First complete review of Haswell i7-4770K

[NTMBK]

Helped by Turbo.

And the 3850 wasnt the top bin like the 5800K. Nor did Llano get any process advantages to the node that had happend since then.

Oh come on, the 3870k was 3.0GHz whereas the 3850k was 2.9GHz... I think we can guess how that 3% different would help. I'm just using the benchmarks I have, I'm afraid. Interestingly the only benchmarks we have comparing it to the 3870k are the ones for power consumption, which I think quite nicely demonstrate that the "Piledriver is more power hungry than K10" line is a bit of a myth:

Also, outside of that one benchmark the 5800k pretty consistently beats the 3850. http://www.anandtech.com/bench/Product/675?vs=399 The ones where it struggles is kind of predictable- the ones where all four threads are doing shedloads of floating point maths.

 

[Enigmoid]

Ahmdahl's law fails the moment you run two tasks at the same time. My Task manager tells me I run close to 70.

Just saying, Ahmdahl's law is a gross oversimplification and is missquoted more often than Moore's law.

Thats not really what ahmdahl's law deals with. One process can only be spilt among so many cores so efficiently.

You can already see the effects of it in games such as crysis 3. Look at the 8350, from the 4300 to the 8350 core utilization goes down despite the 8350 being cpu limited (and still having a fair amount of spare resources).

 

[OBLAMA2009]

the 5800k isnt even twice as fast as the five year old dual core pentium im on right now, thats pretty sad....

 

[lopri]

Is this supposed to be a tick or a tock?


Posted from Anandtech.com App for Android

 

[lopri]

I will never understand why, on one of the premiere review sites on the Internet, people are constantly posting random reviews from random sites done by random people with random track records. Or worse, giving them any credence.

lol


Posted from Anandtech.com App for Android

 

[Idontcare]

Ahmdahl's law fails the moment you run two tasks at the same time. My Task manager tells me I run close to 70.

Just saying, Ahmdahl's law is a gross oversimplification and is missquoted more often than Moore's law.

How can something "fail" when you reference a usage scenario for which it simply doesn't even apply?

Amdahl's law has absolutely nothing to do with running discrete programs in parallel.

Talk about irony and misquoting. (you don't even spell it correctly )

And Amdahl's law is not a gross oversimplification, it is an upper-limit that cannot be exceeded by hardware improvements, to get above the Amdahl limit you must invest in software improvements.

There are people who understand this, and then there are those who don't.

 

[Piroko]

How can something "fail" when you reference a usage scenario for which it simply doesn't even apply?

Amdahl's law has absolutely nothing to do with running discrete programs in parallel.

Talk about irony and misquoting. (you don't even spell it correctly )

And Amdahl's law is not a gross oversimplification, it is an upper-limit that cannot be exceeded by hardware improvements, to get above the Amdahl limit you must invest in software improvements.

Wrote a way too long reply but deleted it. In a theoretical play you're absolutely right, but I don't see how one can draw sensible consumer advice out of it, considering that we're talking about two to eight cores, unknown Windows configurations and complex programs.

Btw my main reason to register here instead of a german board was to improve my english. I'm doing fairly well imho.

 

[mikk]

That is 3dmark . We know from recent review of 4770K (retail part) how well it does in 3dmark and how bad it does (vs GT630 that performs "worse" in 3dmark) in actual games, I expect Richland 6700(65W) to just walk away from HD4600 when it comes to actual fps in games .

Just walk away

 

[sontin]

So, after they lost the high-end desktop and server market to Intel, now they will lose the mid-range market, too?

Wow. And they only needed two years, since Llano, for this...

 

[blackened23]

So, after they lost the high-end desktop and server market to Intel, now they will lose the mid-range market, too?

Wow. And they only needed two years, since Llano, for this...

lol

No comment.

 

[AtenRa]

Just walk away

At what settings ??? where is the link ??

 

[galego]

i7-4770k on Ebay at $600

http://www.tweaktown.com/news/30456/intel-core-i7-4770k-reaches-ebay/index.html

Lists of Haswell-compatible power supplies:

http://www.corsair.com/us/blog/haswell-compatibility-with-corsair-power-supplies/

http://www.xbitlabs.com/news/cases/..._i_Haswell_Compatible_Power_Supply_Units.html

http://www.enermax.com/home.php?fn=eng/press_b_1&no=46

The difference between 4 and 8 cores in games will be very minimal.

That is trivially true for older games using only 2--4 threads, but is not for the next gen.

Average FPS for Crysis 3:

FX-4300 (4C @ 3.8 GHz): 36.3
FX-8350 (8C @ 4.0 GHz): 58.6

A 53% gain is not what I consider very minimal. And Crysis 3 is very far from being optimal code (legacy issues).

Just saying, Ahmdahl's law is a gross oversimplification and is missquoted more often than Moore's law.

It applies to some situations but not at others. In particular, using Amdahl's Law as an argument against massively parallel processing is not valid

http://www.futurechips.org/thoughts-for-researchers/parallel-programming-gene-amdahl-said.html

http://spartan.cis.temple.edu/shi/public_html/docs/amdahl/amdahl.html

 

[blackened23]

i7-4770k on Ebay at $600

So what are you saying here? Pre-release prices are always inflated? Pretty sure you're well aware of what the actual cost will be at release, slightly more than 300$.

A 53% gain is not what I consider very minimal. And Crysis 3 is very far from being optimal code (legacy issues).

[citation required].

 

[Charles Kozierok]

It applies to some situations and fails at others. In particular, using Amdahl's Law as an argument against massively parallel processing is not valid

That doesn't mean it's "failing", just that it's inapplicable.

Amdahl's law is not specifically about parallelism at all.

 

[Idontcare]

It applies to some situations and fails at others. In particular, using Amdahl's Law as an argument against massively parallel processing is not valid

http://www.futurechips.org/thoughts-for-researchers/parallel-programming-gene-amdahl-said.html

http://spartan.cis.temple.edu/shi/public_html/docs/amdahl/amdahl.html

Neither of those articles do justice to Amdahl's Law, it is no wonder the conclusions lend themselves to further incorrect assertions.

Amdahl's Law has never been "an argument against massively parallel processing"...that is a strawman position concocted to then dispute and refute Amdahl's Law on a falsely asserted premises.

Amdahl's Law is a limit, it tells you where you must take your code, at a minimum, if you desire to have a given speedup on your computations.

No different than computing the theoretical IPC of a microarchitecture, knowing full well that it represent an absolute maximum (a limit) in the absence of a host of things that will result in actual IPC being lower than the theoretical IPC.

You will mostly never get better scaling than that of the limit of Amdahl's Law (niche corner-case of cache-resident apps aside), but you can get much worse scaling if you have a poorly performing network topology relative to the granularity of your code.

 

[Duvie]

Haswell is supposed to have 10-15% improved IPC over IB. These results look in line with that, and should not be surprising.

Some of the other anomolous results are completely within the margin of error - anyone expecting amazing increases with Haswell should have known the anticipated IPC increase. 10-15%, same as SB to IB.

Efficiency and graphics are the important metrics today, and Haswell should absolutely deliver on those fronts. Anyway, for desktop users the real wildcard will be overclocking, and we just don't have good data on that. I'd suggest waiting on reviews from reputable websites for that, the CPU reviews here at AT are actually always good.

Brown noser!!!! Lol

I agree though

 

[galego]

That doesn't mean it's "failing", just that it's inapplicable.

Right. Edited.

[citation required].

Citation for what? For the claim about the 53% gain on Crysis 3? Obtained from here

Citation for Crysis 3 being not optimized for 8 threads? Look at the load core percentages given some posts above.

 

[Sheep221]

It's not worth upgrading. If you still run 2500K, don't even bother to buy it as you won't just get sufficient upgrade. I'm curious about the overclocking capabilities of 4770K tho :biggrin:
But it's funny, that CPUs that debuted in January 2011, yes that's almost 2 and half years ago and there is still nothing really replacing them. But it's good at least I can keep my 2550K for at least 10 years! In fact through all of the 2000s the one just didn't have time to buy newer and faster hardware and now, nothing:awe:

 

[Khato]

That is trivially true for older games using only 2--4 threads, but is not for the next gen.

Average FPS for Crysis 3:

FX-4300 (4C @ 3.8 GHz): 36.3
FX-8350 (8C @ 4.0 GHz): 58.6

A 53% gain is not what I consider very minimal. And Crysis 3 is very far from being optimal code (legacy issues).

You do know why some Crysis 3 benchmarks show such massive gains with more cores right? Whereas others show pretty much no difference between the FX-4300 and the FX-8350? It's really simple, I even stated such in a previous reply to this thread - it's real easy to chew up all the extra x86 cores you have available with physics calculations. Never mind that doing such a parallel task on the GPU is both more efficient and has far more processing power available... no, they don't do that because it's difficult to code for comparatively.

Which brings up an interesting question - which future is it going to be? Because the future isn't fusion in gaming if all the non-graphics parallel workloads are being handled by the extra CPU cores that AMD's so fond of providing. But providing extra CPU cores and giving them "multi-threaded" workloads is the only way AMD can get comparable performance to Intel.

I wonder what the "Welcome to the Jungle" Crysis 3 benchmark would look like for CPU scaling if they'd bothered to use OpenCL for the physics calculations? Especially if said OpenCL was run on the integrated graphics leaving the discrete GPU free to handle the graphics. Bet that the dual core i3 would beat the FX-8350 then. That's what I've been waiting for ever since real integrated graphics started being put into the CPU... But yeah, good luck getting programmers to take that route.

 

[blackened23]

Citation for Crysis 3 being not optimized for 8 threads? Look at the load core percentages given some posts above.

Correlation != causation. You haven't proven a case that Crysis 3 is old or un-optimized code, and i'm actually extremely doubtful that you can.

 

[guskline]

Good evening galego. You have time to post to this thread. Take a few seconds and post the specs of your pc (you have one don't you?)

 

[Grooveriding]

i7-4770k on Ebay at $600

http://www.tweaktown.com/news/30456/intel-core-i7-4770k-reaches-ebay/index.html

A friend of mine already has his 4770K sitting in its box, cost him something like $380 I think. Useless as there are no motherboards available, even via the same route he got the CPU he cannot get a moob yet.

Buying that for $600 would be worthless to get it early considering you couldn't even use it without a mobo.

 

[inf64]

@ mikk

So you showed us that 1 year old 5800K has noticeably better iGPU than yet to launch 4770K with HD4600 inside. Richland will offer +15% so yes it will walk away (as I mentioned Richland in my post ). 1.15x1.07=1.23 or 23% better performance on average for just clock bumped Trinity (which is what Richland is). There will be games where Trinity's +20% will turn into +40% on Richland. AMD is positioned nicely on desktop I have to say.

 

[WhoBeDaPlaya]

Good evening galego. You have time to post to this thread. Take a few seconds and post the specs of your pc (you have one don't you?)

Nah, he's posting from his ARM tablet

 

[frozentundra123456]

@ mikk

So you showed us that 1 year old 5800K has noticeably better iGPU than yet to launch 4770K with HD4600 inside. Richland will offer +15% so yes it will walk away (as I mentioned Richland in my post ). 1.15x1.07=1.23 or 23% better performance on average for just clock bumped Trinity (which is what Richland is). There will be games where Trinity's +20% will turn into +40% on Richland. AMD is positioned nicely on desktop I have to say.

If the average is 20 percent different, you cant suddenly make that 40% in one game without some other game then being equal. Otherwise the math does not add up.

That aside, who cares anyway, any of the 3 igps will still be lower in performance than a HD7750, so you are still in limbo of better than good enough for non-demanding uses, but barely adequate (at best) for gaming.

I am somewhat disappointed in the lack of cpu improvement in Haswell, but I could care less about the igp.