Skylake CPU's leaked with specs

[swifty7]

In my case I'm very insterested in upgrading to Skylake.

 

My old Phenom II 965be (4.1GHz) did well all this time

 

i5 6600k + air cooler + motherboard + ddr4 16gb = 600-700€

 

The rest of the my actual computer can wait a bit.

GPU Asus 770 2Gb (1080p gaming)

PSU Corsair TX650

 

 

How many improvement will I get? 100% or more maybe?

actually your GPU does need an upgrade.  Even at 1080p, some games these days/future require more than 2GB worth of vram, that is if you want everything to run at least 60fps maxed out.

[patrickjp93]

That "toothpaste" was used because the solder was starting to kill dies with them becoming so small and the transistors getting smaller. Im pretty sure you prefer them using toothpaste on your 200$ i5 than them using solder but killing half their dies and so having to jack the price up to 400$ for an i5? Stop complaining over stuff you know nothing about.

Do you have a source on this?

[LukaP]

Do you have a source on this?

I remember reading it on either anandtech or ars-technica, but it was a while ago (between Hw and Hw-Refresh launches) and i really cant be bothered to find it now. But i remember it was a quote from an Intel engineer, who said the percentage of dies that were ruined by the heat was too big for the margins.

[swifty7]

so! according to that Swedish article, Skylake desktop will be released between July and September.  I seriously doubt it, at least for the unlocked ones.  I guess I have more waiting to do.

[wachuwey]

actually your GPU does need an upgrade.  Even at 1080p, some games these days/future require more than 2GB worth of vram, that is if you want everything to run at least 60fps maxed out.

I know a 770 is not capable to deliver 60fps at max settings in recent games, but I don't care if I have to turn off some stuff.

Don't want to play games in low settings, but it's not a problem if they are not at max.

That's why I think this gpu is ok for now.

[patrickjp93]

I remember reading it on either anandtech or ars-technica, but it was a while ago (between Hw and Hw-Refresh launches) and i really cant be bothered to find it now. But i remember it was a quote from an Intel engineer, who said the percentage of dies that were ruined by the heat was too big for the margins.

Weird, my dad and I soldered his 4790K without an issue, though maybe in batches there's a collective heating that causes a problem. I wonder if AMD will have to stop soldering at 14nm as well or if an alternate metal combo can be found.

[Opcode]

3DMark CPU scores show a 19.2% IPC increase comparing 4790K vs. 6700K. 11515 vs 10100 with both chips running at full boost tilt.

(1+(1-4.2/4.4))*11515/10100 =1.1919 or 119.2% the performance of Haswell per clock or a 19.2% performance boost, and since multicore scaling isn't perfect, we can conclude it's actually better than this.

Right now "leaked" benchmarks are showing a 14% increase in performance 4790k vs 6700k in threaded workloads clock for clock.

 

 

[LukaP]

Weird, my dad and I soldered his 4790K without an issue, though maybe in batches there's a collective heating that causes a problem. I wonder if AMD will have to stop soldering at 14nm as well or if an alternate metal combo can be found.

Well its logical. Intels last mainstream node to be soldered was 32nm. AMD is soldering at 28nm. We will see though. Would be interesting to see them use a metal with a low melting point (A gallium alloy would be interesting)

[patrickjp93]

Right now "leaked" benchmarks are showing a 14% increase in performance 4790k vs 6700k in threaded workloads clock for clock. That IPC doesn't scale to be 14% across each individual core but all of them as a package for real world performance so we'll probably generally see up to a ~5% increase in single thread.

Different clock speeds. You're comparing raw performance of 4.4GHz vs. 4.2GHz, which is incorrect. You must normalize. Furthermore, you incorrectly calculated per-core IPC. Scaling is imperfect. To have a 15 (19)% increase in multi-core, you must have a greater increase in single core to compensate for less than perfect scaling.

[Opcode]

Different clock speeds. You're comparing raw performance of 4.4GHz vs. 4.2GHz, which is incorrect. You must normalize. Furthermore, you incorrectly calculated per-core IPC. Scaling is imperfect. To have a 15 (19)% increase I multi-core, you must have a greater increase in single core to compensate for less than perfect scaling.

Turbo core does not account for these numbers as when all four cores are being stressed the chip will not go that far beyond its base frequency. Even at a 14% increase over Haswell this new generation are far from impressive on the CPU side over its predecessor. Most people who have Haswell aren't even going to bother upgrading especially considering the requirement of a new platform. The real news in Skylake is the 72 execution units which we have yet to see a result of. Which will be even more so interesting with DDR3 and DDR4 support to see how that influences iGPU performance in gaming having a fatter pipe to work with.

[patrickjp93]

Turbo core does not account for these numbers as when all four cores are being stressed the chip will not go that far beyond its base frequency. Even at a 14% increase over Haswell this new generation are far from impressive on the CPU side over its predecessor. Most people who have Haswell aren't even going to bother upgrading especially considering the requirement of a new platform. The real news in Skylake is the 72 execution units which we have yet to see a result of. Which will be even more so interesting with DDR3 and DDR4 support to see how that influences iGPU performance in gaming having a fatter pipe to work with.

I don't know about you, but at least for my Dad's 4790K, it boosts to 4.4GHz across all cores at stock on AIDA64. No CPU generation has been a worthy upgrade over the previous one since we moved from C2Q to Core. Throughout history until that point every architecture brought increased clocks in addition to IPC gains on the order of 10%. There will not come an architecture so superior to the immediate predecessor as to be a worthy upgrade until we switch from silicon. Without clock boosts it just won't happen.

[wpirobotbuilder]

You're right, but my money is on the usual 10% average increment. No extra cache, no higher clock speeds, and no rumors have indicated a dramatically better ipc (in fact according to Intel they are about to hit the physical limits of silicon, so expecting a higher ipc without moving to graphene is a long shot).

They probably won't be going to graphene, in all likelihood they'll be going to Silicon-Germanium or Germanium transistors. They got a patent on the technology a few years ago:

 

http://www.tomshardware.com/news/intel-sige-transistor-germanium-processor,14664.html

[Sakkura]

I don't know about you, but at least for my Dad's 4790K, it boosts to 4.4GHz across all cores at stock on AIDA64. No CPU generation has been a worthy upgrade over the previous one since we moved from C2Q to Core. Throughout history until that point every architecture brought increased clocks in addition to IPC gains on the order of 10%. There will not come an architecture so superior to the immediate predecessor as to be a worthy upgrade until we switch from silicon. Without clock boosts it just won't happen.

 

That's a motherboard feature, the CPU is not supposed to run at max turbo with all cores loaded. It's supposed to be like this:

 

 

Of course, you'd have to check with the source of that leak to see what clocks the CPU was actually turboing to.

[Sauron]

They probably won't be going to graphene, in all likelihood they'll be going to Silicon-Germanium or Germanium transistors. They got a patent on the technology a few years ago:

 

http://www.tomshardware.com/news/intel-sige-transistor-germanium-processor,14664.html

 

good point

[tsk]

Right now "leaked" benchmarks are showing a 14% increase in performance 4790k vs 6700k in threaded workloads clock for clock.

 

Those benchmark are probably just as fake as the 390x benchmarks from chiphell.

[BiG StroOnZ]

Those benchmark are probably just as fake as the 390x benchmarks from chiphell.

 

They are fake, the original source made them up, if you read the original source they state the following:

 

"We have prepared the following about the potential performance curve of Skylake processors. Intel i7 6700 benchmark test that takes place in this comparison were prepared on the basis of performances given by Intel's processors in the previous year."

 

But people are too lazy to do a Google translate of the original source and figure out where they acquired the information. They just made it up based on previous generational improvements. Nothing in those charts is based on actual samples, just estimated figures based on previous IPC performance increases. Which even then seems far fetched because some of the tests should show absolutely no differences (because there were also "game benches" included in the original source).

[Opcode]

Those benchmark are probably just as fake as the 390x benchmarks from chiphell.

They probably are undoubtedly.