Poll : Quad Core vs Hex-core ?? Choose

[Ben90]

If my 3770k could have been $100 cheaper owing to the much reduced diesize and correspondingly elevated yields and chips/wafer then that would have been a win.

I think we are all severely overestimating the price per mm^2.

Back of the napkin estimates would give us about 350 3770K per wafer. Assuming 80% yields gives us 280 3700K per wafer. A 300mm wafer costs about $2500. That is roughly $10 per chip.

Obviously there are a couple more factors at play here. One is the total cost of the fab, which is a pretty huge and not entirely a flat cost. The more wafers that need to be ran, the more machines need to be purchased and the bigger the fab needs to be.

AMD is actually a good indicator of how much these chips with all the overhead since they break even on good quarters. Keep in mind even the huge Bulldozer chips have some decent margins on them. Intel has no problem selling me a 3770K for $150 NIB.

 

[Ferzerp]

I would vote for neither option, you need the third option - keep the current quad-core layout and get rid of the iGPU and associated production cost, lowering the retail price commensurately.

If my 3770k could have been $100 cheaper owing to the much reduced diesize and correspondingly elevated yields and chips/wafer then that would have been a win.

Yeah, but you would lose out on the super cool core 0 temps due to the lack of extra silicon sitting there acting as a heat sink.

 

[lamedude]

I say replace the iGPU with moar cache. We went from 6MB per 2 cores of L2$ with Penryn to 256K per core. Yorkfield has more cache than IB so much for progress.

 

[2is]

100% YUSH. iGPU is useless for me and should b for anyone who buys a K series processor. Though i think 2 cores cost more than the iGPU.

Very useful for me. I have two monitors and run the secondary off the iGPU. This gives me about 30 watts of power savings at idle. It's also that much less VRAM being used up on my 680.

That's simply wrong..

Enthusiasts who don't use the iGPU, are a drop in the bucket of the market that Intel targets ; mainly OEMs

I'd normally agree but he did specify K series and OEMs would use the non-k variants except for Alienware-esq setups which would fall under enthusiast segment.

 

[Cerb]

Around the globe, people buying performance CPUs like the Intel Core i5 2500K/3570K and Core i7 2600K/3770K but very few if any of them are using the iGPU that comes with them.

More like, "around the globe, people are buying performance CPUs like the i5-2500 and i7-2600S, and most of them don't even know what a GPU is."

I'd get extra cores in a heartbeat, but I also understand that I am not near the middle of the bell curve.

I say replace the iGPU with moar cache. We went from 6MB per 2 cores of L2$ with Penryn to 256K per core. Yorkfield has more cache than IB so much for progress.

Both have 12MB across 4 cores. L2 to L2 is a stupid comparison, since the C2D's L2 was its LLC, and Nehalem through IB have L3 as their LLC.

 

[PhoenixEnigma]

It would depend on the machine. My desktop? Well, it has 6 cores and no iGPU now, and it's just fine - why change what works?

The HTPC? Can I get a dual core + HT and fill even more die space with some hypothetical HD6000 iGPU? Same deal on my laptop, actually.

Interestingly, though, for something like video transcoding, I'm still limited by single threaded performance. GPU transcoding is faster than CPU transcoding for me, but my GPU doesn't pass 30% utilization while one CPU core is running full tilt to feed it. As that's probably one of the few widespread uses of heavy compute for consumers, I really don't think more cores are that useful for the majority of people.

 

[boxleitnerb]

6 cores and no IGP.
But that should not mean that more cores are always better. I would only choose this particular 6 core solution because the per-core-performance is very high.

 

[T_Yamamoto]

Most people don't use igpu so MOAR COARS

 

[Revolution 11]

@T_Yamamoto You mean most enthusiasts. Most people don't have a discrete GPU or even know what that is.

I would take 4 cores with iGPU b/c Ivy Bridge is showing that at 22 nm, we are running into a heat density wall and the dead silicon area would help in this. IPC matters more to me than number of cores as my software needs are very outdated and probably never use more than 2 threads.

But I like Idontcare's answer. Lower the price of the CPU according to the decrease in area and cut the iGPU out completely.

 

[Ferzerp]

That, to me, would be a better option as well. I just don't have a pressing need for more than 4 cores (plus the little HT boost), and my PC is used for the more intensive games (I have 2x 680s), and may be doing that at the same time that it's recording 2 (rarely more than two) cable streams, while serving up a 3rd channel to a media center extender in the living room.

I can run 4 channels at once (each are up to 19Mbit of traffic, and sequential disk writes/reads), while playing games, or doing anything else I want to do with my PC without any of the activities impacting one-another with a quad core+ HT. I really don't need more than that at this time.

I would even say that I'm one of the few people who actually uses a quad+HT on a home system anywhere near enough to justify it as well.

 

[lamedude]

Both have 12MB across 4 cores. L2 to L2 is a stupid comparison, since the C2D's L2 was its LLC, and Nehalem through IB have L3 as their LLC.

15 cycles for Penryn's L2 vs 26-31 for SNB's L3
May have the same amount of total cache but its slower.

 

[Ferzerp]

15 cycles for Penryn's L2 vs 26-31 for SNB's L3
May have the same amount of total cache but its slower.

You can't just compare the number of cycles to decide speed. You have to take in to account clock speed as well. It's still slower (based on time), but not by nearly as much as the raw cycle count suggests.

It's the same phenomenon with RAM. Timings keep going up such that the latency appears to increase, but time-wise, it's really flatlined.

 

[pelov]

You think a poll on one of the biggest desktop enthusiast websites on the internet is going to be a good representation of what people want?

We know what people here want, but we're not the ones that make AMD/Intel the most money, that happens on laptops/pre-made desktops/servers the first 2 make more money with an iGPU than without one. AMD has also neglected to announce another desktop chip and initially planned to have FM2 replace AM3+ entirely.

I would love to live in your fantasy world. I really would. But reality is so much more... real.

 

[lopri]

Hexa-core for me.

 

[Phynaz]

Why stop at 6? Let's have a poll for how many people want Intel to release an 8 core for the same cost and tdp.

I think Tweakboy has been complaining g that he can't get enough cores.

 

[podspi]

What we want and what would happen are two different things. Sure, I'd love two extra cores and no IGP for the same price. But price is not set by die size, its set by what the market will bear. Removing the IGP and adding two more cores will not change that.


Anyway, I am sure OEMs love Intel's current crop of CPUs (APUs, whatever). They are stuck in a race to the bottom (especially in desktops) and probably don't mind being able to use one chip for CPU+GPU.

 

[AtenRa]

Why stop at 6? Let's have a poll for how many people want Intel to release an 8 core for the same cost and tdp.

I think Tweakboy has been complaining g that he can't get enough cores.

You forgetting i have chosen a 6-core design because it would have the same die size as a quad core + iGPU in order to have the same selling price and profit margins for Intel as with 3570K and 3770K.

An 8 core at the same $200-300 would be doable at 16nm

 

[AtenRa]

AMD has also neglected to announce another desktop chip and initially planned to have FM2 replace AM3+ entirely.

Care to provide a link for that ??

 

[pelov]

No, but I urge you to search for a Steamroller-based desktop chip and a desktop platform. If you find it on any AMD slides I'll turn over my car and house to you

(because AMD has only announced server and APUs for Steamroller and it doesn't exist)

 

[AtenRa]

No, but I urge you to search for a Steamroller-based desktop chip and a desktop platform. If you find it on any AMD slides I'll turn over my car and house to you

(because AMD has only announced server and APUs for Steamroller and it doesn't exist)

By your logic, they will not introduce the Excavator for the Desktop or APUs because they haven't released a road map slide with them yet.

We may not see Steamroller for Desktop until 2014 and 20nm.

 

[pelov]

Yet they're quite fine with releasing the APU a year early on 28nm as well as a server-based chip? Leaving out a desktop part because they just feel like it? Add to the tension in the room?

I think AMD was quite serious about leaving the battle with Intel. They're aware they lost the desktop. If a Steamroller-derived server>desktop chip won't work then they won't release it. At the moment they have new server chips and APUs covered through 2013 with no mention of a desktop chip nor even a successor to the AM3+ platform.

 

[AtenRa]

Yet they're quite fine with releasing the APU a year early on 28nm as well as a server-based chip? Leaving out a desktop part because they just feel like it? Add to the tension in the room?

I think AMD was quite serious about leaving the battle with Intel. They're aware they lost the desktop. If a Steamroller-derived server>desktop chip won't work then they won't release it. At the moment they have new server chips and APUs covered through 2013 with no mention of a desktop chip nor even a successor to the AM3+ platform.

You're jumping off the ship too early i believe.

 

[Edrick]

You forgetting i have chosen a 6-core design because it would have the same die size as a quad core + iGPU in order to have the same selling price and profit margins for Intel as with 3570K and 3770K.

It is not all about die space and core count. Other things come into play which have to be accounted for. Currently Intel's 6 and 8 core SB-E run off a quad channel memory controller. IB and SB (4 and 2 core) both run on a dual channel memory controller. How well will a dual channel controller work with 6 cores? Also, more cores may require more L3$, and in doing so will increase latency, as is proof with SB-E (as well as adding more stops to the ring bus). TDP is another issue as 2 additional cores may use more power and create more heat than Intels IGP.

You also assume (incorrectly) that Intel can keep the same clock speeds with 6 cores on the lga1155 platform. I hightly doubt it.

Bottom line, and I don't know why you just can't understand this: Intel does have 6 core CPUs already. $500. Why would they create an entirely new die, and then sell it cheaper? Makes zero business sense. If you want a 6 core, go buy one. If it is too expensive for you, write a letter to Intel.

 

[lopri]

Why stop at 6? Let's have a poll for how many people want Intel to release an 8 core for the same cost and tdp.

I think Tweakboy has been complaining g that he can't get enough cores.

Indeed I'd gladly take 8 cores!

 

[AtenRa]

It is not all about die space and core count. Other things come into play which have to be accounted for. Currently Intel's 6 and 8 core SB-E run off a quad channel memory controller. IB and SB (4 and 2 core) both run on a dual channel memory controller. How well will a dual channel controller work with 6 cores?

Just fine for 99,99999% of Desktop users.

Also, more cores may require more L3$, and in doing so will increase latency, as is proof with SB-E (as well as adding more stops to the ring bus).

No need to add more than 2-2.5MB of L3 per core for Desktop today(without iGPU).

TDP is another issue as 2 additional cores may use more power and create more heat than Intels IGP.

TDP will be lower than a 6-core SB-E(32nm) due to 22nm process. I believe it could be close to ~90W.

You also assume (incorrectly) that Intel can keep the same clock speeds with 6 cores on the lga1155 platform. I hightly doubt it.

Socket has nothing to do with core frequency. They could produce a 22nm 3.3-3.4GHz 6 core for socket 1155 easily.

Bottom line, and I don't know why you just can't understand this: Intel does have 6 core CPUs already. $500. Why would they create an entirely new die, and then sell it cheaper? Makes zero business sense. If you want a 6 core, go buy one. If it is too expensive for you, write a letter to Intel.

Of course they already have a 6-core CPU, they also have an 8-core CPU for $4K. You forgetting that advancements in Lithographic production are made so we have more performance than last litho process or the same performance of yesterday at a lower cost.

Current Intel 6 core and 8 core CPUs (actually they are of the same die) are made with 32nm node process. Intel have started producing at 22nm and even Haswell (mainstream) will only get 4 cores because they only care about raising their GPU market share.
By your logic we will always only have a $500+ 6-core CPU, that its called stagnating performance advancement.