Intel Skylake / Kaby Lake

[Arachnotronic]

Fixed that for you.

Markets can only stay "competitive" for so long before it simply becomes economically pointless for others to compete.

Look at how "competitive" the mobile chip market was; everyone was waxing lyrical about how this market is free from Intel dominance/x86 hegemony because everybody could license ARM cores/architecture.

Yeah, everybody and their cousins tried their hands at the mobile AP game and it's now basically down to one dominant player (Qualcomm) and a distant second (MediaTek).

 

[Magic Carpet]

Not true. the IMC supports up to 4133Mhz and memory modules up to 4000Mhz so far is validated.

2133Mhz is because nothing but 2133 chips existed for broad validation. Today its 2400mhz.

http://www.intel.com/content/dam/ww...emory-for-intel-core-processors-datasheet.pdf

4Ghz at 1.4v, nice. So how much more voltage IMC can sustain then and get validated by Intel?

 

[Arachnotronic]

Personally, I think if AMD ever becomes competitive again, this segmentation could backfire.

How so? Intel can very easily adjust prices to respond.

In any case, the requirement for fast memory will hurt them in the consumer space. Does anyone really think an OEM is going to use a z170 motherboard and fast ram in an off the shelf system? In the vast majority of off the shelf systems they will be leaving a lot of performance on the table. Reminds me on AMDs apus being stuck with slow single channel ram and a bios that needs to be tweaked to eliminate throttling.

It's not a "requirement," it's just the CPU cores are so fast that they actually benefit from being fed data from memory at a higher rate.

 

[ShintaiDK]

4Ghz at 1.4v, nice. So how much more voltage IMC can sustain then and get validated by Intel?

Keep checking that PDF file over time and see what comes up 🙂

 

[frozentundra123456]

Well, it is a balancing act between wringing out the most profit from each sale and giving the consumer better choices which would give more incentive to upgrade. We hear over and over in these forums how even with basically no competition, intel needs to make better products to give consumers a reason to upgrade. Gimping your own product seems just the opposite of this. I know intel has done this a long time with certain instruction sets and so forth, but I don't think most really cared. This memory thing seems a much bigger bottleneck.

Edit: I am referring here to the need to use a z170 motherboard to support sufficiently fast memory to extract the best performance.

 

[Sweepr]

Posting it here too.

Eurogamer's Core i3 6100 Review

Lots of interesting stuff.

For those that still don't believe Skylake benefits from fast memory:

Additionally, we benched the i3 6100 twice, first of all using the full 2666MHz bandwidth of our Corsair Vengeance DDR4 modules, and then paring that back to 2133MHz in order to match the memory restrictions on the H170, B150 and H110 motherboards more likely to be utilised for budget builds. And as the benchmarks came in, the results were fascinating - in many CPU-bound scenarios, the i3 6100 is significantly faster with higher-speed RAM.

The two i3 runs are probably the more fascinating comparison in the table above. Consider the difference that 2666MHz memory makes to performance. That Ryse figure is no error - performance falls through the floor when running with lower levels of bandwidth, while faster RAM offers 11 per cent more performance on GTA 5 and Far Cry 4. And again, those figures are averages spread out across the benchmark run - it's noticeably higher at any given point during 'in the moment' gameplay.

We are hearing rumours that some H170 boards may unofficially support memory overclocking too, which could save some money if true. On top of that, 2133MHz RAM is the absolute baseline - hunt around for 2666MHz sticks.

The Ryse results are particularly interesting. Haswell Core i3 manages 88.7 FPS with DDR3-2133 while Skylake Core i3 58.5 FPS with DDR4-2133. Once they used DDR4-2666, Skylake Core i3 delivered 103.2 FPS. 😱
Also generally Skylake benefited more from 25% higher clocked memory (2666 vs 2133) than Haswell with 33.3% higher clocked memory (2133 vs 1600).

The IPC gain is even more relevant here than what we saw with the first quad-core models:

It's worth repeating that clock speeds are not like-for-like, but we are seeing improvements north of 20 per cent between Skylake and Haswell here, and it's actually the case that (CryEngine apart) a Core i3 6100 with 2666MHz DDR4 is generally on par or even a little faster than an older Core i5 2500K with 1333MHz DDR3 when both systems are paired with a GTX 970. The same set-up also sees Skylake beat the AMD FX-8350 (paired with 1600MHz DDR3) in every game we tested bar Crysis 3 and The Witcher 3. Of course, those chips beg to be overclocked in a way that the i3 never can, but the bottom line is that in many gaming scenarios, the new i3 is capable of performance that belies its dual-core status.

CPU does matter:

The Skylake Core i3 6100 is an accomplished product bearing in mind its price-tag, and in conclusion, it's worth emphasising just how important performance is at this end of the market. High-end games are becoming more CPU-intensive, making it much more likely that the processor will be the bottleneck during any given gaming session - especially so at the budget end of the market. Every cycle matters in the more demanding areas of many modern titles, and in our testing, the Core i3 6100 is best in class.

www.eurogamer.net/articles/digitalfoundry-2015-intel-core-i3-6100-review

There's some video comparisons on YouTube
- Core i3 6100 vs Core i5 2500K (New 2C/4T vs Old 4C/4T)
- Core i3 6100 vs Core i5 6600K vs Core i7 6700K (Skylake Family Match)
- Core i3 6100 vs Core i3 4130 vs FX6300 (Budget CPUs Match)

Core i3 6100 will get a lot of hate because it's a dual-core but let's be honest here, it's a fantastic little chip.

 

[VirtualLarry]

Well, it is a balancing act between wringing out the most profit from each sale and giving the consumer better choices which would give more incentive to upgrade. We hear over and over in these forums how even with basically no competition, intel needs to make better products to give consumers a reason to upgrade. Gimping your own product seems just the opposite of this. I know intel has done this a long time with certain instruction sets and so forth, but I don't think most really cared. This memory thing seems a much bigger bottleneck.

Yeah. If Haswell is actually faster than Skylake, unless you: 1) Use a Z170 chipset mobo, and 2) Buy DDR4-3000+ memory for it, then screw it, I'm sticking with Haswell.

 

[Arachnotronic]

Well, it is a balancing act between wringing out the most profit from each sale and giving the consumer better choices which would give more incentive to upgrade. We hear over and over in these forums how even with basically no competition, intel needs to make better products to give consumers a reason to upgrade. Gimping your own product seems just the opposite of this. I know intel has done this a long time with certain instruction sets and so forth, but I don't think most really cared. This memory thing seems a much bigger bottleneck.

Well think of it this way. Why should I buy a Z170-based board if I can buy an H110-based board that gives me the same experience?

It's not just Intel that's interested in seeing this kind of segmentation, FYI, it's the motherboard makers too. Intel probably doesn't make all that much more extra on a Z170 chipset over an H110, but the mobo makers probably really appreciate the premium that customers are willing to pay for Z170-based boards over H110 based boards.

Also that's why you see Z170 boards launched that include higher quality components than H110 boards, more DIMM slots, more PCIe slots, etc. We going to fault the mobo makers for wanting customers to pay extra for better reliability, higher quality codecs, better warranties, etc. ?

These companies are just trying to make some money. I know that consumers tend to think about wanting to get the best value for themselves, companies be damned, but it's a two way street. If a company makes a worthwhile feature, then why begrudge them for wanting to actually get paid for it? If companies just "gave away" their hard work then they wouldn't be in business long.

 

[VirtualLarry]

Intel probably doesn't make all that much more extra on a Z170 chipset over an H110, but the mobo makers probably really appreciate the premium that customers are willing to pay for Z170-based boards over H110 based boards.

Seeing as how the H110 is highly crippled, I think that the more apt comparison is between the H170 and the Z170. They are basically the same in features, but Z170 allows overclocking. And thus the price difference seems to be primarily driven by the price difference Intel charges. At retail, it's around a $40 price difference, or nearly 2/3 the cost of the whole mobo. Intel clearly seems to be charging a significant premium for overclocking CPU and / or RAM.

These companies are just trying to make some money. I know that consumers tend to think about wanting to get the best value for themselves, companies be damned, but it's a two way street. If a company makes a worthwhile feature, then why begrudge them for wanting to actually get paid for it? If companies just "gave away" their hard work then they wouldn't be in business long.

But where is the value in value components for consumers? If the lower-end products are just designed to be purposefully so crippled, that only "stupid" consumers would choose them over the "premium" (non-crippled) SKUs, then what is the point? And the worst part is, most OEM rigs WILL use the HEAVILY-CRIPPLED SKUs on their boards.

 

[Sweepr]

Well there is chance H170 boards might unofficially support faster memory. H170 + DDR4-2666 + Core i3 6100 might make a nice cheap combo if true.

We are hearing rumours that some H170 boards may unofficially support memory overclocking too, which could save some money if true. On top of that, 2133MHz RAM is the absolute baseline - hunt around for 2666MHz sticks.

 

[ShintaiDK]

The price between Z170 and H110 is less than 10$. Everything else is other components. Overclocking as such cost around 3$ chipset wise vs H170.

 

[Arachnotronic]

But where is the value in value components for consumers? If the lower-end products are just designed to be purposefully so crippled, that only "stupid" consumers would choose them over the "premium" (non-crippled) SKUs, then what is the point? And the worst part is, most OEM rigs WILL use the HEAVILY-CRIPPLED SKUs on their boards.

There are some people that do not give a hoot about memory overclocking and would rather either pocket the money saved by buying something cheaper or apply it towards a different component that is more important to them.

As far as OEM decisions go, it's the same thing; for systems being sold to people who aren't willing to pay the premium, H110 is great. For systems being sold to people willing to pay up for performance/overclockability/etc. Z170 makes the most sense.

The right product for the right job!

 

[frozentundra123456]

How so? Intel can very easily adjust prices to respond.



It's not a "requirement," it's just the CPU cores are so fast that they actually benefit from being fed data from memory at a higher rate.

Or the consumer can just get fed up, say "screw you Intel" and not buy a system at all. I firmly believe in a declining market, companies need to provide products that better accommodate the wants and needs of the consumer rather than limiting their choices. When computer were not powerful, there was a strong need to upgrade with every generation, and it was easy to get away with these tactics. Now, most computers are good enough, and there are strong alternatives from Apple and android. Consumers can easily pick those or just not upgrade at all. Intel needs to do everything within reasonable business practices to give the consumer a reason to upgrade, not artificially limit their choices to try to wring out a few cents more profit.

 

[VirtualLarry]

Or the consumer can just get fed up, say "screw you Intel" and not buy a system at all. I firmly believe in a declining market, companies need to provide products that better accommodate the wants and needs of the consumer rather than limiting their choices. When computer were not powerful, there was a strong need to upgrade with every generation, and it was easy to get away with these tactics. Now, most computers are good enough, and there are strong alternatives from Apple and android. Consumers can easily pick those or just not upgrade at all. Intel needs to do everything within reasonable business practices to give the consumer a reason to upgrade, not artificially limit their choices to try to wring out a few cents more profit.

I agree! I'm not buying Skylake, at least currently, because of these tactics of Intel. No reason to move off of Haswell, that I can see.

This stance may change, should memory overclocking prove possible on H170. At least then, I can buy an i3, and faster DDR4 memory, with the knowledge that Intel isn't arbitrarily crippling 25% of my CPU performance, just because I refuse to pay their "tax" for the Z170 chipset, when I have no immediate plans for an M.2 PCI-E x4 SSD. (Although, the ASRock H170M Pro4 does have an Ultra M.2 slot.)

Edit: I just ran the numbers, and even then, an i3 Skylake isn't worth the upgrade. I'm currently on a Haswell G3258 @ 4.0Ghz, with 8GB of DDR3-1600 (of which, thanks to Intel's crippling, will only run at a max of 1400 in this board / CPU combo). Anyways, my current CPU + board cost $90, and 8GB of DDR3-1600 was around $50-60 when I got it. So a total of $150 for CPU/mobo/RAM.

Looking at a Skylake i3, 16GB DDR4, and an ASRock H170 mATX board, would be around $300-320. So twice the price, for... not nearly twice the performance?

 

[frozentundra123456]

Yeah. If Haswell is actually faster than Skylake, unless you: 1) Use a Z170 chipset mobo, and 2) Buy DDR4-3000+ memory for it, then screw it, I'm sticking with Haswell.

Well said Larry. I dont see the requirement for faster ram and a more expensive motherboard as a problem for high end systems, because they will be using that anyway. It is however a *big* problem for OEM systems, because they will be using cheap motherboards and the slowest, cheapest ram they can find. Hell, they might even just use single channel, which I assume would cripple performance even more. Way to shoot yourselves in the foot, Intel.

 

[ShintaiDK]

Well said Larry. I dont see the requirement for faster ram and a more expensive motherboard as a problem for high end systems, because they will be using that anyway. It is however a *big* problem for OEM systems, because they will be using cheap motherboards and the slowest, cheapest ram they can find. Hell, they might even just use single channel, which I assume would cripple performance even more. Way to shoot yourselves in the foot, Intel.

How did Intel shoot themselves in the foot? And no, Haswell is not faster than Skylake.

We just very recently got DDR4-2400 chips. Anything above that is OC.

And DDR4 is what DDR4 is because of a certain dram cartel.

Why do you think Intel is working heavily on HMC and 3DXPoint along with their EDRAM?

 

[VirtualLarry]

And no, Haswell is not faster than Skylake.

The Ryse results are particularly interesting. Haswell Core i3 manages 88.7 FPS with DDR3-2133 while Skylake Core i3 58.5 FPS with DDR4-2133.

Considering that the fastest Haswell i3 is what, 3.7Ghz? Whereas Skylake i3 is 3.7-3.9Ghz? So Skylake has: 1) higher IPC (?), and 2) clocked higher, and yet... it benchmarks SLOWER?

That only goes to prove how crippled Skylake's default memory sub-system / RAM speeds are.

We just very recently got DDR4-2400 chips. Anything above that is OC.

No, anything above DDR4-2133 is a RAM OC, and requires a Z170 chipset.

Edit: I'm talking about DDR4 speeds, as supported by the chipset. I could care less what the actual DDR4 DRAM chips are themselves rated at. As long as the module as a whole is binned properly at whatever clock speeds.

 

[frozentundra123456]

How did Intel shoot themselves in the foot? And no, Haswell is not faster than Skylake.

We just very recently got DDR4-2400 chips. Anything above that is OC.

And DDR4 is what DDR4 is because of a certain dram cartel.

Why do you think Intel is working heavily on HMC and 3DXPoint along with their EDRAM?

By spending billions to design a new architecture and process node and having it gimped by the chipset would meet my definition. Or if it makes you feel better, you could say they are just trying to "maximize profit" by forcing the consumer to buy a more expensive motherboard. So either "shooting themselves in the foot" or " screwing the consumer". I guess you can pick which ever explanation makes you feel better.

 

[PPB]

The memory scaling only sugest the wiser road for the majority of the post SB users is to wait for kaby lake, a better igp, better ddr4 RAM speed to price ratio and hopefully better binnings a la devils canyon. Haswell refresh to SKL just seems totally stupid.

PD I just built a 4790 rig in a h97m-e board. Not only it overclocks the cpu, but also the ram. The people thinking chipsets are an indicator of OC features are just totally dug into Intel's BS. You are actually getting more OC features with a b85 (b85 vanguard ie) than a b150/h170 board. It is just plain wrong.

 

[ShintaiDK]

By spending billions to design a new architecture and process node and having it gimped by the chipset would meet my definition. Or if it makes you feel better, you could say they are just trying to "maximize profit" by forcing the consumer to buy a more expensive motherboard. So either "shooting themselves in the foot" or " screwing the consumer". I guess you can pick which ever explanation makes you feel better.

By gimped. You mean you want overclocking and running any memory not validated.

Imagine paying for the features you want....

 

[Sweepr]

First Skylake-U Iris™ Graphics 540 results out. 🙂

GFXBench Windows OpenGL - 1080p Manhattan 3.1 Offscreen (Top Scores)
- Iris™ Graphics 540 (15W Skylake-U GT3e): 63.5 FPS
- HD Graphics 520 (15W Skylake-U GT2): 39.6 FPS
- Iris™ Pro Graphics 5200 (47W Haswell-H GT3e): 61.2 FPS
- Gefore GT940M: 52.3 FPS

Note that these are the first Iris submissions, so results may improve. Still, 60% faster would be a very nice boost at the same TDP. Damn, I really want a Core i7 Surface Pro 4 now.

 

[ShintaiDK]

Considering that the fastest Haswell i3 is what, 3.7Ghz? Whereas Skylake i3 is 3.7-3.9Ghz? So Skylake has: 1) higher IPC (?), and 2) clocked higher, and yet... it benchmarks SLOWER?

That only goes to prove how crippled Skylake's default memory sub-system / RAM speeds are.

Could you show me the slower part?

No, anything above DDR4-2133 is a RAM OC, and requires a Z170 chipset.

Edit: I'm talking about DDR4 speeds, as supported by the chipset. I could care less what the actual DDR4 DRAM chips are themselves rated at. As long as the module as a whole is binned properly at whatever clock speeds.

Its nothing to do with the chipset besides overclocking. 2133 is because that's the only thing Skylake was directly validated against during the design. Kaby lake for example may just be 2400Mhz. You cant validate against something that doesn't exist.

This is nothing new.

 

[ShintaiDK]

That specific Benchmark should be flawed somehow. Its theorically impossible, since the best you can hope for is linear increases. You can't pull out nearly twice more performance with a 25% Frequency increase (Which is only reflected as pure Memory Bandwidth since the higher Timmings offsets some of that Frequency increase, even if actual Memory Latency is overally lower).

Considering it stands out so much from all the other benchmarks, even its own 2666. I think we can conclude it was a fluke.

 

[VirtualLarry]

Could you show me the slower part?

"Reading is fundamental."
Sweepr posted it (the benchmark results), and I quoted it. Without overclocked RAM, Skylake is crippled.

Its nothing to do with the chipset besides overclocking. 2133 is because that's the only thing Skylake was directly validated against during the design.

And? Who the F cares what it was validated against. What I'm talking about, is the CHIPSET LIMIT. 2133 is as far as she goes, without hacking the chipset to allow OC. (Which, MSI appears to be the only 1151 mobo maker to do this thus far.)

This is nothing new.

Not true. Haswell barely mattered for benchmarks between DDR3-1600 and 2133 or 2400, other than a few synthetics. In the real world, it barely mattered.

Now, we have real world benchmarks, showing something of a 20-25% deficit in performance for Skylake, so much so, that it's actually slower than Haswell, unless running on a Z170 chipset with overclocked RAM.

Edit: And I happen to believe that Intel did that deliberately.

 

[frozentundra123456]

Yea, I can just see the advertisements now.

"Buy the Dell (insert, OEM of choice) model 150 edition, with the Z150 chipset and GET THE FULL PERFORMANCE DESIGNED INTO INTEL'S SIXTH GENERATION CPUS. Dont be gimped by slow ram like those computers from other manufacturers. We offer the best."

Only 100.00 more expensive.