Anandtech Haswell review is up!

[toyota]

testing games on lowest setting to compare cpus is asinine. low res sure but many settings impact the cpu not just the gpu so a game needs to be tested on highest setting except for AA.

 

[Ketchup]

Haswell looks like the biggest disappoint ever!

Nah, it'd be tough to beat those first P4's, which were actually slower that the P3's they replaced!

 

[Ken g6]

At last, I can post about Haswell!

I'm confused. I've seen two sites (AnandTech and HotHardware) that say the BCLK dividers won't be available on non-K CPUs:

The BCLK adjustment gives you a little more flexibility when overclocking, but you still need a K-SKU to take advantage of the options.

And one (Xbit labs) that said they will be:

It also means that even LGA1150 CPUs with locked multiplier can be overclocked, but only by 25% or 66% above the default frequency.

HotHardware also indicates there might be a fourth divider that does 250%.

 

[frozentundra123456]

Do you have a source?


Here is GT3e: http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested


Big differences to A10-4600M.

Problem I see is the GT3e, while a very nice performance gain, and quite a bit ahead of AMDs current mobile lineup and undoubtedly Richland also, but it is an expensive chip and will almost surely be paired with a 1080p or above screen, so current games will still mostly not be playable at native resolution.

 

[Homeles]

And one (Xbit labs) that said they will be:HotHardware also indicates there might be a fourth divider that does 250%.

I'm pretty sure the guys that are saying that it's unlocked on non-K models are misinformed. XBitLabs isn't exactly the pinnacle of hardware journalism, either.

 

[SiliconWars]

Performance comes at a premium. It's not exactly a secret.

The only thing it's lacking is performance then.

 

[notty22]

The confusion is probably from some sites using ES samples. Which usually are unlocked in all ways. I'd go with they 'won't have' the straps available unless being K part as Anand stated.

 

[Homeles]

The only thing it's lacking is performance then.

Yeah, you're right. Those AMD processors sure are underwhelming.

You just don't get it, do you? Thread crapping will not be tolerated. Please enjoy this extended length vacation
-ViRGE

 

[SiliconWars]

Yeah, you're right. Those AMD processors sure are underwhelming.

I'm sorry what?

 

[frozentundra123456]

The only thing it's lacking is performance then.

Not sure what you really mean by "lacking in performance". Are you talking about cpu or igpu performance?

Anyway, technically I would not say it is "lacking in performance", since in nearly every application it is the fastest mainstream processor. I am actually fairly satisfied with the ipc increases, just wish they would have increased clocks or added a hex core.

I think the days of large IPC increases, especially for intel, which is very efficient already, are over. They need to add more cores or increase clockspeed in spite of the extra power it would use.

 

[cytg111]

.. And K models contain TSX.

And there we have it

 

[SunRe]

Now why exactly are these parts consuming more than IvyBridge? Only because of the integrated voltage regulators?

This baffles me, you do gain some performance over Ivy, but it's no way more efficient. I really hoped otherwise considering all the hype that Ivy was built on a non-optimized 22nm process and Haswell will rock on a greatly improved one.

This is a strange release, I'm not convinced until I see the numbers for the mobile platform with regards to battery life. Right now, these parts are less efficient than the previous ones:

 

[Imouto]

Microsoft and Intel are trying to kill the PC market pretty hard.

Now I wonder what the hell is Intel going to do with the 14nm node shrink. It's crystal clear that temps is a huge issue that it is unable to fix.

 

[USER8000]

Yeah about $300 on the chip alone. D:

It is meant to cost $650 according to the Hardware Canucks editor:

http://www.xtremesystems.org/forums...well-Reviews&p=5190977&viewfull=1#post5190977

Not surprising,as the total package is around 350MM2 in size and Intel wants to maintain its margins too. It seems 65% of the GT3 4C die is for the IGP.

 

[Ventanni]

Not sure what you really mean by "lacking in performance". Are you talking about cpu or igpu performance?

Anyway, technically I would not say it is "lacking in performance", since in nearly every application it is the fastest mainstream processor. I am actually fairly satisfied with the ipc increases, just wish they would have increased clocks or added a hex core.

I think the days of large IPC increases, especially for intel, which is very efficient already, are over. They need to add more cores or increase clockspeed in spite of the extra power it would use.

I agree. I wouldn't expect any kind of major increase in IPC from Intel until a brand new architecture is introduced. Haswell is, after all, a continuation of the Core lineup. Anand's page on the five generations of CPUs compared is a very interesting look at the performance evolution of the Core architecture. As we all can see, some of the early performance gains are clock speed based, but I'd really love to see an article detailing the IPC evolution from Core 2 to Haswell.

http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6

I love AMD, but man, they get pretty beat up in a lot of these performance benchmarks. Even if Steamroller is a modest 20% IPC improvement, they still have a long, long ways to go to catch up to Intel.

 

[IntelUser2000]

Now why exactly are these parts consuming more than IvyBridge? Only because of the integrated voltage regulators?

Well, the TDP is higher on the 4770K.

Only the U and Y series platforms come with the greatly improved power management system. Rest will get the gain you usually get from generation to generation.

 

[Ventanni]

Now why exactly are these parts consuming more than IvyBridge? Only because of the integrated voltage regulators?

This baffles me, you do gain some performance over Ivy, but it's no way more efficient. I really hoped otherwise considering all the hype that Ivy was built on a non-optimized 22nm process and Haswell will rock on a greatly improved one.

This is a strange release, I'm not convinced until I see the numbers for the mobile platform with regards to battery life. Right now, these parts are less efficient than the previous ones:

TDP limitations. The 4770k is an 84w part, while the 3770k is a 77w part. May not seem like much, but the difference you're seeing is that Haswell is allowed to draw more power by design. And, as a result, it does.

Interesting observation though. If we reduced the 4770k to a 77w TDP like the 3770k, how would it perform?

 

[jpiniero]

Now why exactly are these parts consuming more than IvyBridge? Only because of the integrated voltage regulators?

"Intel doesn't care about desktop people"

 

[Ken g6]

I agree. I wouldn't expect any kind of major increase in IPC from Intel until a brand new architecture is introduced. Haswell is, after all, a continuation of the Core lineup. Anand's page on the five generations of CPUs compared is a very interesting look at the performance evolution of the Core architecture. As we all can see, some of the early performance gains are clock speed based, but I'd really love to see an article detailing the IPC evolution from Core 2 to Haswell.

http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6

I love AMD, but man, they get pretty beat up in a lot of these performance benchmarks. Even if Steamroller is a modest 20% IPC improvement, they still have a long, long ways to go to catch up to Intel.

Actually, I think Core can trace its ancestry back through the Pentium M, the PII/III, and back to the Pentium Pro. I don't really expect Intel to introduce another high-performance architecture unless AMD shocks everyone again. After all, Intel's last new architecture was Atom.

 

[WhoBeDaPlaya]

Need bottom line answer to :

1) Whether non-K SKUs will support BCLK OCing (wording in AT review is a little vague)
2) Whether K SKUs will have TSX

 

[oceanside]

I really don't see the point of a K sku anymore. Approx 500-600mHz to play with, elevated temps, reduced instruction set... and we have to pay even more for it. Looks like a big "eff ewe" from Intel to the OC crowd. If we eliminate the K series, it's a decent release with modest but average increases. I feel a bit let down however.

 

[notty22]

Need bottom line answer to :

1) Whether non-K SKUs will support BCLK OCing (wording in AT review is a little vague)
2) Whether K SKUs will have TSX

2 http://ark.intel.com/products/75123/Intel-Core-i7-4770K-Processor-8M-Cache-up-to-3_90-GHz

4th Generation Intel® Core™ i7 Processors

 

[jpiniero]

Need bottom line answer to :
1) Whether non-K SKUs will support BCLK OCing (wording in AT review is a little vague)

Agree that it's vague, but it makes sense that only the K models would only have it.

Whether K SKUs will have TSX

It doesn't. But there's no need to ask, look it up on Ark for yourself.

 

[Sweepr]

10-11% faster than IB at equal clocks according to Hardware.fr. 12-13% IPC improvement over IB according to Hardware Canucks. Wow, actually a little bit better than I expected. If we consider that IB is ~4% faster than SB we're looking at ~15% better performance per clock from SB to Haswell, about the same (if not a little better) generation-to-generation improvement we saw from Nehalem -> Sandy Bridge. The i5 4670K (without HT) is often competitive with AMD's 315mm² 4M/8C FX8350 in MT tasks, surprising.

 

[SunRe]

Interesting observation though. If we reduced the 4770k to a 77w TDP like the 3770k, how would it perform?

Well, I bet it will be even less efficient.

I really, really hope that AVX2 and TXT and the other enhancements really start making a difference in future versions of productivity software and that Haswell really is that great on mobile, otherwise this is a dud.

One would have hopped that Intel worked this through and make sure that the increased power consumption of the new graphics and integrated VR's is compensated by IPC. But it's obviously not the case. Furthermore, these parts are getting more and more difficult to overclock. And I don't see 14nm improving on this in any way.. Power leakage will only go up.