• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Intel Broadwell Thread

Page 108 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Simply put its not a matter of the market needing or demanding it, its a matter of competition. Nobody 'needed' a quad when the first quads were launched either but they were launched because of competition.

Core 2 Quad was at that point really meant to be server parts and should be compared to the Extreme line. It's not really comparable to the mainstream line, which now is 100% focused on mobile. You will see a 6 core mainstream part once it makes sense to do so on mobile, but since Intel wants to kill nVidia you will have to settle for more GPU.
 
How are those 8 core AMD CPUs selling? Where's all the demand for those?

They are not selling because they have worse ST performance than Intel 4 core CPUs, and you know that.

When we get 8 core Zen things will be different, assuming it delivers as promised so we get decent ST performance too from AMD.

Besides, interesting that you consider the AMD FX CPUs to be true 8 cores now. I guess you change standpoint on that depending on what suits your argument at the moment.
 
Then why do people buy Intel 4C/8T CPUs and not just 4C/4T CPUs? The former is about $100 more expensive, so that is just wasted money according to you.

Firstly I doubt the 4C8T cpu's sell as well as the 4C4T cpu's.

I bought my i5 3570K instead of the i7 3770K because I didn't see the value in the extra cost of the 4C8T cpu.

Those who do buy the 4C8T models, almost certainly do so for at least 1 of the below reasons.

1. They like the notion of having the best offering within a certain class
2. Have a variety of workloads that includes highly threaded work, but don't want to sacrifice performance on lowly threaded work.
3. Not everyone is price conscious, so provided the extra cost isn't totally out of whack, they will go for the higher end product.
 
Those who do buy the 4C8T models, almost certainly do so for at least 1 of the below reasons.

1. They like the notion of having the best offering within a certain class
2. Have a variety of workloads that includes highly threaded work, but don't want to sacrifice performance on lowly threaded work.
3. Not everyone is price conscious, so provided the extra cost isn't totally out of whack, they will go for the higher end product.

All those reasons can be applied to 8C CPUs too. So they'd probably also sell well if provided in the mainstream desktop lineup.
 
Since Intel sells the 4x Core + GT3e Broadwell at $250 and get a profit, then they could sell a 6x or 8x core Broadwell (no iGPU, same die size) at the same price.

Ohh, it would even be cheaper to design as well, no R&D for the iGPU, plain copy-paste 8 CPU cores in to a single die.
 
Since Intel sells the 4x Core + GT3e Broadwell at $250 and get a profit, then they could sell a 6x or 8x core Broadwell (no iGPU, same die size) at the same price.

Ohh, it would even be cheaper to design as well, no R&D for the iGPU, plain copy-paste 8 CPU cores in to a single die.

But Intel would need to produce a whole new set of masks, which is a very expensive process.
 
But Intel would need to produce a whole new set of masks, which is a very expensive process.

I dont think the mask cost is the problem, they simple dont want to introduce multi-core CPUs to the masses. They prefer to make a low volume GT3e SKU masks and just sell the server SKUs at higher margins for the HEAD platform.
 
They are not selling because they have worse ST performance than Intel 4 core CPUs, and you know that.

When we get 8 core Zen things will be different, assuming it delivers as promised so we get decent ST performance too from AMD.

Besides, interesting that you consider the AMD FX CPUs to be true 8 cores now. I guess you change standpoint on that depending on what suits your argument at the moment.

Why are you bringing Zen into this? Do you have any proof the 8 core FX chips sell better with better ST performance? After all you and other members post about how great FX is at highly threaded apps. Any Intel 8 core solution would also sacrifice ST performance to give you 8 cores.

I have always said FX are 8 cores, you must be confused.
 
But Intel would need to produce a whole new set of masks, which is a very expensive process.

That isn't the problem. Just look at the number of CPUs intel produces for the consumer market (2+2U, 2+3U, 2+2, 2+1, 4+2, 4+3e, possibly more). I doubt intel would be affected by the cost of a couple masks - with their margins and the demand for a cheaper 6C chip on a mainstream platform they could likely easily recoup the costs.
 
Do you have any proof the 8 core FX chips sell better with better ST performance?

It's impossible to provide proof of that, since there is no such CPU on the market, or any comparable either.

But are you really saying that the relatively poor ST performance of AMD's FX CPUs is not the reason they aren't selling? If so, what do you suggest the reason is that they are not selling as well?
 
It's impossible to provide proof of that, since there is no such CPU on the market, or any comparable either.

But are you really saying that the relatively poor ST performance of AMD's FX CPUs is not the reason they aren't selling? If so, what do you suggest the reason is that they are not selling as well?

No, I never said that poor ST performance is the reason FX CPUs aren't selling.

I'm not going to comment any further about FX sales as this is a Broadwell thread. Feel free to start a thread about FX sales and I'll post there.
 
No, I never said that poor ST performance is the reason FX CPUs aren't selling.

I'm not going to comment any further about FX sales as this is a Broadwell thread. Feel free to start a thread about FX sales and I'll post there.

Well it was you who brought up the AMD FX CPUs in the first place.

Funny that you won't comment on it now any longer when it turned out your FX based argument against 8 cores CPUs did not hold.

Oh well...
 
All those reasons can be applied to 8C CPUs too. So they'd probably also sell well if provided in the mainstream desktop lineup.
Intel's 8 core offering is clocked at 3Ghz vs the 4Ghz of the i7-4790K, despite the 8 core having 59% more TDP to play with.

Why would such a compromised desktop product sell well?
 
FTFY. Now try out your Google-Fu and see how much a 14nm mask set costs.

Its one mask. And it has the potential to sell a lot as well on the high end market as providing a very good step up incentive (ie people will be more likely to spring for a more expensive chip).

How much do you think 14nm intel masks are going to cost? (I want your number).
 
Intel's 8 core offering is clocked at 3Ghz vs the 4Ghz of the i7-4790K, despite the 8 core having 59% more TDP to play with.

Why would such a compromised desktop product sell well?

The 6 core Haswell-E is 3.5 GHz though, the same frequency as the initial 4 core variants were (e.g. 4770K). The 4790K at 4 GHz is from a later generation (Haswell Refresh), with specially binned units.

Intel could probably make similar binned higher clock variants of Haswell-E too if they wanted. They could likely also release higher clocked binned 8 core variant if they wanted to. Also, all this is on 22 nm and we're on 14 nm now, where higher density should allow for higher core count too.

Having said this I do agree that the HEDT CPUs have higher TDP than expected. 130 W TDP for 6 cores and no iGPU is quite high, when the corresponding 4 core + iGPU CPU is at 84 W TDP. I'm not sure why the Intel HEDT TDP is so high though. I have a hard time seeing how the increased core count alone should explain it.
 
Last edited:
Having said this I do agree that the HEDT CPUs have higher TDP than expected. 130 W TDP for 6 cores and no iGPU is quite high, when the corresponding 4 core + iGPU CPU is at 84 W TDP. I'm not sure why the Intel HEDT TDP is so high though. I have a hard time seeing how the increased core count alone should explain it.

Really? What else did you think was major power users in a CPU?
 
Its one mask. And it has the potential to sell a lot as well on the high end market as providing a very good step up incentive (ie people will be more likely to spring for a more expensive chip).

How much do you think 14nm intel masks are going to cost? (I want your number).

According to what I've been able to find, a mask set for Intel's 14nm process consists of 59 individual masks (estimated of course). A 14nm mask set for a high performance SOC is currently around $50M about a million bucks a mask.

Now you need multiple mask sets for each spin of your silicon, even the ones that don't even see the light of production (pun not intended). The costs add up very quickly.

Sources:
n a separate talk, Sehat Sutardja, the CEO of Marvell, said that because of the escalating costs of R&D and lithography (mask sets), beyond 28nm a chip design would need to reach very high volume (25 million units or more) just to keep costs flat.
http://www.zdnet.com/article/chipmakers-face-big-challenges-at-10nm-and-beyond/

the IC design cost for a mid-range 14nm SoC is about $80 million. “Add an extra 60% (to that cost) if embedded software development and mask costs are included,” Gartner’s Wang said. “A high-end SoC can be double this amount, and a low-end SoC with re-used IP can be half of the amount.”
http://semiengineering.com/finfet-rollout-slower-than-expected/

Interview with the CTO of Toppan Photomasks
http://semiengineering.com/behind-the-mask/

Implications of triple patterning for 14nm node design and
patterning
http://www.cerc.utexas.edu/~bei/papers/C10_SPIE2012_TPL.pdf

Code:
FDSOI	FinFET on bulk	FinFET on SOI
Mask layers	44	46	43
Multi patterning masks	8	13	11
Total masks	52	59	54
Substrate cost	High	Low	High
https://www.semiwiki.com/forum/content/3128-soi-future-flop.html

The huge investment needed to be leading edge (an estimated at $110 million to $140 million to set up a 28nm photomask shop, or $350 million to $400 million for a new 10nm mask production site) means that many fall by the wayside. LaPedus reports that there are less than a dozen leading-edge mask shops today.
http://chipdesignmag.com/sld/blog/tag/28nm/
 
The 6 core Haswell-E is 3.5 GHz though, the same frequency as the initial 4 core variants were (e.g. 4770K). The 4790K at 4 GHz is from a later generation (Haswell Refresh), with specially binned units.

Intel could probably make similar binned higher clock variants of Haswell-E too if they wanted. They could likely also release higher clocked binned 8 core variant if they wanted to. Also, all this is on 22 nm and we're on 14 nm now, where higher density should allow for higher core count too.

Having said this I do agree that the HEDT CPUs have higher TDP than expected. 130 W TDP for 6 cores and no iGPU is quite high, when the corresponding 4 core + iGPU CPU is at 84 W TDP. I'm not sure why the Intel HEDT TDP is so high though. I have a hard time seeing how the increased core count alone should explain it.

16 PCIe gen 3 lanes -> 40 PCIe gen 3 lanes adds more power
2 DDR3/4 memory channels -> 4 DDR3/4 memory channels
More complex on-chip interconnect
8MB of L3 cache -> 15-20MB L3 cache

Scaling the rest of the chip to support the increased core count isn't free from a power perspective.
 
Back
Top