Anandtech Haswell review is up!

[willomz]

Hang on, how comes they have Richland data already?

 

[Idontcare]

My two cents to the discussion:

- It seems a chip built from the scratch for mobile. It packs a lot of punch, both in graphics and CPU power, all that in a power sipping power envelope. In

I have a SB-based laptop now, and IB-based desktops (plus FX8350) now...Haswell makes me want to upgrade my laptop. It seems perfect for that marketspace, which for the consumer demographic is >50% marketshare of units shipped so it really is not rocket science that Intel had Haswell focus on the mobile aspects.

- I'm kinda impressed by the top iris part. 70-80% of the GT650 + Quad Core Haswell in a 47W power envelope.It will be nice for very small form factors.

It is impressive performance per watt. I think we are seeing the tip of an iceberg here in terms of what Intel thinks their future will entail.

Crystalwell with its MCM approach to getting L4 cache into the socket reminds me so much of Clarkdale which was their first foray into getting the IGP on-die (first step was getting it on-package, same as crystalwell and its L4$).

It would not surprise me at all if Broadwell had a SKU which included a true on-die L4$.

 

[AtenRa]

Trinity is about 30% faster, Richland performance increase is negligible. Note that these were tested on high settings (where hostorically intel doesn't do well) where almost every single game tested was unplayable even at 1280 x 800.

That is one of the worst reviews i have seen.

 

[AtenRa]

Hang on, how comes they have Richland data already?

Yea, i was wandering that as well. The NDA is not over yet.

 

[monstercameron]

what is the effect of not having to pass through the pcie bus on an igp?

 

[mikk]

It would not surprise me at all if Broadwell had a SKU which included a true on-die L4$.

I would be surprised.

I asked about the potential to integrate eDRAM on-die, but was told that it’s far too early for that discussion. Given the size of the 128MB eDRAM on 22nm (~84mm^2), I can understand why. Intel did float an interesting idea by me though. In the future it could integrate 16 - 32MB of eDRAM on-die for specific use cases (e.g. storing the frame buffer).

At least not with 128MB, this is too big. Broadwell should be very interesting nevertheless. Intel told Broadwell will improve graphics by a bigger margin than Haswell.

 

[Shephard]

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

True but on so many forums I read Sandy Bridge and Ivy bridge users being told to wait for the glorious Haswell. Not even that good... Even the high original core series is still excellent.

 

[Idontcare]

I would be surprised.


At least not with 128MB, this is too big. Broadwell should be very interesting nevertheless. Intel told Broadwell will improve graphics by a bigger margin than Haswell.

42mm^2 die-adder at 14nm is not all that much. Besides, any argument one would make for why they'd keep the 128MB cache a discrete IC using MCM could have been equally applied to the discrete iGPU IC that was fabricated with 45nm...and yet they went ahead and put it all on-die anyways.

Intel has always used MCM packaging as its stepping stone for transitioning to an all-on-die solution. Just look at the history of its MCM products, it has always been a gateway product to putting it all on-die. Why would they stop now?

Especially with something that scales stupidly well (in both yields and die-size) as cache.

 

[jpiniero]

It would not surprise me at all if Broadwell had a SKU which included a true on-die L4$.

I think next is going to be the PCH actually. On Package on all Broadwell models and ondie on Skylake.

 

[mikk]

42mm^2 die-adder at 14nm is not all that much. Besides, any argument one would make for why they'd keep the 128MB cache a discrete IC using MCM could have been equally applied to the discrete iGPU IC that was fabricated with 45nm...and yet they went ahead and put it all on-die anyways.

Not sure if they shrink it to 14nm with Broadwell. I would expect 22nm to be much cheaper initially, makes sense Anands impression. But who knows...

I get the impression that the plan might be to keep the eDRAM on a n-1 process going forward. When Intel moves to 14nm with Broadwell, it’s entirely possible that Crystalwell will remain at 22nm.

 

[mrmt]

I have a SB-based laptop now, and IB-based desktops (plus FX8350) now...Haswell makes me want to upgrade my laptop. It seems perfect for that marketspace, which for the consumer demographic is >50% marketshare of units shipped so it really is not rocket science that Intel had Haswell focus on the mobile aspects.

Same here. I have a SNB laptop, if I can find an Iris convertible part, I'll pay the price.

Crystalwell with its MCM approach to getting L4 cache into the socket reminds me so much of Clarkdale which was their first foray into getting the IGP on-die (first step was getting it on-package, same as crystalwell and its L4$).

I wouldn't be able to say if Intel is going for on-die L4 or for even bigger L4 sizes outside the die. Think about it, what could a 512MB or 1GB part could do to GPU and GPGPU?

 

[jiffylube1024]

42mm^2 die-adder at 14nm is not all that much. Besides, any argument one would make for why they'd keep the 128MB cache a discrete IC using MCM could have been equally applied to the discrete iGPU IC that was fabricated with 45nm...and yet they went ahead and put it all on-die anyways.

Intel has always used MCM packaging as its stepping stone for transitioning to an all-on-die solution. Just look at the history of its MCM products, it has always been a gateway product to putting it all on-die. Why would they stop now?

Especially with something that scales stupidly well (in both yields and die-size) as cache.

Anand implied in his review that the L4 cache memory may not get a die shrink to 14nm right away, as they will have unused 22nm capacity.

 

[azhangvroom]

Any rumours about the upcoming macbook air and its battery life?

 

[Abwx]

Link? Based on all Ivy vs Sandy comparisons I have process scaling was 1,5.

And did you provide a link before throwing your 1.5 ratio.?.

I got the number from SB/IB comparisons.....

 

[mikk]

And did you provide a link before throwing your 1.5 ratio.?.

I got the number from SB/IB comparisons.....

Measured from several die photos from Sandy/Ivy. Ivy and Sandy CPU is almost identical, it is not that hard to calculate the process scaling. You have to isolate the CPU cores.

 

[Enigmoid]

That is one of the worst reviews i have seen.

I agree its stupid that settings where nothing was playable were chosen but its the only review that I could find which the relevant information.

It does show two things

1) Richland for a10-6800k vs a10-5800k = big yawn
2) Haswell is about 35% better than Ivy (with newest drivers, ivy also received a boost with newer drivers).

What we don't know is how the scaling will hold at playable settings.

 

[Arkaign]

It also shows that at least for the rest of 2013, iGPU will continue to be basically worthless for gaming, unless you're going for a dirt cheap notebook to play stripped settings on, or Angry Birds.

 

[allenk09]

I think this would be a good time for current C2D, Phenom II or Core i7 first gen owners to upgrade. 2nd and 3rd gen owners can probaly skip this gen from what I am seeing. I am the only guy that is tired of going from a quad to another quad then another quad

I have a Phenom II X6 and just bought a 4770k today. Should be quite an upgrade.

 

[Lepton87]

It also shows that at least for the rest of 2013, iGPU will continue to be basically worthless for gaming, unless you're going for a dirt cheap notebook to play stripped settings on, or Angry Birds.

Agreed, I consider my overclocked 660M to be the bare minimum to play games. but maybe some games like D3 or SC2 or some CPU-Limited MMOs will be playable on Iris.

 

[Aikouka]

I think I've mentioned this somewhere before, but one of the reasons why I'd like to upgrade is because I'd like to be able to take advantage of faster encoding. Now, I'm not a person that encodes tons of video, but when I do encode videos, I prefer it not to take long. Right now, I have an i5-2500k on a P67 motherboard, so I have no support for QuickSync given the P-series boards don't have access to the iGPU. I could just buy a Z77 board and be done with it, but it seems like a bit of a waste to upgrade "old tech."

Anyway, that brings me to my question! In Ganesh's look at the iGPU for HTPC purposes, he talks about QuickSync and how it isn't that good on Haswell compared to Ivy Bridge:
http://www.anandtech.com/show/7007/intels-haswell-an-htpc-perspective/8

So, I guess I'm wondering... what's up, doc? Is this a software problem, so I could purchase now and expect it to be fixed later (not always the smartest play), or is this some hardware-related problem? From what it looks like, I'd still have a good boost just using x264 with the new AVX2 extensions, and better video quality.

 

[cytg111]

Hmm .. how come it runs hot?

Bigger die than IVB, same process node
Lesser TDP

The chip should output fewer watts and have a larger die size to do so .. By those means alone and other things being equal, it should run cooler than equivalent IVB

 

[WhoBeDaPlaya]

Lesser TDP

The chip should output fewer watts and have a larger die size to do so .. By those means alone and other things being equal, it should run cooler than equivalent IVB

Other way around because of the integrated VRM and upgraded iGPU.

 

[Magic Carpet]

Is the iGPU used as a "cooler" when disabled? Will the iGPU-less parts run cooler?

 

[R0H1T]

Is the iGPU used as a "cooler" when disabled?

Nope

Will the iGPU-less parts run cooler?

There will be minimal difference, if anything at all in the first place.

 

[Magic Carpet]

Nope There will be minimal difference, if anything at all in the first place.

Ah, that sucks. Wouldn't that be cool.