandwidth?

 

[mikk]

New test: http://diy.pconline.com.cn/329/3297240.html

Overall:

System Power consumption test looks improved over Ivy Bridge:

 

[A5]

The switching just generates heat (power-consumption) as well.

Thermodynamically it is a closed system, power in equals power out, no different than your resistive space heater plugged into an outlet.

^ that Pdynamic you see above captures the power consumed in the switching of the transistors.

It is, thermodynamically speaking, no different than you switching on and off a light switch a few billion times per second. Sure some of the energy is momentarily converted into photons, the rest being immediate waste heat (phonon) generation, but the photons also nearly immediately (the speed of light and the distance across the room) become absorbed and converted into phonons (heat) themselves.

The only time this doesn't happen is when some of the energy is temporarily captured and stored as chemical potential energy or kinetic energy. But even then it is just a matter of time, usually on the order of nano to microseconds, before that potential energy then finds its way into becoming heat (phonons or IR photons).

I understand that probably doesn't sit well with some people, but it is the reality as far as science has been able to determine for the past hundred years or so.

Yeah, exactly all of this.

 

[A5]

If anyone is interested -

Seems about what was expected. Hopefully MC will still do their discounts

 

[Remobz]

New test: http://diy.pconline.com.cn/329/3297240.html

Overall:

System Power consumption test looks improved over Ivy Bridge:

Translation please....

 

[moonbogg]

Yeah, exactly all of this.

LOL. Ya, I concur with all of that stuff as well. I mean, it looks like math but its got letters in it, so he made some mistakes cause math don't got no letters but it still looks good to me.

 

[ShintaiDK]

New test: http://diy.pconline.com.cn/329/3297240.html

Looks very nice.

 

[ShintaiDK]

Translation please....

7% faster overall, lower power consumption.

 

[ShintaiDK]

Their testplatform with dGPU gave 28W less in idle and 14W less in load vs 3770K. Using IGP it was 27W idle and 5W load, favouring the Haswell CPU.

Incredible platform advances.

IGP wise it looks like this:

And thats only with GT2. Looks like GT3 and GT3e will be quite the performers.

Platform power summed up:

 

[Idontcare]

The proc send data to other parts though which means energy leaving out of the system

You are conflating entropy with heat and power consumption.

Not trying to be rude but have you had the opportunity to take a college level thermo course?

I'm guessing the answer is "no" and am attempting to dial in my response without needlessly going over your head...but if your answer is "yes" then I won't waste your time (or mine) any further as your college professors ought to have already properly brought you up to speed on how electrical work (i.e. non-PV work) manifests in the laws of thermo.

Data is simply a change in entropy, and depending on the span of time you wish to capture in your closed-loop system the data itself is meta-stable (your hard-drive is not thermodynamically stable, it ought to be metal oxides and it will be metal oxides again at some point in time) so all things end up as heat.

If you want to talk about instantaneous snapshots in time in which you can momentarily and arbitrarily claim "but my electrical system has performed work!" then no one will tell you that you are wrong, we will just point out that owing to your selective chronology the system itself is no longer at equilibrium and as such is not thermodynamically stable...meaning your claims of closed-loop thermo consistency are null and void the moment you said "but..."

 

[Racan]

New test: http://diy.pconline.com.cn/329/3297240.html

Overall:

A "Tick" being more of a performance improvement than a "Tock"

Not impressed

 

[mikk]

IGP wise it looks like this:

And thats only with GT2. Looks like GT3 and GT3e will be quite the performers.

HD4600 efficiency much better than HD4000, very very nice indeed.

A "Tick" being more of a performance improvement than a "Tock"

Not impressed

Not really, i7-2700k is faster than i7-2600k but they didn't test this.

 

[Khato]

Somewhat unfortunate that they didn't have any memory bandwidth measurements as I suspect that the previous reviews had a slight issue in that respect. Regardless, looks much more like expected.

Going to be even more awesome to see what happens on workloads where the new HSW features are actually used.

 

[blackened23]

I know people will chide the 4770k for the negligible IPC improvement, but this was expected - we knew from SB to IB that intel wasn't going for all out performance with each iteration, they were designing the uarch for efficiency. Personally i'm impressed - 4770k has a fairly substantial efficiency improvement even for the desktop part - which has a higher TDP than the 3770k. Count me as impressed. This means that the mobile version will be much better than the IB in terms of efficiency.

Now the wildcard that will clinch a purchase for me is the overclocking ability - as long as it remains fairly similar or perhaps slightly better than the Ivy Bridge in this respect, count me in for a 4770k. And count me in for a Haswell rMBP whenever that is released.

 

[IntelUser2000]

The point is that TDP does NOT mean maximum power draw of a processor. It doesn't even really mean "average" power draw.

Maximum power = no
Practical maximum power = yes

Go and read the datasheets. It says there that TDP can be exceeded, but the maximum is only ~1 minute or so depending on the SKU, and even customizable and the CPU will over longer period of time make sure that it stays at or under TDP.

Because the power manager limits the long term power use as TDP, hence practical maximum power = TDP

And that's also relevant to thermal design as you need to dissipate that heat away.

You also can't compare AMD and Intel TDP definitions, they aren't the same. Intel plays with thermal power definitions a lot, I mean, recently they invented a totally new term with its own weird definition so they could claim they have 7W processors.

LOL, bring up SDP in a TDP conversation.

 

[blackened23]

LOL, bring up SDP in a TDP conversation.

First, he didn't say anything about SDP. Look, this argument is just becoming is stupid and irrelevant to the topic. We corrected you in that TDP does not equal power consumption. Furthermore, every silicon company has different definitions of TDP internally, intel, AMD, nvidia, EVERYONE, has a different way to define TDP. EVERY- SILICON - FIRM has a different definition of TDP.

You're being beyond silly and stubborn at this point to argue. You already acknowledged that TDP isn't power consumption, i'm not sure why you want to argue so badly when you already stated this. I'm pretty sure everyone you're arguing with probably agrees with your viewpoint on FX chips, me included - However facts are facts. Can we talk about the 4770k again now?