First Skylake OC report (bring the salt)

doh cpus these days only use 0.835 volts? i'm still using an i7 920 (nehalem) lol. i'm waiting for skylake because this rig needs an urgent upgrade. got a current ssd and decent video card at least, though. good to see that there may be potential with this cpu.

This has certainly changed a lot since Nehalem. On my 22nm Haswell E chip, stock load voltage is about 1.0v. I think 0.8v is about right for 14nm.

Is still has a 100Mhz bus speed. Welcome to 1999. Meantime my silly 8350 is running 250Mhz bus speed.

Frankly I'm underwhelmed. :rolleyes::eek:

A more accurate term is base clock. That's not the 1999's bus speed you are thinking about. The memory controller runs at a completely different speed, which along with other components' clock speed, are derived from the base clock and multiplier setting.
 
I think it's time for us to upgrade, fellow x58 owner.

Don't temp me. i'm itchy as! I'm in it for the system though. I don't expect much improvements in the CPU department but it'll be all very welcomed. I just wish i could get a Z170 with DDR3! I'm in AUS so DDR prices are like house and land packages cost wise!


(running a 950 stock).
 
Only 4 cores seriousily...Isn't time to keep 6 and 8 cores now with 5ghz speed. How about something innovative Intel. Not impressed. Step backwards from x99.
 
Only 4 cores seriousily...Isn't time to keep 6 and 8 cores now with 5ghz speed. How about something innovative Intel. Not impressed. Step backwards from x99.

wat.. "WTH commentary" of the year?. This isn't a replacement for X99 platform... :confused:
 
if skylake dont do 5ghz on all the cpus then its a big epic fail again.
we should been at 10ghz by now intel what are you guys doing?
Zen is coming for you
 
CPUS don't really take advantage of water cooling like GPUS do. as in, going from air to water probably won't net you any real difference in clock speeds

this is of course assuming you're using a really nice air cooler and not a stock cooler
 
Only 4 cores seriousily...Isn't time to keep 6 and 8 cores now with 5ghz speed. How about something innovative Intel. Not impressed. Step backwards from x99.

Uhh, x99 is the workstation professional line. Skylake is the next gen consumer line.

You are thinking Skylake-E.

That being said, most people will not benefit at all from anything above 4 cores. Completely wasted on most workloads, including games, so there is no point for Intel to pursue higher core counts in the consumer space.

Personally, I'd happily trade down from my -E platform (granted I have an older x79 Sandy-Bridge-E) to a 4 core platform if they only ditched the paste under the heat spreader and replaced it with solder as in the past, as well as added more PCIe lanes. The biggest shortcoming of consumer parts to me is that they have so few PCIe lanes. Even the 40 in -E parts seems stingy to me.
 
This is not impressive at all after looking at leaked benches on other sites :\ Another haswell.

Really shouldn't be surprising at all.

We have known for a long time that massive performance increases is not what Intel is targeting these days. As far as they are concerned, high performance desktops are a shrinking market. (note I said shrinking market, not dead market as many so called "journalists" say.)

Intel is instead struggling to stay relevant in a market that is increasingly mobile. All of their designs are now (and will be going forward) focused on low power optimizations, not on high clocks or high performance high wattage parts.

Since they share the same architecture across their product lines, this means lesser performance gains generation over generation.

If you optimize your architecture for power efficiency (or at least performance per watt) you can - of course - still add voltage and increase clock speeds on binned parts to get higher performing desktop parts, but they are never going to be as high performance as if the overall architecture were optimized for performance instead, at the expense of power use.

Just like in the old days, when the architectures optimized for high end desktops, when downvolted and downclocked in mobile form, sure used less power, but where never as low power as they could have been, had the architecture been optimized for it.

This is Intel's reality. They OWN the x86/Desktop/Laptop CPU market, but it is a shrinking market. It's not going away, by any means, but it is already much smaller than the phone/tablet market, and this difference is only going to grow, so since they only design one architecture (not a separate one per application) they are going to optimize that architecture for low power consumption, and try to scale it up to fit our needs. This is the new reality we live in, and no one should be surprised by it.

And they are succeeding at it too. It is amazing actually. I have three Asus Chromeboxes with Broadwell Celeron 2955U's

Two Desktop style cores (not that Atom crap as in the Celeron Nxxx models), granted clocked at a low 1.4Ghz, but using only 12W under load, 3W-5W idle, for the entire system, not just the CPU/GPU package. (They are rated 15W TDP, but I have never seen them register above 12W at the wall using my Kill-A-Watt.

So, it's a measure of the times, and I wouldn't expect any changes ever in the future. The game today is performance/watt, not outright performance, no0 matter how much we enthusiast builders jump up and down and throw temper-tantrums that we want it.

That being said, this doesn't mean we don't get improvements generation over generation. They just don't come at the pace we were used to in the 90's and early 2000's (part of that is also due to less competition from AMD)

Even single digit IPC improvements generation over generation add up when you compound them.

Let's assume an average 5% IPC increase every generation from bloomfield to skylake, which is 5 generations if I am not mistaken.

(1.05)^5 ~28%

So, approximate 28% increase in IPC

Now lets add that the top end bloomfield, the i7-975 had a base clock of 3.3Ghz, and turbo of 3.6. The tire shredding overclockers maxed those things out at ~4Ghz.

Then we look at Haswell parts which have base clocks of 4ghz, and turbo to 4.4Ghz, with max overclocks at ~5Ghz.

So we are talking another 25% from clocks. So in other words, a total increase since bloomfield of approximately 60%.

Not as impressive as back in the heyday of things, but it is not like it is no increase at all.


Anyway, the TLDR version is, Intel's focus today is on performance/watt, not outright performance, and you will never again see the type of performance increases generation over generation that you seem to expect, so you should probably readjust your expectations to the new reality.

I haven't upgraded CPU/Motherboard since 2011, the longest I have ever gone since getting into this hobby in the early 90's. While I too miss the heyday of rapid advances, this sure does have one benefit though. Our hobby has gotten less expensive to keep up with (at least on the CPU/Motherboard side) :p
 
Last edited:
[QUOTE='Zarathustra[H]Personally, I'd happily trade down from my -E platform (granted I have an older x79 Sandy-Bridge-E) to a 4 core platform if they only ditched the paste under the heat spreader and replaced it with solder as in the past, as well as added more PCIe lanes. The biggest shortcoming of consumer parts to me is that they have so few PCIe lanes. Even the 40 in -E parts seems stingy to me.[/QUOTE]

Skylake has 20 PCIe 3.0 lanes on the chipset. That give you a total of 36 lanes:

http://techreport.com/news/27761/report-skylake-chipsets-getting-massive-pcie-upgrade

The DMI bus will get a speed doubling, meaning you'll have enough bandwidth to move 4GB/s peak over that link. That's more than enough to satisfy two current PCIe nVME drives!

Thus, you can run dual-GPU without it affecting your storage upgrade options :D
 
Skylake has 20 PCIe 3.0 lanes on the chipset. That give you a total of 36 lanes:

http://techreport.com/news/27761/report-skylake-chipsets-getting-massive-pcie-upgrade

The DMI bus will get a speed doubling, meaning you'll have enough bandwidth to move 4GB/s peak over that link. That's more than enough to satisfy two current PCIe nVME drives!

Thus, you can run dual-GPU without it affecting your storage upgrade options :D

This is more interesting to me than whether these new CPUs will hit 5+ GHz. I already know it's going to be hit or miss on the overclock. Intel hasn't improved the technology enough to make it anything but luck, which is why I still have a 2500k.
 
I think it's time for us to upgrade, fellow x58 owner.

comparable clocked skylake to i7 920

Cinebench R15
skylake - 618
i7 920 - 389

Both quadcore hyperthreading enabled. both at stock speeds.
The skylake was an engineering sample that was default at 2ghz and turbo'd to 2.8ghz.
 
Zarathustra[H];1041744412 said:
That being said, most people will not benefit at all from anything above 4 cores. Completely wasted on most workloads, including games, so there is no point for Intel to pursue higher core counts in the consumer space.

I think this is a chicken and the egg problem. If Intel would release 6 cores we'd start having an opportunity to write software to take advantage of it.

Also with both consoles this generation being 8 thread, 8 IPU, 4 FPU chips we should see game makers start to dabble in >4 thread programming, and with DX12 and vulcan coming spreading out the rendering load may also be more feasible.

I think the time for highend consumer chips to go 6 core is here, but looks like it'll still be locked to the prosummer skylake-E procs.

When I bought my 4core 4ghz i7 920 I was thinkgin I'd wait till I could get a 6core 6ghz something, but looks like I'm going to settle for skylake.
 
I think this is a chicken and the egg problem. If Intel would release 6 cores we'd start having an opportunity to write software to take advantage of it.

Also with both consoles this generation being 8 thread, 8 IPU, 4 FPU chips we should see game makers start to dabble in >4 thread programming, and with DX12 and vulcan coming spreading out the rendering load may also be more feasible.

I think the time for highend consumer chips to go 6 core is here, but looks like it'll still be locked to the prosummer skylake-E procs.

When I bought my 4core 4ghz i7 920 I was thinkgin I'd wait till I could get a 6core 6ghz something, but looks like I'm going to settle for skylake.

I already write software to take advantage of multiple cores.

Developers can, and usually don't, write effective multi-threaded applications.
 
I already write software to take advantage of multiple cores.

Developers can, and usually don't, write effective multi-threaded applications.

I do as well, but I still think having more multicore machines out there would make dev's want to step up their game.

I mean one of the things I've seen proven out is if you put the hardware in the hands of developers, if it effects their daily work, they'll code for it.
 
I do as well, but I still think having more multicore machines out there would make dev's want to step up their game.

I mean one of the things I've seen proven out is if you put the hardware in the hands of developers, if it effects their daily work, they'll code for it.

Not this again.

It's not like it is just a matter for developers to buckle down and do it, and we will have multithreaded code. Most of these tasks are already multithreaded today. There are lots of tasks, most tasks even, which simply do not lend themselves well to multithreading.

Some tasks are highly parallelized , and work very well multithreaded, but other tasks simply don't, won't scale well with muktithreading, not today, not ever, no matter how much money and how many talented coders you throw at it.
 
Zarathustra[H];1041757640 said:
Not this again.

It's not like it is just a matter for developers to buckle down and do it, and we will have multithreaded code. Most of these tasks are already multithreaded today. There are lots of tasks, most tasks even, which simply do not lend themselves well to multithreading.

Some tasks are highly parallelized , and work very well multithreaded, but other tasks simply don't, won't scale well with muktithreading, not today, not ever, no matter how much money and how many talented coders you throw at it.


Sure.

I write code, professionally, I know how hard threading can be and the problems that it simply can't solve. I would surely choose a 150 gigaflops on a single core cpu over 160 gigaflops on a 6 core cpu.

But I'd rather have intel delete the integrated graphics on their high-end desktop i7s and give us 6 cores.
 
Yeah that is so tempting, but like we were both saying single thread perfomance is still king for many applications and I want to see what retail skylake looks like before I pull the trigger.

The X99 mobo costs are bit off putting too.

That Cpu is only $299 at Microcenter!!!
 
Yeah that is so tempting, but like we were both saying single thread perfomance is still king for many applications and I want to see what retail skylake looks like before I pull the trigger.

The X99 mobo costs are bit off putting too.

Overclocking is not so hard. On x99 you do not pay extra for it :D

And yeah the motherboards are more expensive, but that's because very few people need more than 4 cores. They have to make up for the small quantities by charging more. Or don't tell me you're one of those people who think they should be able to get 8 cores @ 4GHz @ 45w on a $60 motherboard?
 
It's been along time since I bought my last AMD system.

:eek: :rimshot:

Exactly the sort of cheap ass complainer I was thinking of :D

I'm just saying the single-threaded performance difference can be covered by an overclock. I'm not expecting a night-and-day difference with Skylake.

Or, you could wait another year to see if Broadwell-E drops prices any further. It will also have TSX enabled, if you desire such things. They MIGHT move an 8-core processor down to the $500 mark (like they did with the 6-core option on Ivy-E). But the cheap option should remain unchanged.

One thing's for sure: the die shrink will allow them to bump-up the clocks of those Haswell-E 8-core processors :D
 
Last edited:
I'm just saying the single-threaded performance difference can be covered by an overclock. I'm not expecting a night-and-day difference with Skylake.

I've got an ES unit in arms reach of me. It's a nice 10% or so IPC bump on haswell.

But if skylake overclocks to 4.5ghz you'd need to get haswell up over 4.75 even if you rate it conservatively at a 5% IPC bump to have the same perf.

Now granted you have 50% more cores.

But until we see retail chips and a full suite of benchmarks I'm inclined to site tight.
 
Last edited:
I've got an ES unit in arms reach of me. It's a nice 10% or so IPC bump on haswell.

But if skylake overclocks to 4.5ghz you'd need to get haswell up over 4.75 even if you rate it conservitively at a 5% IPC bump to have the same perf.

Now granted you have 50% more cores.

But until we see retail chips and a full sweet of benchmarks I'm inclined to site tight.

Fair enough, I was only making a fuss because it sounded like you "needed it yesterday." The waiting game is just as fun :D
 
I've got an ES unit in arms reach of me. It's a nice 10% or so IPC bump on haswell.

But if skylake overclocks to 4.5ghz you'd need to get haswell up over 4.75 even if you rate it conservatively at a 5% IPC bump to have the same perf.

Now granted you have 50% more cores.

But until we see retail chips and a full suite of benchmarks I'm inclined to site tight.


Imo it will be all about the oc'ing with this one. If SL can push some much higher clocks than the standard 4.5 - 5.0 we've been seeing, it's 10% IPC will look much better.
 
INB4 a massive flood of 2500/600/700Ks flood ebay. Temped to sell mine now while they're going for around 200 USD.
 
Back
Top