skylake processor, which one???

[David Bisset]

I suggest wearing glasses when viewing the link - the i5 6600k is the same silicon as the 6700k - both support those ram kits. As you can see, Intel has officially validated 4000Mhz DDR4 kits on Skylake - and just look around at what people are overclocking their memory to. Of course the ARK page for each CPU only list the fastest JEDEC speeds available - which has been the case for many, many years.

This is significant improvement from Haswell, where the 4790k struggled to run XMP 3000Mhz kits, unless you got the luck of the gods and won the silicon lottery. I've seen X99 users who couldn't get their boards/CPU to work properly with 3000Mhz kits also. Skylake just supports it out of the box, thanks to it's memory controller being significantly upgraded.

Just ignoring that your own source demonstrated that apart from one very small capacity kit the dual-channel skylake ratings were not faster than quad-channel x99 options were managing, despite x99 also handling double the capacity at that speed? Making skylake according to your own source a bit worse? Bearing in mind there are two main areas of interest in RAM - bandwidth and latency. The timings on the skylake options are comparable to those on the x99 options, so latency is similar. To match the bandwidth of x99 skylake would need much, much higher frequencies due to dual not quad channel. Luckily for skylake we don't often gain much from increased bandwidth at the high end so it's only a small disadvantage.

Then you go onto anecdotal evidence that disagrees with your own source - if that datasheet proves skylake can run fast memory then it proves x99 options can too.

 

[tipes]

As you appear to think otherwise, to spell this out clearly:
A Skylake i5 is no more future proof than a Haswell i5, just costs more. In very rare situations it's fractionally faster but those are few and far between and the difference even then is too small to worry about.

A Skylake i7 is LESS future proof than the 5820k having fewer cores, fewer threads and fewer PCI-E lanes and still costs more. The 5820k is part of the top tier of Intel products, Skylake is not.

You're paying a premium for a new name, not for 'future-proof' tech.

If you want to 'future-proof' a build (hint: It's basically impossible, the future hasn't happened yet so we don't know what it'll need) the best options would either by go X99 as it's much more capable and so will last longer on current performance or try to keep costs down so you can replace it sooner - in which case the Haswell i5 fits the bill nicely.

+1. Bang on I think, and as others have said x99 is the better way to go - I would go for 5930K and have done with it. You have the best of all worlds this way. Good for gaming, rendering etc etc

Mark

Mark

 

[r7slayer]

+1. Bang on I think, and as others have said x99 is the better way to go - I would go for 5930K and have done with it. You have the best of all worlds this way. Good for gaming, rendering etc etc

Mark

Mark

+2

Others will argue that games don't utilise more than 4 threads so even a i5 is financially more viable for gaming. The skylakes are slightly faster than a 5820k clock for clock but in gaming this is so ever minimal that in real word terms your probably talking 1-5 FPS TOPS which id be surprised if you would notice even with a FPS counter tbh unless you record what you get in bechmarks etc.

But the problem is like David said we don't know what the future holds and this is the same for games. Games used to utilise around 2 cores on average but now a lot or most do 4 cores. What's to say especially with DX12 around the corner more than 4 cores will become the norm? Having those extra 2 cores (4 threads) is more future proofing than getting a slightly faster clock for clock but unnoticeable yet expensive i7 skylake is it not?

When these cores (threads) become used you will notice it more than the slightly faster clock for clock skylake lol.

 

[Cern]

+2

Others will argue that games don't utilise more than 4 threads so even a i5 is financially more viable for gaming. The skylakes are slightly faster than a 5820k clock for clock but in gaming this is so ever minimal that in real word terms your probably talking 1-5 FPS TOPS which id be surprised if you would notice even with a FPS counter tbh unless you record what you get in bechmarks etc.

But the problem is like David said we don't know what the future holds and this is the same for games. Games used to utilise around 2 cores on average but now a lot or most do 4 cores. What's to say especially with DX12 around the corner more than 4 cores will become the norm? Having those extra 2 cores (4 threads) is more future proofing than getting a slightly faster clock for clock but unnoticeable yet expensive i7 skylake is it not?

When these cores (threads) become used you will notice it more than the slightly faster clock for clock skylake lol.

Will Intel release any socket 1151 Skylake CPUs in future with more than 4 cores? Or is Skylake always limited by architecture to quad core? I can't seem to find this out from any roadmap.

I'm considering a new system based around the i5-6600k (when the price becomes sane again) as this is plenty for me for now. But I wonder if I'm heading into an upgrade cul-de-sac with Skylake and Z170.

Nothing is ever future proof of course, but some platforms have more scope for future upgrades than others. Is Z170 / Skylake a relative dead end?

 

[setter]

I highly doubt it tbh. 4 core 8 thread chips always seem to be the max on the mainstream platform. Anything higher is on the enthusiast X series. Kabylake will be the next cpu range on Z170. Similar to how devils canyon replaced the first range of haswell chips.

 

[Cern]

I highly doubt it tbh. 4 core 8 thread chips always seem to be the max on the mainstream platform. Anything higher is on the enthusiast X series. Kabylake will be the next cpu range on Z170. Similar to how devils canyon replaced the first range of haswell chips.

Thanks. I wasn't aware of Kabylake. So looks like Z170 is always going to be 4/4 max.

If I do go Skylake I guess there'll be little real advantage from any future upgrades like Kabylake. So, I should expect to swap out the mobo and CPU later if I need more cores.

I use my PC mostly for gaming, digital art (Photoshop, Painter etc) and home entertainment. Mainstream usage really. I only do light overclocking, nothing at enthusiast levels. I tend to upgrade my PC infrequently, trying to make a system last 5+ years if I can.

The i7-6700k doesn't seem to be good value (even at its original lower price) but I wonder if the i5-6600k will be enough to last me the next 5 years or so.

 

[Dave2150]

Will Intel release any socket 1151 Skylake CPUs in future with more than 4 cores? Or is Skylake always limited by architecture to quad core? I can't seem to find this out from any roadmap.

I'm considering a new system based around the i5-6600k (when the price becomes sane again) as this is plenty for me for now. But I wonder if I'm heading into an upgrade cul-de-sac with Skylake and Z170.

Nothing is ever future proof of course, but some platforms have more scope for future upgrades than others. Is Z170 / Skylake a relative dead end?

It's a possibility, especially if Zen releases on time and if it's faster than Intel's offerings at the time.

Either way, there are still no games that warrant a 6 core CPU, plus developers know it would be silly to make a game need a 6 core, when less than 1% of the PC gamer world have a 6 core CPU.

Most PC gaming systems out there are quad cores without HT, some are still dual cores. Probably 5-10% have a I7 quad core with HT, then less than 1% have a 6 core 12 thread Intel CPU - what would you develop for if you were making a game that had to run well on the average PC?

 

[Dave2150]

Then you go onto anecdotal evidence that disagrees with your own source - if that datasheet proves skylake can run fast memory then it proves x99 options can too.

Err, did you even look at the link I provided? http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xmp-memory-for-intel-core-processors-datasheet.pdf

It shows the fastest DDR4 kits that Intel have validated for a variety of CPU's. Skylake (6600k) is there, the 5960X is there, 5830k is there, 5820k is there. Just press CTRL-F to seach and enter the CPU you want to view. It will then show you the fastest DDR4 kits Intel has validated that for that CPU.

Validated = confirmed to work with the voltages, frequency and timings listed here.

The 5820k has only been validated with 2400Mhz DDR4. This support my argument that for example, not all 5820k's can use 3000Mhz DDR4 - sure some golden sample chips can, though most will struggle to get it stable.

The much more expensive 5960X has been validated with 3000Mhz kits, though of course it's a much better binned CPU and is much more expensive.

Meanwhile Skylake has been validated to work with 4000Mhz DDR4 kits.

As to why Skylake has a better IMC than Haswell - this is due to it being a newer architecture. Intel obviously upgraded the speeds that the IMC is able to handle - why you can't understand this is quite strange.

 

[Orangey]

Either way, there are still no games that warrant a 6 core CPU

Stop ignoring the people correcting you.

 

[setter]

Fir a gaming setup, which the op stipulated. 4000mhz ddr4 is as useful as an ashtray on a motorbike. Spend the money on a gpu.

 

[David Bisset]

Err, did you even look at the link I provided? http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xmp-memory-for-intel-core-processors-datasheet.pdf

It shows the fastest DDR4 kits that Intel have validated for a variety of CPU's. Skylake (6600k) is there, the 5960X is there, 5830k is there, 5820k is there. Just press CTRL-F to seach and enter the CPU you want to view. It will then show you the fastest DDR4 kits Intel has validated that for that CPU.

Validated = confirmed to work with the voltages, frequency and timings listed here.

The 5820k has only been validated with 2400Mhz DDR4. This support my argument that for example, not all 5820k's can use 3000Mhz DDR4 - sure some golden sample chips can, though most will struggle to get it stable.

The much more expensive 5960X has been validated with 3000Mhz kits, though of course it's a much better binned CPU and is much more expensive.

Meanwhile Skylake has been validated to work with 4000Mhz DDR4 kits.

As to why Skylake has a better IMC than Haswell - this is due to it being a newer architecture. Intel obviously upgraded the speeds that the IMC is able to handle - why you can't understand this is quite strange.

That's hilarious - so 6600k can proxy for 6700k, but the 5930k can't for the 5820k even though it's even more similar, being exactly the same bar PCI-E lanes while hyperthreading can and does cause various performance differences - if we're arguing they've got the same IMC then it has to work both ways. Also, this is not showing what is runnable, just showing what intel happen to have done & not any kind of guarantee that another skylake chip (even another 6600k) would be able to run that - again, it's NOT a supported speeds list. To be clear, it's not any more useful than a user posting their settings that work for them in here.

Just to reiterate as reading is proving tough - the 4GHz kit is 8GB with very poor timings. Comparing any other kit with decent capacity (we're having a future-proofing debate here after all) and they're much the same in terms of frequency and timings between the 6600k and 5820k.

Even with that sole small-capacity example it's effectively slower than the 5930k manages as frequency primarily dictates bandwidth and the dual-channel disadvantage leaves it behind here. Latency is about the same for both as the higher timings go hand in hand with the frequency increase.
Again, this matters little as systems don't use the bandwidth offered by either when gaming so I'm not denegrating Skylake for being poorer in this regard - but to pretend it's better when the data you use proves no such thing is annoying.

 

[Caracus2k]

Will Intel release any socket 1151 Skylake CPUs in future with more than 4 cores? Or is Skylake always limited by architecture to quad core? I can't seem to find this out from any roadmap.

Is Z170 / Skylake a relative dead end?

Z170 will be stuck with four cores/ eight threads. Its going to have one refresh 'Kabylake' that will mainly concentrate on improving iGPU power as it will neither be a die shrink nor a redesign of the CPU layout. It will be a 'tweak' ala Haswell -> Devils Canyon (read native USB 3.1 support and HDCP 2.2 that wont matter if you're not using the iGPU).

If you anticipate needing a more heavily threaded CPU in the lifetime of the motherboard you are intending to buy then Z170 would in deed be a bad choice. Kabylake is highly unlikely imo to be a worthwhile upgrade from a 6700k Skylake and will mostly appeal to people upgrading from far older platforms who did not jump on the Haswell-E, Skylake or Broadwell-E (as this is currently expected to be out before Kabylake) bandwagons.

see here for some details

http://wccftech.com/intel-14nm-kaby...16-256-mb-edram-hseries-91w-kseries-unveiled/

https://en.wikipedia.org/wiki/Kaby_Lake

 

[Cern]

Z170 will be stuck with four cores/ eight threads. Its going to have one refresh 'Kabylake' that will mainly concentrate on improving iGPU power as it will neither be a die shrink nor a redesign of the CPU layout. It will be a 'tweak' ala Haswell -> Devils Canyon (read native USB 3.1 support and HDCP 2.2 that wont matter if you're not using the iGPU).

If you anticipate needing a more heavily threaded CPU in the lifetime of the motherboard you are intending to buy then Z170 would in deed be a bad choice. Kabylake is highly unlikely imo to be a worthwhile upgrade from a 6700k Skylake and will mostly appeal to people upgrading from far older platforms who did not jump on the Haswell-E, Skylake or Broadwell-E (as this is currently expected to be out before Kabylake) bandwagons.

see here for some details

http://wccftech.com/intel-14nm-kaby...16-256-mb-edram-hseries-91w-kseries-unveiled/

https://en.wikipedia.org/wiki/Kaby_Lake

Thanks for the info, very helpful.

Seems if I go with Z170 I should buy the best system my budget allows and then sit on it for as long as it continues to meet my needs, because future upgrades on that platform will just be incremental and not worth it.

If only x99 motherboards weren't so much more expensive, it might make my decision easier.

 

[Dave2150]

That's hilarious - so 6600k can proxy for 6700k, but the 5930k can't for the 5820k even though it's even more similar, being exactly the same bar PCI-E lanes while hyperthreading can and does cause various performance differences - if we're arguing they've got the same IMC then it has to work both ways. Also, this is not showing what is runnable, just showing what intel happen to have done & not any kind of guarantee that another skylake chip (even another 6600k) would be able to run that - again, it's NOT a supported speeds list. To be clear, it's not any more useful than a user posting their settings that work for them in here.

Just to reiterate as reading is proving tough - the 4GHz kit is 8GB with very poor timings. Comparing any other kit with decent capacity (we're having a future-proofing debate here after all) and they're much the same in terms of frequency and timings between the 6600k and 5820k.

Even with that sole small-capacity example it's effectively slower than the 5930k manages as frequency primarily dictates bandwidth and the dual-channel disadvantage leaves it behind here. Latency is about the same for both as the higher timings go hand in hand with the frequency increase.
Again, this matters little as systems don't use the bandwidth offered by either when gaming so I'm not denegrating Skylake for being poorer in this regard - but to pretend it's better when the data you use proves no such thing is annoying.

I'll stop replying since you just demonstrated you have no clue how CPU binning works.

Of course the 5820k, 5930k and 5960x are similar, they are the same silicon...... It's very easy to understand this.

Where they differ is binning. Intel put the best clockers, IMC into the 5960X. Then the next best into the 5930k, and then finally the runts of the litter go into the 5820k's (hence the price). Also as you mentioned, the 5820k is further crippled by only having 28 PCI-E lanes (not enough to run two GPU's in SLI 16X/16XX, same as Skylake).

So, the average 5820K will clock worse than the average 5930k, etc etc. This also applies to it's IMC, hence Intel won't validate the cheapest 5820k for the fastest speed memory, as it will be hit and miss whether it's stable or not.

 

[Dave2150]

Z170 will be stuck with four cores/ eight threads. Its going to have one refresh 'Kabylake' that will mainly concentrate on improving iGPU power as it will neither be a die shrink nor a redesign of the CPU layout. It will be a 'tweak' ala Haswell -> Devils Canyon (read native USB 3.1 support and HDCP 2.2 that wont matter if you're not using the iGPU).

If you anticipate needing a more heavily threaded CPU in the lifetime of the motherboard you are intending to buy then Z170 would in deed be a bad choice. Kabylake is highly unlikely imo to be a worthwhile upgrade from a 6700k Skylake and will mostly appeal to people upgrading from far older platforms who did not jump on the Haswell-E, Skylake or Broadwell-E (as this is currently expected to be out before Kabylake) bandwagons.

see here for some details

http://wccftech.com/intel-14nm-kaby...16-256-mb-edram-hseries-91w-kseries-unveiled/

https://en.wikipedia.org/wiki/Kaby_Lake

There's nothing stopping Intel making an 6 or 8 core CPU for Z170 if they want to. They could even do the same for z97.

Though the only way they'd do so is if Zen completely outclasses Intel and forces their hand, something very unlikely.

 

[Caracus2k]

There's nothing stopping Intel making an 6 or 8 core CPU for Z170 if they want to. They could even do the same for z97.

Though the only way they'd do so is if Zen completely outclasses Intel and forces their hand, something very unlikely.

Zen is currently due for an October 2016 release assuming it doesn't slip. Kabylake will be out at about the same time. If Zen somehow manages to wow you **might** see a six core mainstream processor with Cannonlake.

That and I'm not sure it would be as simple as Intel being able to add more cores to socket 1151 with a dual channel memory controller and a socket and die arrangement that was built to accommodate 4C/8t chips. Socket 2011-3 was built to accommodate 6+ cores with a quad channel memory configuration and extra CPU pins to match.

I cant find anything definitive but I would imagine that there would be multiple technical considerations to consider that would mean it would make sense just tape out a whole new CPU architecture (with a new chipset and motherboard line-up of course) if Intel wanted to have a six core + consumer line up.

bottom line - it **may** be technically possible to make a six core + chip for Z170, Z97 etc but it isn't at all likely to happen and there's nothing on the Intel roadmap that indicates that you'll be getting more than four cores on a consumer CPU any time before 2017.

I have been on hex cores for the past four years with Intel's enthusiast boards ............

(980 then 5820k)

 

[David Bisset]

I'll stop replying since you just demonstrated you have no clue how CPU binning works.

Of course the 5820k, 5930k and 5960x are similar, they are the same silicon...... It's very easy to understand this.

Where they differ is binning. Intel put the best clockers, IMC into the 5960X. Then the next best into the 5930k, and then finally the runts of the litter go into the 5820k's (hence the price). Also as you mentioned, the 5820k is further crippled by only having 28 PCI-E lanes (not enough to run two GPU's in SLI 16X/16XX, same as Skylake).

So, the average 5820K will clock worse than the average 5930k, etc etc. This also applies to it's IMC, hence Intel won't validate the cheapest 5820k for the fastest speed memory, as it will be hit and miss whether it's stable or not.

That'll be why all the 5820k's can't hit any decent clocks. Oh, wait - so far there has been no evidence of that so we really don't know much at all about how binning between those three is done (hint: given the rated supported memory speeds it's highly unlikely to be done by IMC).

Even if we assume everything you wrote here was entirely right it doesn't change that you don't understand what RAM is used for & what it's relevant limitations are so are still mistaken about it being better to right slightly higher freq with loose timings in dual channel vs slightly lower freq with tighter timings in quad. And it still doesn't matter as it makes stuff-all difference to performance.

Edit: To start over - the reason we want to increase RAM frequencies in the first place is to increase bandwidth. More bandwidth has no downsides, but beyond a certain point for most uses it has very few gains. However, very few are not none so why not?
What we gain: Bandwidth. But frequency is not the only factor here - number of channels acts as a multiplier, hence quad channel is better than dual.
What we lose: Timings need to be looser. However, timings on their own are not very interesting, it's latency we care about and that's a function of both frequency & timings e.g. 1600MHz C8 is the same latency as 2400MHz C12 (ish, as there is more to timings than just CAS of course)
So with the chips we're talking about, Skylake is miles behind in bandwidth and in a similar place with latency. However, we've already enough bandwidth for what we're doing anyway hence performance is so similar. iGPUs make better use of increased bandwidth but we're not caring about iGPU performance so it's irrelevant.

 

[Orangey]

Why does Skylake mean so much to you, Dave? Do you work for Intel? It's time to own up.

You keep ignoring when you've been corrected on things like mem OCs making no difference to gaming, needing to run both cards in 16x, etc. Textbook rep tactics.

 

[mikelkite]

As far as I can see buying into X99/Haswell-e is way more expensive than buying into Skylake.

I want to buy into a DDR4 machine, and want to have a socket that lasts for a while, so the obvious choice would be an i5 6600k, it works out at decent price. To buy into x99 it looks like the cpu is £299, the motherboard is about £150 and then you add the ram.

I think I'll keep my i7 870 + 16 GB DDR3 going for a while, with my GTX 970 I really don't have issues, I have yet to be limited in anything I can do and I currently am 3rd in class in Firestrike benchmark using the same kit.

I got a new PSU coming Wednesday (sorry OCUK you couldn't match the price and I had a voucher ) Thats the only part I am looking to replace now because it's been unstable and the PSU is 6 years old so naturally it needs to go.

I wanted to invest in DDR 4 but I don't know if I actually need to upgrade yet. I game, I work I do 3d modelling, I have 0 issues.

I am not fussed about SLI/X-fire I think the diminishing returns aren't worth it and everything scaled with SLI, not all games use it etc. So I would rather hold out for dual-gpu cards than multi-cards so PCI lanes do not bother me. 1 card for 4k gaming when it comes and I'll buy into it - for now 1440p is fine.

 

[Caracus2k]

As far as I can see buying into X99/Haswell-e is way more expensive than buying into Skylake.

I want to buy into a DDR4 machine, and want to have a socket that lasts for a while, so the obvious choice would be an i5 6600k, it works out at decent price. To buy into x99 it looks like the cpu is £299, the motherboard is about £150 and then you add the ram.

I think I'll keep my i7 870 + 16 GB DDR3 going for a while, with my GTX 970 I really don't have issues, I have yet to be limited in anything I can do and I currently am 3rd in class in Firestrike benchmark using the same kit.

I got a new PSU coming Wednesday (sorry OCUK you couldn't match the price and I had a voucher ) Thats the only part I am looking to replace now because it's been unstable and the PSU is 6 years old so naturally it needs to go.

I wanted to invest in DDR 4 but I don't know if I actually need to upgrade yet. I game, I work I do 3d modelling, I have 0 issues.

I am not fussed about SLI/X-fire I think the diminishing returns aren't worth it and everything scaled with SLI, not all games use it etc. So I would rather hold out for dual-gpu cards than multi-cards so PCI lanes do not bother me. 1 card for 4k gaming when it comes and I'll buy into it - for now 1440p is fine.

No sensible person is going to argue that a the cheapest Z170 /6600k setup is not cheaper than the cheapest x99/ 5820k combo but its an apples vs oranges comparison for the following reasons

1) the cheaper x99 boards have generally superior features to the cheaper z170 boards. When comparing boards with similar features there is far more price parity

2) you're comparing a six core/ twelve threaded CPU with a four core/ four thread cpu that's 50% more cores and 100% more threads in cpu's that will have a comparable overclock.

I want to buy into a DDR4 machine, and want to have a socket that lasts for a while, so the obvious choice would be an i5 6600k, it works out at decent price .

If you want to buy into a socket that may 'last for a while' the z170 could be an awful choice

Currently it has modern chipset features but is limited to 4c/8t chips now and and in the future it also has less potential pci-e lanes in a more restrictive layout then x99 (with the lanes being split between cpu for - gpu's only and chipset for everything else - but not for gpu's). X99 gives you up to 40 lanes to allocate as you wish and has provides access to the same features as z170.

So sure if you want a cheap to buy system now go for skylake (why stop at the 6600k there's even cheaper cpu's you could slap on a z170 board) But don't be under any illusions that you are not buying into an inferior setup both now and for the future.

Sticking with the 6600k and assuming you buy a semi decent motherboard with feature parity to an x99 motherboard the difference in price between this setup and an x99 5820k setup will solely be in the CPU (you could even use the same dual channel memory kit on x99 with little performance loss assuming its not a super slow kit to start with). The difference would therefore be circa. £80 which given the overall cost of the respective platforms is hardly 'way more expensive'