why i7 6700k always OUT OF STOCK while 6600k is available

[Hulk]

Bogus launch of Broadwell followed by a ghost launch of Skylake; it is something unusual from Intel.

Pretty much on target but I would word it as follows.

Ghost launch of Broadwell and slow launch of Skylake.


Broadwell for the desktop is still pretty much MIA. Sure they aren't fulfilling demand for all Skylake parts but at least they are out there to some extent. I'm sure yields will ramp up in the next month or so. Once the Intel machine gets rolling supply never seems to be a problem. 14nm seems like it continues to be a bee-atch for them though.

 

[Spydermag68]

I have my 6700k. It took a month before it showed up long enough to that I could order one. Right place at the right time.

 

[Pwndenburg]

Meh, I'm on a locked haswell 4770, I'll just see what this little kaby lake thing brings to the table. I like the new stuff on the boards, just don't want to blow $$ on it. I have to have the "itch." Lord knows there is no legit reason to upgrade for the foreseeable future for my usage case (gaming).

 

[iSkylaker]

Meh, I'm on a locked haswell 4770, I'll just see what this little kaby lake thing brings to the table. I like the new stuff on the boards, just don't want to blow $$ on it. I have to have the "itch." Lord knows there is no legit reason to upgrade for the foreseeable future for my usage case (gaming).

So what would you say about your chip, have you felt the need to upgrade or get an unlocked variant during the time you have had it?. I'm now considering more the i7-6700(locked) due its optimized TDP and what not, It gives me a reason to get the more mature 14nm Kabylake whenever it drops next year.

 

[Idontcare]

The 6700K is frequently out of stock whereas the 6600K is frequently in stock for one reason and one reason alone - price.

Intel priced the 6700K too low, and the 6600K too high.

Had Intel priced the 6700K at $400 or $450, and the 6600K at $200 or $150, the availability situation would be exactly the opposite between those two SKUs.

We can all make rational arguments regarding the supply disparity, but the bottom line is that the demand disparity is solely due to Intel mispricing the 6700K.

 

[frozentundra123456]

The 6700K is frequently out of stock whereas the 6600K is frequently in stock for one reason and one reason alone - price.

Intel priced the 6700K too low, and the 6600K too high.

Had Intel priced the 6700K at $400 or $450, and the 6600K at $200 or $150, the availability situation would be exactly the opposite between those two SKUs.

We can all make rational arguments regarding the supply disparity, but the bottom line is that the demand disparity is solely due to Intel mispricing the 6700K.

I dont really understand your reasoning here. The 6700k is priced (except for gouging retailers) where every unlocked i7 has been priced at introduction. In fact one could argue that the 6700k is priced a bit higher than usual because they are not shipping it with a cooler. And unlocked i5 chips have never been priced below where the 6600k is.

Now I agree that if Intel had priced the 6700k high enough, they could have hit the sales level that would have allowed it to be in stock. But that would be just artificially lowering demand to bring it into line with a limited supply of chips. I suppose one could argue that this would be pricing it "properly", but still begs the question of why supply is so low. Not to mention it would make the 5820k an even more compelling alternative.

 

[Idontcare]

I dont really understand your reasoning here. The 6700k is priced (except for gouging retailers) where every unlocked i7 has been priced at introduction. In fact one could argue that the 6700k is priced a bit higher than usual because they are not shipping it with a cooler. And unlocked i5 chips have never been priced below where the 6600k is.

Now I agree that if Intel had priced the 6700k high enough, they could have hit the sales level that would have allowed it to be in stock. But that would be just artificially lowering demand to bring it into line with a limited supply of chips. I suppose one could argue that this would be pricing it "properly", but still begs the question of why supply is so low. Not to mention it would make the 5820k an even more compelling alternative.

What does the price of prior products have to do with the demand curve of existing products?

You aren't "artificially lowering demand" if you set the price such that the supply/demand curve is nearer to equilibrium.

Whoever chooses to pay the higher price obviously feels the product is worth the higher price, regardless what a predecessor product may or may not have been valued at.

For those individuals, charging a lower price nets them a bargain (they get something for less cost than what they would have otherwise been willing to pay), but that still doesn't justify the price ever having been the traditionally expected value in the first place.

As for why the supply is low, we don't really know if the supply is low, or if demand is high. All we do know (from observation) is that there is an imbalance in the supply versus demand, one which traditionally would be addressed by elevating the pricepoint (even if temporarily until such time as more supply comes available).

 

[SOFTengCOMPelec]

What does the price of prior products have to do with the demand curve of existing products?

You aren't "artificially lowering demand" if you set the price such that the supply/demand curve is nearer to equilibrium.

Whoever chooses to pay the higher price obviously feels the product is worth the higher price, regardless what a predecessor product may or may not have been valued at.

For those individuals, charging a lower price nets them a bargain (they get something for less cost than what they would have otherwise been willing to pay), but that still doesn't justify the price ever having been the traditionally expected value in the first place.

As for why the supply is low, we don't really know if the supply is low, or if demand is high. All we do know (from observation) is that there is an imbalance in the supply versus demand, one which traditionally would be addressed by elevating the price point (even if temporarily until such time as more supply comes available).

The Skylake releases seem at least a bit ODD to me, compared to my usual amateur observations.

For example.

The previous tick/tock generation, has usually been readily available for about the last 12 months. Yet Broadwell, has mostly been a no-show. (with some exceptions).

The initial price (6700K) seems to be a bit higher than it usually is on launch day, in recent times.

Usually they have been producing the new chips, in significant quantities for a while (something like the previous 3 months). So that they will have plenty of stock, to fulfil lots of peoples purchasing demands of them. Yet this does NOT seem to have happened for Skylake. (Relatively speaking).
EDIT: Clarification. When I say that the 3 months has NOT happened. What I mean is that whatever pre-release day production took place. It has NOT been enough to allow day one world wide-stock and/or As Idontcare says, it was NOT priced high enough. (But I think there is more to it than that, compared to previous releases.)

The at release range of models, seems extremely limited.

A long time ago. Skylake was suppose (NOT officially, but by speculation) to introduce AVX3 to the (mainstream) world. (Ignoring that the non-mainstream Knights stuff, was suppose to use it first). And also ignoring that it was NOT official said which Intel consumer cpus would have it installed.
But anyway, it does NOT have AVX3, at the moment. (Usable). This could have given it a reasonable speed up, for software which has been written to use AVX3.

tl;dr
Even if Intel priced the parts (higher), so that they could be sold world-wide, currently. There are still many anomalies, compared to previous releases.
Potentially implying, that even with the Skylake generation, 14 nm is still proving to be very challenging and difficult to produce with good yields.

But in fairness, I'm speculating here.

 

[ShintaiDK]

AVX512 was never to be in client SKUs.

And the supply depends a lot on region. Look on Europe:

Skylake:
http://geizhals.eu/?cat=cpu1151

Broadwell:
http://geizhals.eu/?cat=cpu1150&xf=1133_Core+i5-5000#xf_top
http://geizhals.eu/?cat=cpu1150&xf=1133_Core+i7-5000#xf_top

All stock and free flow. Ever since releases.

 

[SOFTengCOMPelec]

AVX512 was never to be in client SKUs.

And the supply depends a lot on region. Look on Europe:

Skylake:
http://geizhals.eu/?cat=cpu1151

Broadwell:
http://geizhals.eu/?cat=cpu1150&xf=1133_Core+i5-5000#xf_top
http://geizhals.eu/?cat=cpu1150&xf=1133_Core+i7-5000#xf_top

All stock and free flow. Ever since releases.

We kind of knew, beforehand that Skylake was likely, to have little speed improvement, compared to previous generations.

But AVX512, was one of the few things, which could have meant a modest speed up capability, for at least some software. Especially in the future, after software developers have had time to utilize AVX512.

Without AVX512 (mainstream), Skylake is of even less improvement, compared to Haswell. Which at least introduced AVX2 for the first time (mainstream).

Just releasing two models (SKUs), in a limited range of world wide locations. Seems rather small fry, compared to previous releases.

EDIT2:
But in fairness. I guess in time, full (world wide) availability, and plenty of SKUs, will happen. We (some of us), just need to be patient.

 

[ShintaiDK]

We kind of knew, beforehand that Skylake was likely, to have little speed improvement, compared to previous generations.

With fast memory Skylake is 20-30% faster than Haswell.

But AVX512, was one of the few things, which could have meant a modest speed up capability, for at least some software. Especially in the future, after software developers have had time to utilize AVX512.

Without AVX512 (mainstream), Skylake is of even less benefit, compared to Haswell. Which at least introduced AVX2 for the first time (mainstream).

We need broad AVX2 utilization first. And AVX512 would increase power consumption and/or lower clocks.

Haswell also increased the speed of AVX1 instructions that are 256bit due to the cache changes.

Just releasing two models (SKUs), in a limited range of world wide locations. Seems rather small fry, compared to previous releases.

What?

 

[VirtualLarry]

We kind of knew, beforehand that Skylake was likely, to have little speed improvement, compared to previous generations.

But AVX512, was one of the few things, which could have meant a modest speed up capability, for at least some software. Especially in the future, after software developers have had time to utilize AVX512.

Without AVX512 (mainstream), Skylake is of even less benefit, compared to Haswell. Which at least introduced AVX2 for the first time (mainstream).

Just releasing two models (SKUs), in a limited range of world wide locations. Seems rather small fry, compared to previous releases.

It makes more sense, if you consider that Desktop, is really just Mobile leftovers. No reason for AVX512 to be in Mobile SKUs (only Server), so we don't get it in Desktop either.

 

[ShintaiDK]

It makes more sense, if you consider that Desktop, is really just Mobile leftovers. No reason for AVX512 to be in Mobile SKUs (only Server), so we don't get it in Desktop either.

And maybe note that Haswell Xeons (E5/E7) down clock when running AVX loads.

 

[SOFTengCOMPelec]

With fast memory Skylake is 20-30% faster than Haswell.

The bulk of sources that I have seen, seem to claim a 5% improvement, over Haswell. But they may have NOT put in the fastest DDR4 memory, and/or the fastest DDR4 memory is probably NOT generally available yet, and may not be, for many months or even years, to come.

Your 20% to 30%, sounds way too high, for most general things. With the possible exception, of VERY high memory bandwidth benchmarks and stuff.

Have you got any sources/links to your claimed speed up figures ?

Sadly, my own references, would be via www.cpubenchmark.net. Which seems to be claiming, that the new Skylake is 5% SLOWER, than the alternative, previous generation, Haswell, which I consider to be the top, mainstream part.

I.e. 4790K vs 6700K.

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-4790K+%40+4.00GHz

Which seems to give the Haswell (Devil's Canyon) a rating of 11238 (6519 samples)

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-6700K+%40+4.00GHz&id=2565

Which seems to give the Skylake I7-6700K a rating of 10695 (69 samples).

Skylake-speed / Haswell-speed = 10695/11238 = 0.95168 (Approx)

I.e. two complete generations (Broadwell, Skylake, a pair of tick, tocks), have speeded up by - (MINUS) 5%.

Sorry, if you DON'T like what I am saying. But that seems to be the measured performance. At least for THIS benchmark.

We need broad AVX2 utilization first. And AVX512 would increase power consumption and/or lower clocks.

Haswell also increased the speed of AVX1 instructions that are 256bit due to the cache changes.
What?

I'm glad that Haswell speeded up AVX1 performance.
But if Intel insists on waiting until software has already significantly used older instruction set improvements.
We could end up with a log-jam situation, and a VERY slow and long cpu improvement future.

Also I DON'T agree that Intel has released AVX2 yet. Sorry to sound nasty. But it is my opinion.

Until Intel release AVX2, across ALL potential desktop parts (and probably many others). I do NOT consider AVX2, a generally available standard.

E.g.
If *ONLY* Mac (Apple) computers had USB at the moment. No other computers had USB, except the Apple Macs.
Then I suspect, USB would NOT be the useful standard (interface) it is. With a HUGE abundance of all sorts of USB devices, at nice low prices. Such as USB flash pens.

AVX2 seems to NOT be on many/most of the lower end, Intel desktop parts.

I consider this a bad idea, and bordering on being unacceptable (my opinion). It has probably contributed to the relative lack of take-up, of AVX2. I have seen many complaints about this situation on the Internet.

tl;dr
I think Intel are being short sighted, as regards new instruction sets, such as AVX2. They should be as widely available, as possible/practicable. This then gives software developers the most incentive, to write software, which usefully utilizes, the new (and hopefully faster/better) instruction sets.

 

[ShintaiDK]

Just check the Skylake thread. And please dont use passmark as useless and wrong as it is. They dont even validate clock.

It seems with your AVX2 comment you already decided on the matters. So I dont see any reason to continue.

 

[SOFTengCOMPelec]

It makes more sense, if you consider that Desktop, is really just Mobile leftovers. No reason for AVX512 to be in Mobile SKUs (only Server), so we don't get it in Desktop either.

Good point. But I think it is a mistaken strategy of Intels, to do that. If they are not careful (Intel). They could find that the ever increasingly capable Arm processors, end up having more powerful, AVX2/AVX512 capabilities. One day, sooner or later.

E.g. I've been somewhat close to buying a mobile phone, with 8 cores (probably 4 big ones, and 4 small ones). Yet my Intel (Desktops), are still *ONLY* 4 core.

My mobile phone(s), surely should NOT have more cores, than my best, desktop computer. In my opinion.

 

[toyota]

What does the price of prior products have to do with the demand curve of existing products?

You aren't "artificially lowering demand" if you set the price such that the supply/demand curve is nearer to equilibrium.

Whoever chooses to pay the higher price obviously feels the product is worth the higher price, regardless what a predecessor product may or may not have been valued at.

For those individuals, charging a lower price nets them a bargain (they get something for less cost than what they would have otherwise been willing to pay), but that still doesn't justify the price ever having been the traditionally expected value in the first place.

As for why the supply is low, we don't really know if the supply is low, or if demand is high. All we do know (from observation) is that there is an imbalance in the supply versus demand, one which traditionally would be addressed by elevating the pricepoint (even if temporarily until such time as more supply comes available).

you seem to have an odd view on things. saying Intel should have priced a product higher because it is temporarily selling out quickly, likely due to a short term low supply, makes no sense. so should Ford have raised the MSRP on the 2015 Mustang GT because the first ones that trickled into dealerships sold out quickly? of course not. the dealerships could certainly not cut you a deal and maybe add a surcharge but Ford themselves raising the actual MSRP would have been a crazy move.

 

[ShintaiDK]

Good point. But I think it is a mistaken strategy of Intels, to do that. If they are not careful (Intel). They could find that the ever increasingly capable Arm processors, end up having more powerful, AVX2/AVX512 capabilities. One day, sooner or later.

E.g. I've been somewhat close to buying a mobile phone, with 8 cores (probably 4 big ones, and 4 small ones). Yet my Intel (Desktops), are still *ONLY* 4 core.

My mobile phone(s), surely should NOT have more cores, than my best, desktop computer. In my opinion.

ARM is further away from Core performance than ever. So ARM isnt going anywhere. If you think "moar cores" is better, get yourself an FX CPU.

And it seems you overrate the potential of AVX vastly.

 

[SOFTengCOMPelec]

Just check the Skylake thread. And please dont use passmark as useless and wrong as it is.

It seems with your AVX2 comment you already decided on the matters. So I dont see any reason to continue.

I've had VERY good results, from passmark. Do you have any sources/links, or is this just your opinion ?

I am pleased with passmark, because it seems to usually fairly accurately predict, the kind of performance I will get. From a new computer.

BUT I don't think passmark, is particularly good, when big many cored, server cpus, are being considered. It seems to under-rate their possible performance, are my early indications.

 

[SOFTengCOMPelec]

ARM is further away from Core performance than ever. So ARM isnt going anywhere.

And it seems you overrate the potential of AVX vastly.

I have NOT published any results yet. But my early indications, of trying my own software, to benchmark, arm processor(s), has shown that they are amazingly fast (taking into account the significantly lower clock frequency). Considering their low cost and low power consumption, requirements.

I've NOT seen any information, explaining how much of a speed improvement, we will get, when AVX512 finally surfaces on desktop and/or server parts.
AVX2 is usually quoted as being about 10% to 20% of a speed improvement (when/if it can be usefully used). If I recall correctly.

 

[ShintaiDK]

I've had VERY good results, from passmark. Do you have any sources/links, or is this just your opinion ?

I am pleased with passmark, because it seems to usually fairly accurately predict, the kind of performance I will get. From a new computer.

BUT I don't think passmark, is particularly good, when big many cored, server cpus, are being considered. It seems to under-rate their possible performance, are my early indications.

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-5960X+@+3.00GHz&id=2332

Try scroll down and explain to me why one 5960X can be 34.4% faster than another 5960X if you think its such a good benchmark.

22211 vs 16525 on the same chip.

I can tell you why, one is running 4.7Ghz and the other at stock. And this is why this benchmark is completely rubbish to even remotely compare with.

 

[SOFTengCOMPelec]

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-5960X+@+3.00GHz&id=2332

Try scroll down and explain to me why one 5960X can be 34.4% faster than another 5960X if you think its such a good benchmark.

22211 vs 16525 on the same chip.

I can tell you why, one is running 4.7Ghz and the other at stock. And this is why this benchmark is completely rubbish to even remotely compare with.

I agree, that they (passmark) should be robustly differentiating, between Standard cpus, and over-clocked ones.

I thought that they were already doing that. But you seem to be showing an example, where they are NOT doing that.

I am disappointed, to see any over-clocking stuff, mixed up with default setting benchmarks.

So I concede that the Skylake 6700K, may well be +5% faster than the previous Haswell 4790K. (At stock settings).

The Cinebench scores (32 bit), seem to show that the 6700K is +5% faster than the 4790K.

Since it seems to be becoming so difficult, to get much (especially single core) performance improvements. I am actually pleased that Intel have achieved the +5%.

 

[VirtualLarry]

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-5960X+@+3.00GHz&id=2332

Try scroll down and explain to me why one 5960X can be 34.4% faster than another 5960X if you think its such a good benchmark.

22211 vs 16525 on the same chip.

I can tell you why, one is running 4.7Ghz and the other at stock. And this is why this benchmark is completely rubbish to even remotely compare with.

I suspected, but didn't have any proof, that their G3258 Passmark scores might have been inflated by overclocking.

Hmm, I guess this link does prove Passmark is useless.
http://www.cpubenchmark.net/compare.php?cmp[]=2346&cmp[]=2523&cmp[]=2267

Comparison between scores of G3250, G3260, and G3258.

TL;DR: G3260 is SLOWER, than G3250, and G3258 should be the same as G3250, but it's quite a bit higher (indicating overclocking).

 

[SOFTengCOMPelec]

I suspected, but didn't have any proof, that their G3258 Passmark scores might have been inflated by overclocking.

Mostly, I look up, generally non-over-clockable parts.

Such as 4770, rather than a 4770K. Or a 3770, rather than a 3770K, etc etc.
Or Xeons, which (modern ones), are NOT especially over-clockable.
That has probably helped me get more consistent, results, from the passmark benchmark.

Their averaging, probably helps improve the accuracy, as well. If there are enough samples.

The problem with most (all) benchmarks, is they vary so widely. It is difficult to interpret them.

E.g. The FX8350 is almost on par with the 3770/4770 etc, with some benchmarks (especially multi-cored ones).
But most benchmarks show disappointing results with the FX8350, compared to Intel.

 

[SOFTengCOMPelec]

I suspected, but didn't have any proof, that their G3258 Passmark scores might have been inflated by overclocking.

Hmm, I guess this link does prove Passmark is useless.
http://www.cpubenchmark.net/compare.php?cmp[]=2346&cmp[]=2523&cmp[]=2267

Comparison between scores of G3250, G3260, and G3258.

TL;DR: G3260 is SLOWER, than G3250, and G3258 should be the same as G3250, but it's quite a bit higher (indicating overclocking).

One thing, which I have found VERY important, to do if you use passmark. Is to check that there are plenty (lots) of samples. If there are NOT, then it is best to take the results with a pinch of salt. (I.e. ignore them).

The example you have given here, has ONLY 23 samples for the G3260. This is probably too small, to get an accurate figure. EDIT: (The G3258 has got a total Samples: of 1216, which is potentially significantly more accurate.)

Some of the cpus they have listed, have only got a VERY small number of samples. E.g. 3 or 6.
Such small numbers can give VERY inaccurate results indeed.