Intel Gen9 Skylake

Paran

Regular
(AlEdit: img src: http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/4)

gen9gt2tuuhh.png


Intel just launched the first two Skylake SKUs i5-6600k and i7-6700k. No Gen9 details from Intel available at the moment, probably at IDF. There are many reviews, here some of them.

http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/
http://www.tomshardware.com/reviews/skylake-intel-core-i7-6700k-core-i5-6600k,4252-9.html
http://nl.hardware.info/reviews/622...getest-benchmarks-igpu-crysis-3-1920x1080-low
http://www.pcgameshardware.de/Core-...ylake-Test-Core-i7-6700K-i5-6600K-1166741/#a3
http://www.pcper.com/reviews/Proces...irst-Enthusiasts/Integrated-Graphics-Performa
http://www.pcworld.com/article/2955...esents-for-gamers-and-enthusiasts.html?page=4
http://www.hardware.fr/articles/940-7/hd-graphics-530-pratique-jeux.html


2xMSAA in hardware definitely works since Gen8, a big difference can be seen in HL2 at Tomshardware. Also Gen9 seems to support Feature Level 12_1.
 
Last edited by a moderator:
Wow.
The performance increase over Haswell is really impressive. In some applications, not all. I do wonder what they did to accomplish that?
Did Intel solve a bottleneck (memory bandwidth, for example), and if they did, then which? Or is it that the performance increase is only in ALU-limited workloads?
 
If only Skylake had included the eDRAM L4... :( I would have had to upgrade my rig immediately. Now I am forced to hold off, blah!
 
Any idea why Skylake...at 14nm....is using 1.3v-1.5v on the vcore..?
I thought the trend with lowering nm..the vcore lowers too?
 
@Ryan Smith I was hoping to see a comparison for integrated graphics with DDR4 having more bandwidth (DDR4 3000 in particular), is something like that in the works by chance?
I can certainly suggest it to Ian. Obviously there was a great deal to get done for this launch and not a ton of time to do it in.
 
I can certainly suggest it to Ian. Obviously there was a great deal to get done for this launch and not a ton of time to do it in.

Yeah, I particularly liked the fact that you used 4 different samples for overckloging assesment! I'm sure it eats a LOT of extra time for not as much benefit (usually you get only slight differences). But I appreciate the fact that it gives more weight to the actual results

Waiting for a high end liquid cooler re-test!
 
Last edited:


There is no info in your Test Setup what drivers versions have been used. No info about the used RAM for Haswell and no timing infos for the used DDR4. Also no further tests like an AF flower comparison, in case Gen9 differs. No HEVC test with some HEVC videos either. Only hardware.fr did a small HEVC test. These things are missing in most tests.
 
I too was hoping for some HEVC tests, particularly HEVC encoding which can't AFAIK is hard to do in real-time using software.
 
I was apparently wrong when I said Skylake "had" to include the FIVR to achieve its power savings compared to previous gens of processors - it seems this is not the case, and that Intel managed anyway somehow.

Has anyone checked if Skylake's clockspeed and core voltage is as dynamic as Haswell's, or more in the vein of Ivy Bridge and predecessors, where clock speed/voltage typically changes on the order of a second or so rather than completely dynamic every tenths of a second. Also, the presence or absence of the new power saving modes introduced with Haswell. This would be really interesting to know!
 
There were, previously, other swings like this. For example, the uncore had a separate clock with Nehalem, coupled to the core with Sandy, and decoupled with Haswell again. (http://www.anandtech.com/show/6355/intels-haswell-architecture/10)
Intel may, for example, have decided that FIVR still is the future, but Skylake has requirements that Intel cannot implement in a FIVR before Skylake's release; or Haswell was a huge verification/validation of the FIVR concept, it is now being improved/refined, but for cost/benefit reasons the current FIVR was not included in Skylake.
Also, the huge (IMHO) voltages being supplied to Skylake suggest that some parts/elements of the IVR may still be there... but this is wild speculation.
 
Can't wait to see what the 72 EU + L4 Skylake can do... I hope we see a high end desktop version of it, I'd expect it to be very popular, especially given the general performance advantage that the L4 can offer + the promise of multi adapter.
 
Wow.
The performance increase over Haswell is really impressive. In some applications, not all. I do wonder what they did to accomplish that?
Did Intel solve a bottleneck (memory bandwidth, for example), and if they did, then which? Or is it that the performance increase is only in ALU-limited workloads?
If you look At Anands results you will see R. Smith actually used slower memory than what was used on Haswell. Causing really bad scores . This is a K chip for enthusiast . Why would you run SKL DDR4 at slower speeds than DDR3 and think your going to get a true IPC difference. Do the Math the DDR3 is running faster than the DDR4. That was the worst review of the day . It was odd also that only 2 sights could overclock to 4.5 GHz both sites being owned by Punch. Toms review was much better as it showed good performance over Haswell. My grandson is buying the Skylake I will wait for kabylake and the ANGEL canyon cpu releasing next year around this time . Hopefully I can get 1 stick of Xcross to put in 1 ddr4 slot . Along with the higher speed DDR4 in 2 slots . I like the density of Xcross use a little e dram with the Angel Canyon CPU and GT4 . I will never have to buy a Dgpu again. The fact that Intel is using gen10 graphics should be Large boost in performance over SKL GT4e. Gen 9 Intels IGPs are looking good . I do believe if Intel wanted to build a Dgpu now they have what it takes. You have to admit the GT2 530 is pretty good
 
Why would you run SKL DDR4 at slower speeds than DDR3 and think your going to get a true IPC difference.
Are you sure about this? I thought the ddr3 was at 1600mhz and the ddr4 was at 2133mhz.
edit - the DDR3 is at 1866 cl9 and the DDR4 is at 2133 cl15.
edit 2 - IIRC DDR3 latency = 4.82ns DDR4 will equal 7.03ns
edit 3 - I was corrected below haswell @ 1600 which makes sense as it's the default mem clock for it.
 
Last edited:
If you look At Anands results you will see R. Smith actually used slower memory than what was used on Haswell. Causing really bad scores . This is a K chip for enthusiast . Why would you run SKL DDR4 at slower speeds than DDR3 and think your going to get a true IPC difference.
I beg your pardon? All of our platforms used memory at their normal, supported frequencies and timings. That's DDR3-1600 for Haswell, and DDR4-2133 for Skylake (with DDR3L-1866 for the DDR3L vs. DDR4 comparison). Those are the highest officially supported frequencies for both platforms.
 
Are you sure about this? I thought the ddr3 was at 1600mhz and the ddr4 was at 2133mhz.
edit - the DDR3 is at 1866 cl9 and the DDR4 is at 2133 cl15.
edit 2 - IIRC DDR3 latency = 4.82ns DDR4 will equal 7.03ns
Latency typically means squat for the IGP (or at least way less important than bandwidth).
FWIW I've wondered about some gpu scores in the reviews - that is not just individual benchmarks but some reviews seemed to show not much of an improvement in general wheras others said it's like a 40% improvement in in just about anything...
Also I think intel only officially supporting up to ddr4-2133 is "slightly" on the low side and disappointing - that's literally the slowest ever built! Sure the server platform doesn't support more neither but it would be quite expected this platform lags behind on that aspect. Granted for the cpu part it doesn't really matter but the IGP could always benefit from it...
 
Last edited:
I was apparently wrong when I said Skylake "had" to include the FIVR to achieve its power savings compared to previous gens of processors - it seems this is not the case, and that Intel managed anyway somehow.

Has anyone checked if Skylake's clockspeed and core voltage is as dynamic as Haswell's, or more in the vein of Ivy Bridge and predecessors, where clock speed/voltage typically changes on the order of a second or so rather than completely dynamic every tenths of a second. Also, the presence or absence of the new power saving modes introduced with Haswell. This would be really interesting to know!

As i said..the voltage is huge/gone upwards. Unless the software are reading wrong. 1.1v jump to 1.3v is not imho only. This already of interests...i thought die shrinks come with lowering voltages?

AFAIK, the fivr will still appear in Skylake mobile processors...it will be now located underside of the package... to help with the heat or otherwise. Did Intel removed it to enable better overclocking since the IPC gains are small?

Skylake desktop power savings are not as impressive....Skylake desktop has lost the low C states?
http://techreport.com/review/28751/intel-core-i7-6700k-skylake-processor-reviewed/5

Hopefully some sites will go into such details wrt to power clock switching...it is interesting.
Just like this one for instance
http://www.legitreviews.com/what-en...ates-do-on-the-core-i7-4770k-haswell-cpu_2217
 
Back
Top