First complete review of Haswell i7-4770K

[ShintaiDK]

Chipsets != processors.

But yeah, I may well be wrong and we may see desktop Broadwell. I just don't think we've had proper confirmation either way yet.

There was never any doubt until a "techsite" decided it needed some extra clicks from people seeking sensationalism.

 

[mikk]

Broadwell is not mobile only. We also got H97 and Z97 chipsets confirmed. So lets not keep spinning the fear^H^H^H^H rumour mill.

New chipsets is not a confirmation.

 

[ThePeasant]

The kind where 256/32 = 8, one was stuck with scalar x87 or scalar SSE2, and is lucky enough to be able to process a whole cache line without interdependence, and arrange the data to be read in one cache line, and use adds and/or multiplies a lot. Hence why it's a range. You're not going to flip a compiler switch and get an 8x speedup, but that's about where a potential speedup would top out, for already-tight loops, though there is more potential for int cases, due to fewer instructions, registers, and operands needed for the front-end to keep track. Theoretically, Haswell can also perform twice as fast as Ivy Bridge, when it comes to vectors, so its not all just from AVX2 itself, though AVX2 will be needed to take full advantage of it.

I don't know why you bring up x87 when you mentioned SSE2 vs AVX2. I'm still waiting to hear the cases where AVX2 can give you a 20x speedup over General purpose x86 and 8x over SSE2.

 

[AtenRa]

3D Mark 11

• AMD A10-6700 “Radeon HD 8670D” - P1508
• Intel Core i5-4670K “HD 4600 GPU” - P1285
• Intel Core i5-3570K “HD 4000 GPU” - P748

http://wccftech.com/intel-hd-4600-graphics/#ixzz2THGqf7K5
​

 

[Cerb]

I don't know why you bring up x87 when you mentioned SSE2 vs AVX2. I'm still waiting to hear the cases where AVX2 can give you a 20x speedup over General purpose x86 and 8x over SSE2.

32-bit to 256-bit (potentially) is an 8x (for 64-bit ints, only 4x). Haswell+AVX2 offers a 256-bit integer multiply, add, shift, etc., which standard x86 does not have. Then there's a more nebulous potential from gather, which will still need some tuning (but with sufficiently high-level code in languages like C++, maybe not much), but not require either several loads or packed operands, ideally (FI, load several red values from an array of pixels as packed RGB values, with a single instruction, which should also mean only a single load per cache line). Also, FMA can offer a 2x speedup, to any code currently doing FMUL+FADD, and reduce register writes and reads in the process.

Add to that that some of the speedups are Haswell v. IVB, not merely AVX2. FI, HSW has 2 FMAs, allowing 2x w/ AVX for FADD or FMUL, or 4x v. AVX for FMA cases, on top of AVX2 itself as an ISA extension, whereas IVB has 1 FMUL and 1 FADD, and no FMA. I'm not sure about scheduling with SSE, though, so it may only offer the real benefits for AVX and AVX2.

For code that already works with SSE2 or AVX, there may be a very low upper limit (2x-4x, plus some for gather, just improvements from gather, or nothing). For code that could be vectorized in theory, and works well on other ISAs that way, but hasn't been on x86, it depends on how pessimisal the scalar x86 version is, in comparison to how efficiently an AVX2 implementation can be, and/or how easily it can be made to use AVX2 well. With OOOE and speculative loads and stores, pinning a single broad value down is all but impossible--"at least," and, "up to," are good as it will get until the chips become widespread, and applications can be tested.

x87 and SSE2, because SSE2 is often faster for scalar FP than x87, due to x87 being based off a stack (indexable, but stack nonetheless), while SSE has named registers. Binaries made with MS' compiler may use both, for scalar FP arithmetic, FI.

 

[inf64]

3D Mark 11

• AMD A10-6700 “Radeon HD 8670D” - P1508
• Intel Core i5-4670K “HD 4600 GPU” - P1285
• Intel Core i5-3570K “HD 4000 GPU” - P748

http://wccftech.com/intel-hd-4600-graphics/#ixzz2THGqf7K5
​

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 .

 

[Phynaz]

Statistics?

Business junior. We buy a lot more PC's than the home user and power consumption is very important to us.

 

[AtenRa]

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 .

Thats for sure, Core i3 haswell (when it will come) has no chance at all.

 

[Enigmoid]

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 .

At unrealistically high AA settings for an igp (4x or 8x) compared to a 630 with GDDR5 (newegg has 19 630 chips and only one has gddr5).

Richland (desktop) should still walk away from it.

 

[inf64]

GT630 is 30-130% faster (in actual games,AA or no AA) than HD4600 @ 1350Mhz with DDR3-2600(!). Imagine HD4600 using normal 1600/1866Mhz memory...

 

[mikk]

If a 1200 Mhz GT2 can get below 50% to the top Richland it would be a shame for AMD because it means they are lost in the mobile segment.

 

[beginner99]

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 .

true but then no one here cares about iGPU performance on desktop part, or at least not me. And mobile parts will ship with GT3 and hence would fare better than the 4770k in 3d Mark. besides that I'm more than content with my SB laptop and HD3000. Even that is over powered if you don't play games. And games are just niche.

iGPU only becomes relevant when the iGPU can actually be used for something else than gaming. And those uses cases are at least for now even more niche.

 

[inf64]

true but then no one here cares about iGPU performance on desktop part, or at least not me. And mobile parts will ship with GT3 and hence would fare better than the 4770k in 3d Mark. besides that I'm more than content with my SB laptop and HD3000. Even that is over powered if you don't play games. And games are just niche.

iGPU only becomes relevant when the iGPU can actually be used for something else than gaming. And those uses cases are at least for now even more niche.

You might not care but most buyers ae clueless and they get that crappy IGP .

 

[AtenRa]

At unrealistically high AA settings for an igp (4x or 8x)

The GT630 gets more than 30fps across all the games in that review even with that high AA settings.

 

[Khato]

GT630 is 30-130% faster (in actual games,AA or no AA) than HD4600 @ 1350Mhz with DDR3-2600(!). Imagine HD4600 using normal 1600/1866Mhz memory...

Oh, they added some fps numbers without AA enabled? Or are you referring to another review for the no-AA portion of that statement?

There's a reason why Intel's graphics suck at AA other than memory bandwidth.

 

[Smartazz]

true but then no one here cares about iGPU performance on desktop part, or at least not me. And mobile parts will ship with GT3 and hence would fare better than the 4770k in 3d Mark. besides that I'm more than content with my SB laptop and HD3000. Even that is over powered if you don't play games. And games are just niche.

iGPU only becomes relevant when the iGPU can actually be used for something else than gaming. And those uses cases are at least for now even more niche.

For high resolution laptop displays not even the HD4000 is enough. Look at the Retina Macbook Pros.

 

[aigomorla]

sigh...

im pretty close in getting dual xeons Ivy's and saying FUUUUUUUUUUUUUUU to first gen haswell.. :\

 

[blackened23]

For high resolution laptop displays not even the HD4000 is enough. Look at the Retina Macbook Pros.

If you're going to present an argument, at least have your facts correct. The 13 inch retina macbook pro uses HD4000. The native resolution is 2560x1600. Specifications for you:

http://www.apple.com/macbook-pro/specs-retina/

Every ultra portable Apple product, except the 2799$ configuration of the macbook pro retina and 15 inch standard pro, uses HD4000. The 15 inch MBP uses the GT650M. I can tell you, that if intel matches or comes close to the 640M with the next gen ULV parts that Apple won't use discrete in the 2013 MBP. Already, nearly all of apple's products use HD4000. Now why would they use discrete when nearly all of their stuff (sans mac pro and imac) uses HD4000 _already_? They won't. Except perhaps in the high end configuration of the imac, which oddly enough costs nearly 3000$ - they would likely use a 780M for that (if they even release an imac this year, rumors indicate they won't.)

And I know you'll jump to the gaming argument. I've seen it here all too much - the thing is, the mass market doesn't care. Macbooks are not gaming machines, OSX isn't a good OS for gaming, and macs have never been designed for gaming. Period. So don't bother citing gaming performance of the 650M. Apple merely wants a display processor that can make the UI and experience snappy - apparently they thought it (HD4000) wasn't enough for 2880x1880. But this will change with Haswell.

 

[blackened23]

true but then no one here cares about iGPU performance on desktop part, or at least not me. And mobile parts will ship with GT3 and hence would fare better than the 4770k in 3d Mark. besides that I'm more than content with my SB laptop and HD3000. Even that is over powered if you don't play games. And games are just niche.

iGPU only becomes relevant when the iGPU can actually be used for something else than gaming. And those uses cases are at least for now even more niche.

Yep, gamers are a niche for ultra portable products. If you're a gamer, you're buying a different category of machine - you'll want a large full size laptop that is hot and loud, that has a quad core CPU with a mobile GPU such as the 780M.

That is a gaming laptop. Completely different category of machine.

Ultrabooks and macbooks aren't designed for gaming, period, so it's truly confusing when someone automatically jumps to the gaming argument. With these devices such as the macbook air, truly no one gives a F. They're a different category of machine designed for a different purpose. The premium here is on battery life, not gaming performance. A gaming laptop may get 30 minutes of battery life while playing crysis 3, while the next generation Haswell ultra portables will get around 10-12 hours of battery life for normal usage scenarios.

 

[NTMBK]

If you're going to present an argument, at least have your facts correct. The 13 inch retina macbook pro uses HD4000. The native resolution is 2560x1600. Specifications for you:

http://www.apple.com/macbook-pro/specs-retina/

Every ultra portable Apple product, except the 2799$ configuration of the macbook pro retina and 15 inch standard pro, uses HD4000. The 15 inch MBP uses the GT650M. I can tell you, that if intel matches or comes close to the 640M with the next gen ULV parts that Apple won't use discrete in the 2013 MBP. Already, nearly all of apple's products use HD4000. Now why would they use discrete when nearly all of their stuff (sans mac pro and imac) uses HD4000 _already_? They won't. Except perhaps in the high end configuration of the imac, which oddly enough costs nearly 3000$ - they would likely use a 780M for that (if they even release an imac this year, rumors indicate they won't.)

And I know you'll jump to the gaming argument. I've seen it here all too much - the thing is, the mass market doesn't care. Macbooks are not gaming machines, OSX isn't a good OS for gaming, and macs have never been designed for gaming. Period. So don't bother citing gaming performance of the 650M. Apple merely wants a display processor that can make the UI and experience snappy - apparently they thought it (HD4000) wasn't enough for 2880x1880. But this will change with Haswell.

The 13" MBP struggled with the HD4000:

The 2.5GHz Core i5 in the 13-inch Pro offers terrific raw CPU performance in benchmarks, running Geekbench at a solid 6700-6800 range, but the integrated Intel HD 4000 graphics chip can struggle driving such a high-resolution display. Oddly, it showed up for me more during day-to-day usage than under any crazy test situation I came up with: RAW files in Aperture scroll around just fine while QuickTime is playing back 1080p movie trailers, but Safari and Chrome both stutter a little while scrolling simple web pages. And they stutter a lot with image-heavy sites like The Verge and Polygon.

You’ll also notice some general lag when you start multitasking heavily. Open more than a few tabs and apps and you’ll start to notice things slowing down around the system as a whole. My standard workday set of 15-20 tabs open in Chrome, music playing in Spotify, email and IRC open, and Skype, iPhoto, and Messages running in the background never pegged the CPU meter, but I could clearly feel the system running just a hair behind me. You might not notice it if you’re coming from an Air or a much older MacBook, but I’m used to my 15-inch Pro with 2.3GHz Core i7 and Radeon HD 6750M graphics, and and the 13-inch Retina is definitely a little slower. If you’re a pro looking to step down to a smaller machine, you’ll almost certainly notice the performance dropoff as well.

The simple fact is that Retina display is a lot of pixels to push around for that graphics chip — just consider that Apple uses the same GPU in the MacBook Air, which has a third of the pixels and can still be made to drop frames under heavy multitasking loads. Plug in an external monitor and you’ll feel the hit come even quicker in day-to-day use — I tried both my 1080p TV over HDMI and my 24-inch 1920 x 1080 display over mini DisplayPort, and I could make YouTube videos stutter on one display simply by loading The Verge in the other.

http://www.theverge.com/2012/11/1/3585082/13-inch-macbook-pro-with-retina-display-review

 

[Phynaz]

So the stuff that's graphics intensive (Aperture & video) runs fine, a web browser with many tabs open runs slow, and he blames the graphics.

Reviewer is a moron.

 

[aigomorla]

So the stuff that's graphics intensive (Aperture & video) runs fine, a web browser with many tabs open runs slow, and he blames the graphics.

Reviewer is a moron.

LOLOL

The blind will always follow the blind.

 

[frozentundra123456]

true but then no one here cares about iGPU performance on desktop part, or at least not me. And mobile parts will ship with GT3 and hence would fare better than the 4770k in 3d Mark. besides that I'm more than content with my SB laptop and HD3000. Even that is over powered if you don't play games. And games are just niche.

iGPU only becomes relevant when the iGPU can actually be used for something else than gaming. And those uses cases are at least for now even more niche.

I agree. Igps all suck for gaming, but are adequate for the normal user. I have a sb laptop with a discrete card and in normal usage I have never seen the discrete card kick in. Granted I don't have a huge number of tabs open as well as so many other apps like that guy from the virge was talking about, but honestly, close out an app once in a while.

The only place I see igp being that significant is in a tablet or ultrabook where heat and battery life is of utmost importance and batteries must be small.

 

[NTMBK]

I agree. Igps all suck for gaming, but are adequate for the normal user. I have a sb laptop with a discrete card and in normal usage I have never seen the discrete card kick in. Granted I don't have a huge number of tabs open as well as so many other apps like that guy from the virge was talking about, but honestly, close out an app once in a while.

But what resolution was your display? 768p, 1080p? That MBP was pushing 2560x1600 around with a HD4000.

 

[inf64]

Thanks to Olivon on XS, a link to retail 4770K Haswell @ 5Ghz and 1.6V for vcore XD (presumably on air). No data on the temps or stability.

Coolaler's comment says it all in that thread:

5GHz的電壓非常醜陋

Translated : "5GHz voltage is very ugly".