Broadwell GT3 48EUs? TDP range 4.5W-47W

[mikk]

Mobile i5 and i7 are planned for Q3, mobile i3 and Broadwell-K Q4.

 

[jpiniero]

Mobile i5 and i7 are planned for Q3, mobile i3 and Broadwell-K Q4.

That was before the delay; it's more like Q4 for the mobile and very end of Q4 for Broadwell-K. And it's only the DC Ultrabook models along with the QC GT3e.

 

[Homeles]

Intel doesn't seem to care about their "fat margins" with their Fail Trail fire-sale do they?

It's pretty bad that you can't see that the most important thing for Intel right now is to break into the market.

Actually, I'm sure you can see it, you'd just rather do what you do best and spread lies to support your agenda.

 

[Ajay]

I suppose I should have a point relevant to the thread - this one is easy, Intel doesn't have the capability to beat Nvidia or AMD in graphics perf/$. If they tried to it would cost them more than they can afford to lose elsewhere in other areas that matter more to them.

Intel isn't really trying to best AMD or NV in GFX right now. They are just incrementally improving their own GFX. First off die and now on die. The result of this is that they are hurting AMD and NV at the low end. Eventually, as technology improves, Intel will hurt AMD and NV in the mid-range. If Intel can maintain their advantage in process tech, they will probably shut out AMD and NV from everything but the high end gaming market.

Intel is able to amortize these improvement over many more units and AMD and NV. I don't like it, but it looks to be inevitable - as game devs will optimize for the product that has the most consumers.

 

[Shivansps]

Im not sure if IGP will overtake everything up to high end right now, even with GT4e, eventually 4K will become mainstream and IGPs will be a joke again, and mid to high end dgpus will envolve around 4K. At least for now, not sure what gona happen when Intel moves out of silicon and stacked dram hits mainstream.

But Intel IS targeting Nvidia and AMD, thats the whole reason for crystalwell, in mobile they targeted at 650M and below with Haswell GT3e, and now Broadwell-K is a obvious attempt to overtake AMD on desktop IGPs, 16EU on Cherry Trail (more than Broadwell GT1), they are trying to overtake AMD low end APUs IGP power.

 

[Ajay]

But Intel IS targeting Nvidia and AMD, thats the whole reason for crystalwell, in mobile they targeted at 650M and below with Haswell GT3e, and now Broadwell-K is a obvious attempt to overtake AMD on desktop IGPs, 16EU on Cherry Trail (more than Broadwell GT1), they are trying to overtake AMD low end APUs IGP power.

Yes, for iGPU, but not for dGPU - that out of their reach till HBM becomes available (IMO).

 

[krumme]

Hmm... what do you mean?

What i mean is we are always told the next gen will be fantastic

Look back at the facts. I think its better to lower expectations. We are not getting a second sb igpu jump for perf and perf/power. But i mean thats okey. What is important is its priority for Intel and we will get dgpu perf but give it a few more years.

 

[krumme]

Intel doesn't seem to care about their "fat margins" with their Fail Trail fire-sale do they?

Most people try to make it as simple as possible so their feeble brains can cope with it, but business is multifaceted by it's very nature. You just see Intel's money and manufacturing prowess and automatically assume that they can do anything they want - but the more they try to hammer home their advantages everywhere, the thinner spread they get.

Intel cannot just walk all over AMD and Nvidia in graphics, while walking all over Qualcomm in mobile, while maintaining their server market share, while continuing to spend $10 billion on fab CAPEX every year just to fend off TSMC, while still keeping their "fat margins".

*Trying* to do this is the reason they are failing in most of them. He who defends everything defends nothing.

I suppose I should have a point relevant to the thread - this one is easy, Intel doesn't have the capability to beat Nvidia or AMD in graphics perf/$. If they tried to it would cost them more than they can afford to lose elsewhere in other areas that matter more to them.

Its very good point.
But right now they have no competition in x86 server. The notebooks profit end and the same for desktop.
With a reduction in capex as they do they can give away failtrail for free and bet on gfx. Its a tough road for them but for the next years they have the cash for it.
As for the gfx i am sure they will get a long way as its just a matter of technology and cash. Unlike fighting an arm ecosystem and huge brands like samsung, google and apple that will not pay the intel tax or be dependent on Intel or any other single supplier.

 

[ShintaiDK]

Yes, for iGPU, but not for dGPU - that out of their reach till HBM becomes available (IMO).

HBM, MCDRAM, etc.

Its just not one term

 

[mikk]

That was before the delay; it's more like Q4 for the mobile and very end of Q4 for Broadwell-K. And it's only the DC Ultrabook models along with the QC GT3e.

Not it is after the delay. Before the delay Broadwell was planned for H1 2014 (Q2 probably).

 

[witeken]

Intel doesn't seem to care about their "fat margins" with their Fail Trail fire-sale do they?

You can't ask premium prices if you have 0 market share. Intel tries to let companies change to x86 so x86 will have the biggest possible market share. After the market has consolidated, margins will increase.

Most people try to make it as simple as possible so their feeble brains can cope with it, but business is multifaceted by it's very nature. You just see Intel's money and manufacturing prowess and automatically assume that they can do anything they want - but the more they try to hammer home their advantages everywhere, the thinner spread they get.

Intel cannot just walk all over AMD and Nvidia in graphics, while walking all over Qualcomm in mobile, while maintaining their server market share, while continuing to spend $10 billion on fab CAPEX every year just to fend off TSMC, while still keeping their "fat margins".

I don't think competing with AMD and Nvidia in graphics is a big priority for Intel. At this moment, their biggest priority is to gain market share in the smartphone and tablet market. Atom is now treated the same as Core. Just like ARM has 3 architectures with A7 and A53, A9 and A57, and A12.

*Trying* to do this is the reason they are failing in most of them. He who defends everything defends nothing.

I'm not sure why you think they are failing in most of them. According to Wikipedia, they had 50% GPU market share in 2009 (when you include IGPs). It also seems like they will indeed walk over Qualcomm in the near future and will continue to have a process advantage over TSMC and Global Foundries.

I suppose I should have a point relevant to the thread - this one is easy, Intel doesn't have the capability to beat Nvidia or AMD in graphics perf/$. If they tried to it would cost them more than they can afford to lose elsewhere in other areas that matter more to them.

Why doesn't Intel have that capability? Their process advantage allows them to make the same GPU as Nvidia/AMD with a much higher margin (e.g. 50% vs. 70%).

 

[witeken]

What i mean is we are always told the next gen will be fantastic

Look back at the facts. I think its better to lower expectations. We are not getting a second sb igpu jump for perf and perf/power. But i mean thats okey. What is important is its priority for Intel and we will get dgpu perf but give it a few more years.

Which facts? It is expected that gen8 will be 40% faster, so that's what I use in my calculations. I read that Broadwell-K will have GT3e while Haswell-K has GT2, so performance increase should be even higher than 40%.

 

[mikk]

Which facts? It is expected that gen8 will be 40% faster, so that's what I use in my calculations. I read that Broadwell-K will have GT3e while Haswell-K has GT2, so performance increase should be even higher than 40%.

Could be lol. Haswell GT3e is already 60-70% faster than GT2 and these tests are all based on mobile tests which are slower (power restricted, memory timings restricted to usually 1600-CL11) compared to a desktop GT2. That's why I said more than once that the 80% claim from GT3e over GT2 from CPU-world would be pretty bad. Gen7.5 with 20% more shaders could easily achieve the same increase over Haswell GT2, not to mention that these predictions are often cherry picked from 3dmark. Because I expect a healthy architectural improvement with Gen8 I think there is something wrong with the CPU world claim. Maybe Broadwell-K doesn't get GT3e or maybe 80% ist just wrong.

 

[krumme]

Could be lol. Haswell GT3e is already 60-70% faster than GT2 and these tests are all based on mobile tests which are slower (power restricted, memory timings restricted to usually 1600-CL11) compared to a desktop GT2. That's why I said more than once that the 80% claim from GT3e over GT2 from CPU-world would be pretty bad. Gen7.5 with 20% more shaders could easily achieve the same increase over Haswell GT2, not to mention that these predictions are often cherry picked from 3dmark. Because I expect a healthy architectural improvement with Gen8 I think there is something wrong with the CPU world claim. Maybe Broadwell-K doesn't get GT3e or maybe 80% ist just wrong.

And get 7870 and 760 performance in actual games? Lol.

Take you 3dmark nonsense and dump it. Nothing but marketing bs cheating customers.

 

[mikk]

And get 7870 and 760 performance in actual games? Lol.

Did I say something about 7870 and 760 performance? What are you smoking?

Take you 3dmark nonsense and dump it. Nothing but marketing bs cheating customers.

Stop your fanboy attitude and learn to read. Do better next time.

 

[386user]

i doubt an hd5000 -> 6000 would be anywhere near 40% increase

sure, the highest sku could do this..but most people wont be buying these

 

[Homeles]

i doubt an hd5000 -> 6000 would be anywhere near 40% increase

sure, the highest sku could do this..but most people wont be buying these

Why's that? There's a 25% increase in units from GT2 -> GT2. It doesn't take much improvement at all to cover the rest.

 

[386user]

15 watt..low tdp form factors

 

[AtenRa]

Which facts? It is expected that gen8 will be 40% faster, so that's what I use in my calculations. I read that Broadwell-K will have GT3e while Haswell-K has GT2, so performance increase should be even higher than 40%.

So, a new 14nm process(low yields, more expensive), more EUs that eat die space, and eDRAM on top of that for the Broadwell-K Desktop ??? and how much will that be ??? 1K USD ???

 

[Homeles]

15 watt..low tdp form factors

I'm not understanding your point. At equal TDPs, Broadwell's IGP should provide a 40%+ improvement over Haswell's, regardless of whether it's a ULV product or not.

So, a new 14nm process(low yields, more expensive), more EUs that eat die space, and eDRAM on top of that for the Broadwell-K Desktop ??? and how much will that be ??? 1K USD ???

Broadwell is not more expensive than Haswell. It might temporarily be so, but in the long run, it should be substantially cheaper, given that it's somewhere around 60% of the size of Haswell (HSW GT3 to BRW GT3). That's the point of Moore's Law -- you should know this.

As far as eDRAM goes, it'll be a second generation product. That means its costs will be much, much lower.

Broadwell will have the smallest die size of Intel's in over 7 years (100mm2 or less). It's not hard to see the cost savings there.

 

[386user]

well...i will have to assume the usual 'up to 40%'

and people who are concerned with igp are likely concerned with gaming performance...( its adequate for day to day since hd3000, and even prior)

its just not a huge real life jump...

 

[mikk]

I wonder how Intel is doing the socket Broadwell GT3e, I mean the edram is pretty big how does it fit in a LGA1150 socket. I wouldn't be surprised if the GT3e claim from CPU-world turns out wrong and it's just a GT3 without edram. The 80% estimate would make much more sense then.

 

[ShintaiDK]

I wonder how Intel is doing the socket Broadwell GT3e, I mean the edram is pretty big how does it fit in a LGA1150 socket. I wouldn't be surprised if the GT3e claim from CPU-world turns out wrong and it's just a GT3 without edram. The 80% estimate would make much more sense then.

It should fit in LGA1150, Haswell GT3e was close to fit, but just a tad too big. With 14nm its easy to fit.

 

[AtenRa]

I'm not understanding your point. At equal TDPs, Broadwell's IGP should provide a 40%+ improvement over Haswell's, regardless of whether it's a ULV product or not.

Broadwell is not more expensive than Haswell. It might temporarily be so, but in the long run, it should be substantially cheaper, given that it's somewhere around 60% of the size of Haswell (HSW GT3 to BRW GT3). That's the point of Moore's Law -- you should know this.

As far as eDRAM goes, it'll be a second generation product. That means its costs will be much, much lower.

Broadwell will have the smallest die size of Intel's in over 7 years (100mm2 or less). It's not hard to see the cost savings there.

So are you telling me that a 14nm new process, bigger die (CPU + eDRAM) will be cheaper than Haswell 4c 8T GT2 selling low volume high end Desktop products (Broadwell-K) ???

 

[Homeles]

So are you telling me that a 14nm new process, bigger die (CPU + eDRAM) will be cheaper than Haswell 4c 8T GT2 selling low volume high end Desktop products (Broadwell-K) ???

So you're telling me that you can't read? Got it.

If you could go one day without making a straw man, that'd be great.