Everything We Know About Intel's Skylake Platform

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.


True, that is a list, but it is a list for a Gigabyte motherboard. Motherboard OEMs are listing support for RAM at higher voltages even though Intel said you shouldn't use anything above 1.35v on Skylake. In the long run, it likely won't matter too much as the majority of Skylake systems will use DDR4, and if DDR3 above 1.35v does cause damage, it will likely be gradual and not cause a hardware failure for several years. As it stands, we can't test to know for sure because right off DDR3 at 1.5v doesn't cause any issues, we know that for sure, but it is the long term effects that we need to be concerned about, and at the point enough time has passed for us to answer that question it won't matter anymore.
 




Gigabyte absolutely released beta BIOSs that support non-k OC (although I don't think they're listed on the Gigabyte product pages anymore). I was able to get my i3-6100 up to 4.5 GHz, although it wasn't stable enough for more than a few minutes of P95 and the voltage was too high for comfort.
 


Then it should be easy to get them from the HWBOT forum (they archived all of the non-k BIOS).
 
Nope, you must be thinking of Cannonlake "expected to be released in the second half of 2017"

However, Kaby does appear to have been pushed back to the 3rd quarter of 2016 when they were originally coming in the 2nd.

I am far more interested in articles about the soon to come Z270 motherboards

200-Series Union Point Motherboards
http://www.tomshardware.com/forum/id-2983311/200-series-union-point-motherboards.html
Click to expand...
You will be lucky to see Kaby Lake mobile before the end of 2016. It will be mid 2017 at the earliest before consumers can get their hands on desktop motherboards with Union Point.
Very hard to expect a detailed breakdown of that platform at this time.
Click to expand...
Click to expand...

I'm taling about when consumers can buy kaby lake. You won't see desktop mobos for sale in 2016.
 
Sooo... Could you also mention Asus, Gigabyte and MSI in the non-K OC section? And perhaps link some sites that have archived these BIOS revisions? Like for example a HUGE list of MBOs @ http://overclocking.guide/intel-skylake-non-k-overclocking-bios-list/
 
Sorry, just read all the comments ... so to sum it up:
- ASUS, Gigabyte and MSI are all afraid of Intel and are acting like they didn't make any of the beta BIOS files for Skylake non-K OC
- Tom's Hardware is afraid of both Intel and MBO manufacturers, because they wouldn't be happy if such a big site posts the truth, because there is no "hard proof that they did it"
- article is still titled "Everything We Know About Intel's Skylake Platform"

A suggestion - change the title to "Everything We Know About Intel's Skylake Platform - and are allowed to say publicly", thanks.
 


Wow!!! Didn't know about that!!! I think that when I was shopping for a Skylake board it wasn't possible, but now it is available, at least in this very new H170 board, it goes up to DDR4-2800, not bad:
http://www.asrock.com/mb/Intel/Fatal1ty%20H170%20PerformanceHyper/index.la.asp?cat=Memory
I looked at another older one with H170 chipset and it doesn't have it.
Thanks for the info!
 
"H170 is often advertised as supporting multiple graphics cards, but motherboards based on it can only run a multi-GPU setup in a x16/x4 configuration."

At least for the price range I'm interested in, graphics boards that don't require an x16 socket are very scarce. I haven't found any so far.
 


They're referring to bandwidth, not slot size. I'm pretty sure all graphics cards require a slot that's physically x16. But the actual bandwidth to that slot can vary. So you can have a slot that's x16 in size, but only has the bandwidth of a x4 connection.
 
Actually, some boards (e.g. some Gigabyte) have three-sided PCIe X1 slots in which you could conceivably plug a X16 card and have it work, although choked.

OTOH, AM1 boards have a PCIe X16 slot, but it is only X4 electrically.
 
Sooo many people here justifying Skylake's limited PCIe lanes by saying x16/x16 isn't necessary for decent SLI/CF. Funny how review sites and users were, by complete contrast, so enthusiastic about the better PCIe provision offered by X58 when it launched (ditto X79); as VR becomes more popular and GPUs evolve to push a heck of a lot more pixels, we're going to need all the bandwidth we can get. Why the change of attitude? What happened to future proofing? This is why used-X79 is still a good deal IMO, or even Z68, etc., since many boards have NF200 or equivalent switches to offer broader PCIe provision, while X79 has more lanes anyway.


"... a huge evolution in connectivity, overclocking and, ultimately, system performance."

Er, no. The connectivity setup with Skylake is a mess. Boards are misleading buyers about their features, and what is possible in basic terms is not as good as the comparative leaps introduced with X58. Even P55 boards often used PCIe switches to offer excellent connectivity (I have one that can do x8/x8/x8/x8, I hold most of the P55 3DMark records with it); the only thing lacking back then was a decent 3rd party SATA3 controller (they all suck), Intel having no PCIe SATA3 card of its own because it would wreck sales of newer boards. Lack of decent SATA3 is the only thing really holding back older mbds from extending their usefulness, as in many cases the CPU performance is more than adequate, a point many forum posters confirm. USB3 can be added via PCIe cards and they seem to work well.

For those still delaying an upgrade, it may be that M.2 NVMe with a PCIe card could be an excellent way to hold on to a decent older platform while having access to the much greater performance offered by PCIe SSDs. I plan on testing this shortly.

It just feels like ever since Z68, Intel has been kinda jerking around with all this. Poor thermal material inside chips, 8-core CPUs with low base clocks when XEONs prove they could do much better, PCIe-crippled entry X99 CPU (it's nuts that the older 4820K has more lanes and could thus be better for a 3-way SLI/CF build than 5820K/X99), limited performance increments with each CPU release (has toms forgotten its own headline, "desktop enthusiasts yawn"?), M.2 ports that are actually only for SATA (a gotcha that's caught out many laptop buyers), Skylake's limited and confusing connectivity, etc. I don't know how anyone can look back at the way X58 was received and by comparison say that Z170 is a giant leap for enthusiasts. The cost of a K-series Skylake CPU alone should be a red flag that competition is sorely lacking in this market atm. The last CPU which was genuinely an enthusiast's dream for meddling was SandyBridge (every 2700K I've tested runs at 5GHz with just decent air cooling). Nothing since then has behaved as well, Skylake included. Where is the critical journalistic eye? Hw sites are being far too kind to Intel and other PC tech makers atm.

So I join the chorus of those hoping AMD can get back into the game with Zen, the market badly needs the competition.

And btw, anyone noticed the recent outbreak of cost-cutting by mbd makers via the lack of POST debug LED, power/reset switches and other such features on numerous Z97 and Z170 boards? These things ought to be standard by now, but so many mbds suddenly don't have them. Is the penny-pinching really that tight? The Asrock P55 Deluxe had all of these features, yet it was only 75 UKP, so it doesn't sound like they are particularly costly items to include on a mbd. Just another sign that things seem to be going backwards IMO. Intel mbd/CPU reviews these days contain references to features and options being restricted or locked out in some way far too often. I doubt the buying public would be so tolerant if car companies or TV companies made their products in the same way, yet somehow it's deemed acceptable in the tech world.

I will be impressed by a new chipset or CPU when a review spends most of its time telling me what I can do, not what I can't do.
 
I don't think it's realistic to expect grand jumps like Sandy Bridge every generation. And no, x8/x8 isn't a big deal for SLI/CFX when you're talking PCIe 3.0 bandwidth ( I'd actually wonder what kind of GPU performance you can get over 3.0 x4 right now. Anyone want to test x8/x8/x8 against x8/x8/x4? ). Combine that with how powerful and efficient cards are right now and it's all but unnecessary to run more than two GPUs for any real-world gaming, even at 4K. You're always going to have some people trying to push the bounds and it's just as difficult and expensive to do that now as it was five years ago. Is it any harder to find a proper lane splitting board and CPU for four-way GPUs now than back in X58, Z68?


This is something I have pointed out in many of my reviews. It's something that seems to have moved to only the higher-end boards and not so much in the mainstream anymore. But while it's disappearing from mainstream boards, you still see the near-pointless VGA outs on Z170 boards. If it's a matter of money, I'd rather see most of the video outputs removed to be replaced with error code displays and back panel clear CMOS buttons. The internal power/reset buttons would be nice, but I can understand those being mostly on the premium overclocking boards since those are the ones ( I think ) that more often than not find themselves on an open bench case. If you use a lot of boards of various quality on the bench, it's not difficult or expensive to rig a power/rest switch that can plug to the front panel header. Such a switch can also transfer to other boards, making it handy to the frequent OCer / tinkerer.
 
Mapesdhs makes a good point about motherboards that disable some features if others are used. Losing SATA ports is particularly irritating.
I suspect part of it is that, sadly, the enthusiast segment who cares about these things is tiny. After all, look at the prebuilts sold to consumers and businesses by the thousands or tens of thousands; most are the cheapest, sparsely-featured "H" boards being made, yet that's where the volume is.
 


I didn't say I expect such a shift every generation, that would be unreasonable. But it's been a long time now and nothing like it has happened since. Issues with each change are never fixed, eg. limited Intel SATA3 with X79, lack of native USB3 for so long, etc. I remember reading review articles about board refreshes in which the author moans about the lack of proper updates to basic tech like SATA3 provision and USB3, but these issues seem to have been largely put aside. Z170 is including what should have been available 2 years ago, but in a manner that comes with all sorts of restrictions one must take into account.




That's only for existing games, doesn't include VR demands (quadrupled fill rates) and it doesn't cover compute needs.




Of course that sort of thing can run into stutter issues. And I thought NV didn't allow SLI with x4? Not sure.

For conventional gaming (varies by title), the performance loss isn't as great as many might think (eg. check my P55 Firestrike result with three 980s), but it does matter more at higher resolutions, etc., and since that's where the tech is heading (with 4K and VR), it seems bizarre to me that mainstream chipsets have not evolved in this side of things, eg. I was shocked to find newer versions of ROG boards that don't have PCIe switches to support broader CF/SLI modes, whereas it was a key feature of older models like the M4E. Surely it's crazy that the latest ASUS M8E only supports up to x8/x4/x4, whereas the M4E (albeit with the older PCIe v2) can do x8/x16/x16 because of its NF200 switch; in most other respects the M8E is a nice board, but the PCIe config just seems lame compared to what one can get with even a used X79.




If VR takes off this year as so many hope, that will no longer apply at all. Remind me again, what are the minimum suggested specs for the OR?... 😉 Quad-buffered stereo needs a lot of RAM and fill rates, a lot more than a normal display in order to achieve the same frame rate performance.




And that's what for five years Intel has not been doing in the same way. At every stage they've been holding back, locking things up, crippling this or that. It started with the hobbled 3930K (an 8-core chip with 2 cores disabled, as reported by toms initial review) and has become way too common with nowhere near enough critical review comment IMO.




The price range is the same. There's no excuse for not having the same equivalent functionality as older boards. How much was the M4E at launch? How much is the M8E? So why no decent PCIe provision with the latter?




Obviously after Z68, X79 took the crown for 4-way, but the point is there was a plethora of choice back then. Now there isn't in the same way at all. It's not as if I'm saying anything here that hasn't been said many times by review writers, where they have to write multiple paragraphs explaining the PCIe caveats of the 5820K, the limitations of Z170, and other scenarios where limitations must be taken into account in a manner which was not the case with X58 and Z68. That's my key point.

I look at the M8E, its price, and just balk. I mean really, only x8/x4/x4? It's because of this and other factors (like the looney cost of Skylake) that for my own new gaming system I said to heck with that, I bought a used ASUS R4E, used 3930K, now I have a very potent 4.8GHz 6-core setup which is faster and a lot more flexible than a Z170 (I did buy a new 980 for it though). For my video system I bagged a P9X79-E WS which supports x16/x16/x16/x16 @ v3.0 spec (not so relevant for gaming in this case, but ideal for fitting a kickbutt Quadro, two GTX 580 3GB for extra CUDA and an M.2 PCIe).

Note I'm picking on the M8E here (and don't get me wrong, I really like ASUS boards, I have loads of them), but I could highlight other brands instead to convey the same issues.

However, some boards have been built the way they once were, eg. the GIGABYTE G170X-Gaming G1 has a PLX8747 to support x8/x8/x8/x8 (same price as the M8E, though its feedback on Amazon isn't as good, which doesn't surprise me, that's why I buy ASUS).





What's crazy is that older boards like the P55 Deluxe prove that these features do not cost very much to include, yet now they're being marketed as luxuries. With the complexity of modern tech, they ought to be compulsory on every board IMO.




Hear hear!




It would still be incredibly useful given the hordes of forum comments from people who are having problems with mainstream or budget boards.

Like I say, how much does it really cost to include these buttons and LED when the P55 Deluxe had them and it only cost 75 UKP?

I can understand biz-focused and entry/non-K chipsets not having them, but I was surprised to see so many Z97 boards without them.




The numbers are comparatively small, but the money involved is huge. The problem with many of the volume items is there's often little or no margin. I talked to the owner of a typical small high street store in CA, he told me that without premium sales of GPUs, mbds, CPUs, etc., his store could not survive, because those were the items with the significant margins, and of course it's the profit which pays the bills. By contrast, he said he often sells budget items, especially HDDs, at no profit or even a loss just to ensure people keep coming in.

The issue feeds itself though. Were the CPUs for X99 really that much of leap over IB-E for those with money to burn? I don't think so. PCIe-crippled entry chip, costly intermediate, low-clocked top-end; the whole lineup is 2 cores and a performance level lower than it should be. 5820K should have been a non-crippled 6-core (ie. what the 5930K is now), middle should have been an 8-core, top-end a 10-core (which was perfectly possible, the XEONs prove this) at a relevant price. That would have given X79 owners a reason to consider it. In the end, for many, the benefits were not worthwhile, lots of people on forums saying they'd wait for BW-E or even the next chipset.

If Intel made something genuinely worth the money, people would buy it, and a good bump over what was available before would ensure volume demand of such premium parts, helping drive the market overall. I was already disappointed by 5960X only having 8 cores, but I was shocked at its low base clock.

So again like many others, I hope AMD can give the market a good kicking with Zen.

Ian.

 
Sorry I haven't checked back in for a few weeks on this article, I'm currently away in China, and haven't had a chance to check in on comments.

I personally agree with mapesdhs about Skylake's connectivity being a being a mess, but part of the issue is more or less unavoidable for the time being. I look at the massive amounts of connections on Z170 boards and then see half are disabled when the other half are in use, and then I feel like I'm only getting half of what I paid for. But what or who is to blame for this problem? We shouldn't look at Intel as causing the issue, as all they did was try to make their chipset more versatile with a wider range of potential connections.

At first I thought it is the motherboard OEMs to blame, but then I asked representatives from one "Why add all of these connections when you can't use them all at the same time?". The rep pointed out they could only place as many connections as the board can run at the same time, but then they would get complaints about not having enough of something. For example, if they enabled all of the USB 3.0, two M.2 Key M slots, a few other PCI-E or on-board devices and then placed only two SATA-III ports on the board, then everyone would complain about the lack of SATA-III. That argument really makes sense, as we are transitioning to new and faster connections.

Ultimately, the situation is caused by the current transitional state of the market, and unavoidable. I do, however, feel that Intel should have implemented a faster interface to connect the PCH and CPU together, similar to Intel's QPI. As for PCI-E performance, another OEM claimed that a PCI-E 3.0 x4 connection could supply a GTX 980 Ti class GPU with 99 percent of its bandwidth requirements. I haven't done any testing or seen any test results from a reliable source for me to know for sure.
 
I don't think what we have now was unavoidable. Intel could have started increasing the number of HSIO/PCIe lanes in its chipsets ages ago. It's nuts that an i7 5960X has the same no. of PCIe lanes as an i7 3820, and the contrast between the 4820K and 5820K is even more weird. John Mashey (of STREAM fame) once said, "It's the bandwidth, stupid!". Seems like ever since Z68, Intel has been letting this side of its tech slide somewhat.

Let me put it another way: as mentioned, I have a P55 board that can run 3-way GTX 980 SLI (here's my Firestrike result, the fastest P55 system on 3DMark atm); it's bizarre that a supposed top-end-chipset 5820K system can't do this.

From Z77 onwards, it feels like Intel has been treading water, or even going backwards. Based on platform potential, the 4820K is a better chip than the 5820K; the latter limits what one can do with a board, the former doesn't. But now the designed-in limitations in newer CPUs are feeding into mbd limitations aswell (ie. if the CPU wasn't limited, the mbd wouldn't be either), which makes it very confusing for a buyer to work out what a particular config is capable of. Not helped of course by yet more obscure jargon and tech blurb (yay for M.2! But wait, is that port AHCI or NVMe? SATA? Er...) So now we have the worst of both worlds: confusing technobabble combined with restricted hardware.

In short, we need another X58/Nehalem-style tech bump IMO. Otherwise, the sense that each tech iteration just isn't worth the money will persist or increase, and the enthusiast PC market (where the money is) will decline further.

Ian.

 
What we need is for AM4 to be decent. It doesn't have to take the performance crown if it grabs the connectivity crown; for most tasks, CPU performance is "good enough" across the board, even if one benchmarks notably faster than another. That will provoke Intel and/or its board partners to add more lanes and/or go back to using plex switches.
 
The GIGABYTE G170X-Gaming G1 has such switches, but it costs $455! That's 80 more than a R4E Black. 😀 Heck, it's only $25 less than an X99 R5E, in which case an R5E + 5820K makes more sense (the latter CPU is only $30 more than a 6700K on Newegg).
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom