[THG]Core i7-4770K: Haswell's Performance-Previewed

[AtenRa]

http://chinese.vr-zone.com/58507/in...70r-and-i5-4570r-with-graphics-5200-04052013/


Looks like BGA GT3e for desktop is coming.

Even with a 10% more IPC they will perform lower than Ivy 3570K(base 3.4GHz, Max Turbo 3.7GHz) and 3770K(base 3.5GHz, Max Turbo 3.9GHz) in CPU tasks.

It seems to me that Intel its taking the AMD(APU) road focusing more on the iGPU parts.

 

[AtenRa]

Finfet isn't expensive.

It is more expensive than planar.

 

[Homeles]

Even with a 10% more IPC they will perform lower than Ivy 3570K(base 3.4GHz, Max Turbo 3.7GHz) and 3770K(base 3.5GHz, Max Turbo 3.9GHz) in CPU tasks.

It seems to me that Intel its taking the AMD(APU) road focusing more on the iGPU parts.

Funny thing -- Intel had an on package GPU first. Guess that would mean it was AMD's copying Intel then, and not the other way around like you're trying to portray it.

 

[AtenRa]

Funny thing -- Intel had an on package GPU first. Guess that would mean it was AMD's copying Intel then, and not the other way around like you're trying to portray it.

That was not integrated in to the CPU die. Sandybridge was the first Intel iGPU released on January 2011 followed by AMD Llano launched in June of the same year.

And they are still behind including a node advantage.

Edit: When im talking about the AMD way, im talking of Intel focusing more and investing more logic for the iGPU than the CPU part.

 

[mikk]

It is more expensive than planar.

If you say finfet is expensive, planar is as well.

 

[AtenRa]

If you say finfet is expensive, planar is as well.

OK chief,



nexxxtttt

 

[GlacierFreeze]

Haven't kept up with Intel much in a while, especially since my last build was in '09.

Planning on building a new one later this year, when the next gen video cards come out. Haswell is CPU + integrated GPU correct? If so, since I'm going discrete GPU and not going to be interested a chip with integrated GPU, what would be the alternative that I'd need to look at towards the end of this year? IB-E? Haswell version with no iGPU?

 

[cytg111]

Haven't kept up with Intel much in a while, especially since my last build was in '09.

Planning on building a new one later this year, when the next gen video cards come out. Haswell is CPU + integrated GPU correct? If so, since I'm going discrete GPU and not going to be interested a chip with integrated GPU, what would be the alternative that I'd need to look at towards the end of this year? IB-E? Haswell version with no iGPU?

you would be looking at haswell-e.

 

[Idontcare]

IDC, Intel Ivy Core i7 3740QM (base clock 2.7GHz, Max Turbo 3.7GHz) is at 45W TDP not 50W.

According to the slides, Haswell 4950HQ (base clock 2.4GHz, max turbo 3.6GHz) is at 47W TDP. So it seems that new Haswell mobile parts featuring GT3 graphics, will not provide more CPU performance vs Ivy parts. It seems to me that Haswell design was focused on the iGPU part and we will see this trend continue with Broadwell.

It will become increasingly difficult for Intel to continue doubling the iGPU logic and at the same time keeping the same CPU performance while lowering the power consumption.

Also, with the GT3 graphics the die size should be close to 180-200mm2(from 160mm2 of Ivy). That is a huge die size for an expensive 22nm FinFet node and prices will not be cheap.

Intel's performance/cost for their graphics is not competitive with AMD's, that is for sure. I think that is why Intel is trying so hard to segment their iGPU SKUs, to recoup that cost through forced market segmentation.

I know Anand keeps saying Intel is dead serious about improving their graphics but it seems to me that if that were true then they'd have released a 350mm^s IB chip with mammoth iGPU area. But they didn't, they are interested in improving graphics but only within reason and on a budget.

 

[james1701]

In about 7 to 9 weeks, I need to build a new pc for class. Is there any chance the 4770K and mobo's will be released by then, or should I plan on a 3770K build?

 

[jpiniero]

I know Anand keeps saying Intel is dead serious about improving their graphics but it seems to me that if that were true then they'd have released a 350mm^s IB chip with mammoth iGPU area. But they didn't, they are interested in improving graphics but only within reason and on a budget.

OEMs still have to want it, or be willing to pay for it. I'm still not convinced you will see GT3e outside of that fruit company.

 

[mrmt]

I know Anand keeps saying Intel is dead serious about improving their graphics but it seems to me that if that were true then they'd have released a 350mm^s IB chip with mammoth iGPU area. But they didn't, they are interested in improving graphics but only within reason and on a budget.

And how are they going to monetize that huge iGPU?

Nobody seem too keen to pay too much for an iGPU, as we can clearly see in AMD line up. The price for AMD APU is basically dictated by direct comparison of the performance of the CPU part with correspondent Intel CPU part, the iGPU part is basically added cost. AMD simply cannot charge a price premium for that. With Intel, same thing. They can't charge a huge premium for their better iGPU part, they can raise the price less than 10% for what is a 15%+ in die size area, even more if we factor the reduced yields.

Until the bandwidth problems of the iGPU are solved, or until we move to a very space/thermal constrained form factors, APU's won't take off, because you can just spend a bit more on a discrete card and have a lot more or performance, or the MPU company can just sell substandard graphics and pocket more money from the reduced costs.

 

[Homeles]

OEMs still have to want it, or be willing to pay for it. I'm still not convinced you will see GT3e outside of that fruit company.

Apple is easily going to make use of GT3e, though. Their 11" and 13" MB Airs, 13" RMBP, Mac Mini and 21.5" iMac (possibly) could all use the crap out of GT3e. Well, I think the airs probably don't need it; GT3 would more than suffice.

There are plenty of ultrabooks that could use it as well. I would kill for a GT3e ultrabook. But I guess the usefulness of it all depends on the cost and how well it performs.

 

[jpiniero]

Apple is easily going to make use of GT3e, though. Their 11" and 13" MB Airs, 13" RMBP, Mac Mini and 21.5" iMac (possibly) could all use the crap out of GT3e. Well, I think the airs probably don't need it; GT3 would more than suffice.

Yeah. Apple can drop the discrete and make it thinner. And they may be the only ones who can sell notebooks that expensive. There's plenty of evidence they want a good gpu - there were rumors that they were interested in AMD for the Macbook Pro Retina contract, but AMD blew it. Imagine how much money that would have brought in...

 

[rgallant]

And how are they going to monetize that huge iGPU?

Nobody seem too keen to pay too much for an iGPU, as we can clearly see in AMD line up. The price for AMD APU is basically dictated by direct comparison of the performance of the CPU part with correspondent Intel CPU part, the iGPU part is basically added cost. AMD simply cannot charge a price premium for that. With Intel, same thing. They can't charge a huge premium for their better iGPU part, they can raise the price less than 10% for what is a 15%+ in die size area, even more if we factor the reduced yields.

Until the bandwidth problems of the iGPU are solved, or until we move to a very space/thermal constrained form factors, APU's won't take off, because you can just spend a bit more on a discrete card and have a lot more or performance, or the MPU company can just sell substandard graphics and pocket more money from the reduced costs.

if the patents laws didn't apply to APU's - intel with it's fab tech . would dominate this market.
-but you have different ip locked to 3 players nv\amd\intel. so unless one of these guys can make a APU work @ 100% with only 33% of the possible ark.that works ,they will all suck.

 

[Homeles]

Yeah. Apple can drop the discrete and make it thinner. And they may be the only ones who can sell notebooks that expensive. There's plenty of evidence they want a good gpu - there were rumors that they were interested in AMD for the Macbook Pro Retina contract, but AMD blew it. Imagine how much money that would have brought in...

Well, all of the products I listed, with the exception of the iMac, already use IGPs (in fact, they all use HD4000). I don't think Intel will have a part that's capable of supplanting discrete graphics until Broadwell. Broadwell should easily take care of it, though.

The main thing that GT3e would be useful for is Apple going all out retina display on their products. It's definitely got the horsepower for a 13" retina display. The iMac's a bit of an edge case... I can only see GT3e being useful if it were higher performing thanks to lifted thermal and power limits.

 

[Hulk]

Intel's performance/cost for their graphics is not competitive with AMD's, that is for sure. I think that is why Intel is trying so hard to segment their iGPU SKUs, to recoup that cost through forced market segmentation.

I know Anand keeps saying Intel is dead serious about improving their graphics but it seems to me that if that were true then they'd have released a 350mm^s IB chip with mammoth iGPU area. But they didn't, they are interested in improving graphics but only within reason and on a budget.

Exactly, within reason and on a budget. Just a few generations ago Intel graphics were only good for being able to connect a display and run 2D applications. And for their sales needs that was enough. Their partners could sell very inexpensive desktop and mobile systems without the added cost and complexity of discrete graphics.

But the market is changing. With GPU capabilities of ultra-mobile devices having reached gaming levels Intel needed to catch up. There are a lot of casual gamers who would consider 30 or 40fps at low quality "good enough." Or ones that don't really game but would like to have the option in the future.

In addition in order to hit system power targets SoC has to happen.

If Haswell GT3 can provide a doubling of HD4000 that's another step in the right direction. And Intel said that "Broadwell" would be the "big one for graphics." Imagine another 2x increase in performance. That'd be 4x what we're seeing in Ivy's graphics now.

I think they are serious but in the cautious Intel way. A couple small steps rather than one giant leap.

 

[Homeles]

And they are still behind including a node advantage.

Really, the reason for that is primarily due to the GPU simply being smaller. They're licensing graphics IP from Nvidia, so Intel's not exactly using awful tech.

Edit: When im talking about the AMD way, im talking of Intel focusing more and investing more logic for the iGPU than the CPU part.

Alright, I can give you that one.

 

[Sally600]

Oh my dear 2500K, you are going to be with me for another 3 years at least.

 

[Vesku]

Intel's performance/cost for their graphics is not competitive with AMD's, that is for sure. I think that is why Intel is trying so hard to segment their iGPU SKUs, to recoup that cost through forced market segmentation.

I know Anand keeps saying Intel is dead serious about improving their graphics but it seems to me that if that were true then they'd have released a 350mm^s IB chip with mammoth iGPU area. But they didn't, they are interested in improving graphics but only within reason and on a budget.

Agreed, they are seriously interested in improving graphics WITHIN their desired cost model.

 

[AtenRa]

Really, the reason for that is primarily due to the GPU simply being smaller.

If you take a look at the xbit review, you will see that A6-5400K with 192 SPs(half of the entire number in the die,384) is faster than HD4000.

AMD HD6450(160SPs) GPU is 67mm2 at 40nm, with 192SPs at 32nm i will estimate the die size of the iGPU to be close to 50-60mm2(Trinity has more Texture units and bigger front end).

Now, Ivybridge(HD4000) iGPU die size is close to 40mm2 at 22nm, at 32nm it should be close to 80mm2.

It is clear that at the same size Intel cannot compete and they are behind even with a node advantage.

 

[Khato]

If you take a look at the xbit review, you will see that A6-5400K with 192 SPs(half of the entire number in the die,384) is faster than HD4000.

AMD HD6450(160SPs) GPU is 67mm2 at 40nm, with 192SPs at 32nm i will estimate the die size of the iGPU to be close to 50-60mm2(Trinity has more Texture units and bigger front end).

Now, Ivybridge(HD4000) iGPU die size is close to 40mm2 at 22nm, at 32nm it should be close to 80mm2.

It is clear that at the same size Intel cannot compete and they are behind even with a node advantage.

First off, the i3-3225 beats the A6-5400K in 6 out of the 9 tests (have to find it amusing that AMD is the one with compatibility problems on 2 of the tests.) How exactly does that equate to the A6-5400K being faster than HD4000?

Second, looking at die shots of the respective CPUs I'd agree that if a 192 SP Trinity used exactly half the die space for graphics as a 384 SP version then it'd be approximately 50mm^2. That said, I'd expect that it'd probably be higher than that as about half of the die space is used for the SPs and the other half goes to other logic, some of which doesn't scale linearly with the number of SPs.

Third, to arrive at an 80mm^2 die size estimate for HD4000 on a 32nm process you're assuming perfect node scaling. This is never the case. An actual estimate for node scaling arrived at by comparing both core and cache sizes yields somewhere between 1.5x (core) and 1.67x (cache) - quite a bit short of 2x. Using those more correct estimates you get something in the 60-67mm^2 range for HD4000 on a 32nm process.

Anyway, the above isn't really meant to defend Intel... After all, they're at best comparable to AMD on performance per area when process normalized, but only achieve such by running at higher frequencies and hence using more power. At worst they're at half the performance...

And that's the indisputable issue with Intel graphics - for some games their architecture just plain sucks right now.

 

[Phynaz]

If you take a look at the xbit review, you will see that A6-5400K with 192 SPs(half of the entire number in the die,384) is faster than HD4000.

AMD HD6450(160SPs) GPU is 67mm2 at 40nm, with 192SPs at 32nm i will estimate the die size of the iGPU to be close to 50-60mm2(Trinity has more Texture units and bigger front end).

Now, Ivybridge(HD4000) iGPU die size is close to 40mm2 at 22nm, at 32nm it should be close to 80mm2.

It is clear that at the same size Intel cannot compete and they are behind even with a node advantage.

Wow that's pretty desparate, you have to compare against a CPU that never existed to make AMD look better?

 

[ShintaiDK]

http://news.cnet.com/8301-1001_3-57578193-92/intels-next-gen-haswell-chip-now-shipping-to-pc-makers/

Haswell is already shipping to OEMs and distributors. So warehouses are ready with Haswell products well in advance. Cant be long before some shop starts sell them before launch as usual.

 

[Homeles]

http://news.cnet.com/8301-1001_3-57578193-92/intels-next-gen-haswell-chip-now-shipping-to-pc-makers/

Haswell is already shipping to OEMs and distributors. So warehouses are ready with Haswell products well in advance. Cant be long before some shop starts sell them before launch as usual.

Can't be long before we see leaks, either. In fact... we should get an AnandTech preview soon.