Who's buying Skylake-X? (You may now change your vote)

[moonbogg]

So I wonder if these would even be a gaming upgrade over a broadwell 6 core @ 4.3. That French site shows lower gaming IPC, and its kind of significant.

 

[frozentundra123456]

According to jonnyguru, if der8auer used the SuperFlower modular PSU cables he showed in the video, it might just be terrible cable design causing the 12V rail to drop too far under load.

PSU problem causing this uproar would be hilarious.

Yea, definitely a sloppy launch, but I think we need more information from various test sites and real users, as well as some time for bios updates before all the sky is falling talk.

 

[Arachnotronic]

So I wonder if these would even be a gaming upgrade over a broadwell 6 core @ 4.3. That French site shows lower gaming IPC, and its kind of significant.

Probably not worth the upgrade, TBH, unless you just want the new shiny.

 

[BrainEater]

Gimmie Gimmie !!

 

[raghu78]

It looks like Intel rushed the launch of Skylake-X. Maybe this is what competition does even to a dominant company like Intel. Skylake and X299 could benefit from future BIOS revisions but the motherboard manufacturers need to fix the issues with poor VRM design. Anyway Intel should learn from this launch and not repeat the same mistakes with Coffeelake.

 

[Edrick]

It looks like Intel rushed the launch of Skylake-X. Maybe this is what competition does even to a dominant company like Intel. Skylake and X299 could benefit from future BIOS revisions but the motherboard manufacturers need to fix the issues with poor VRM design. Anyway Intel should learn from this launch and not repeat the same mistakes with Coffeelake.

First, Intel didnt make the motherboard vendors use a poor VRM cooling design.

Second, at stock & mild OC speeds, the platform is stable and efficient. (low voltage and heat)

Just because you can not OC a 10C cpu to 4.8ghz without a better VRM cooling solution does not mean they "rushed" this release. In fact, the Xeon versions of Skylake-X cores have been shipping to customers for months already (according to sweepr).

 

[raghu78]

First, Intel didnt make the motherboard vendors use a poor VRM cooling design.

Second, at stock & mild OC speeds, the platform is stable and efficient. (low voltage and heat)

Just because you can not OC a 10C cpu to 4.8ghz without a better VRM cooling solution does not mean they "rushed" this release. In fact, the Xeon versions of Skylake-X cores have been shipping to customers for months already (according to sweepr).

See the video properly. He states the blame is to be shared . Intel pulled in the launch by roughly 3 months . This basically did not give sufficient time to motherboard manufacturers to design proper motherboards. Do not compare Xeon with Skylake-X. Server platforms have much more longer design and validation cycles. btw he said the VRMs are good but the VRM cooling is just really bad. So I think a rev 2 by the motherboard manufacturers which fixes VRM cooling issues should not be a major problem.

 

[Arachnotronic]

See the video properly. He states the blame is to be shared . Intel pulled in the launch by roughly 3 months . This basically did not give sufficient time to motherboard manufacturers to design proper motherboards. Do not compare Xeon with Skylake-X. Server platforms have much more longer design and validation cycles. btw he said the VRMs are good but the VRM cooling is just really bad. So I think a rev 2 by the motherboard manufacturers which fixes VRM cooling issues should not be a major problem.

Seems to me the boards were designed just fine and it was just a matter of the board vendors messing on the cooling.

 

[Edrick]

See the video properly. He states the blame is to be shared . Intel pulled in the launch by roughly 3 months . This basically did not give sufficient time to motherboard manufacturers to design proper motherboards. Do not compare Xeon with Skylake-X. Server platforms have much more longer design and validation cycles. btw he said the VRMs are good but the VRM cooling is just really bad. So I think a rev 2 by the motherboard manufacturers which fixes VRM cooling issues should not be a major problem.

I saw the video and that is 1 persons opinion. I agree with some of what he says, but I put a lot of the blame on the motherboard makers. I mean how hard is it to test VRM heat and apply a proper heat sink. If I really felt it to be an issue with my MB, I could simply point a fan at the VRMs.

 

[Edrick]

Seems to me the boards were designed just fine and it was just a matter of the board vendors messing on the cooling.

Exactly!

 

[imported_ats]

Thanks! All the retailers are advertising 4.3 on all cores, 4.5 on up to 2 cores - I must be misunderstanding something there. Are you at least seeing Turbo Boost 3.0 work for 4.5 GHz on 2 cores?

Also what PSU are you using? I was looking at the EVGA SuperNOVA 850 or 1000 G3 however they are made by SuperFlower which scares me because der8auer was apparently having his temperature issues with a SuperFlower unit, but I am not sure if it was an EVGA or exactly what unit it was.

If you want pure peace of mind, go Seasonic. They are basically the standard by which everything else is judged at this point.

 

[Arachnotronic]

If you want pure peace of mind, go Seasonic. They are basically the standard by which everything else is judged at this point.

Agreed, Seasonic makes great stuff. They serve as ODM for many of the "popular" brands, too.

 

[raghu78]

Seems to me the boards were designed just fine and it was just a matter of the board vendors messing on the cooling.

In your opinion Intel is never to blame.The argument here is about Intel rushing the launch and not providing sufficient time to motherboard partners for design and proper testing.

 

[Arachnotronic]

In your opinion Intel is never to blame.The argument here is about Intel rushing the launch and not providing sufficient time to motherboard partners for design and proper testing.

http://www.jonnyguru.com/forums/showpost.php?p=140451&postcount=3

A sample size of one isn't sufficient to draw such conclusions. This is something that obviously needs to be monitored, but more than one datapoint is needed before we can draw such sweeping conclusions.

The board hardware seems fine, but the "rushed" bit is obvious in that the mobo makers kept giving the reviewers newer BIOSes during review testing.

 

[jj109]

mlc --c2c_latency -c0 -w1 Result: Latency = 16.5 core clocks (4.6 ns)
mlc --c2c_latency -c0 -w2 Result: Latency = 16.1 core clocks (4.5 ns)

mlc --c2c_latency -H -c0 -w1 Result: Latency = 14.7 core clocks (4.1 ns)
mlc --c2c_latency -H -c0 -w2 Result: Latency = 14.9 core clocks (4.1 ns)

Not sure how these numbers compare to Skylake-S or Broadwell-E. Do you have those numbers to compare?

These numbers were run at 4.2ghz core and 2.8ghz northbridge

Not Skylake-S, but Haswell(-S?) @ 4.8/4.4... Wow, 76 clocks to L2<->L2. I guess it does have to go up-and-down the L3 at 36 cycles/ea.
Command line parameters: --c2c_latency -c0 -w1
Latency = 11.8 core clocks (2.9 ns)

Command line parameters: --c2c_latency -c0 -w2
Latency = 75.7 core clocks (18.9 ns)

Wow core-to-core is pretty slow. According to the manual running without arguments does a full suite of tests
Intel(R) Memory Latency Checker - v3.3
Measuring idle latencies (in ns)...
Memory node
Socket 0
0 42.8

Measuring Peak Memory Bandwidths for the system
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using all the threads from each core if Hyper-threading is enabled
Using traffic with the following read-write ratios
ALL Reads : 29581.4
3:1 Reads-Writes : 28253.1
2:1 Reads-Writes : 27367.6
1:1 Reads-Writes : 26495.0
Stream-triad like: 27863.3

Measuring Loaded Latencies for the system
Using all the threads from each core if Hyper-threading is enabled
Using Read-only traffic type
Inject Latency Bandwidth
Delay (ns) MB/sec
==========================
00000 114.07 29061.1
00002 114.92 29101.3
00008 112.63 28528.5
00015 111.14 28643.6
00050 96.87 27858.2
00100 73.92 25218.6
00200 59.07 15920.1
00300 54.11 11798.4
00400 52.64 9392.6
00500 51.81 7845.8
00700 50.05 6131.6
01000 49.20 4738.8
01300 48.10 4033.5
01700 48.49 3380.9
02500 48.05 2747.7
03500 50.01 2281.0
05000 45.99 2111.8
09000 45.85 1798.4
20000 44.86 1608.6

Measuring cache-to-cache transfer latency (in ns)...
Using small pages for allocating buffers
Local Socket L2->L2 HIT latency 17.1
Local Socket L2->L2 HITM latency 18.9

 

[AdamK47]

*Update* Found where I can change the NB clock (Now called CLR in BIOS). Re-ran the test with the NB set to 3200mhz. Big improvement.

I ran the test to see how my system compared. Ran it with my normal everyday clocks/settings.

 

[jj109]

I ran the test to see how my system compared. Ran it with my normal everyday clocks/settings.

Run MLC, too.

https://software.intel.com/en-us/articles/intelr-memory-latency-checker

 

[TheGiant]

does anyone know if the uncore can be oced higher on 6 core than 8/10 core?

Edrick's result with uncore at 3200 Mhz is nice- the goal for intel- get it up to 4000 Mhz and game performance will go significantly up I think

 

[wildhorse2k]

does anyone know if the uncore can be oced higher on 6 core than 8/10 core?

Edrick's result with uncore at 3200 Mhz is nice- the goal for intel- get it up to 4000 Mhz and game performance will go significantly up I think

Uncore 3200 should be more than enough. Ideal memory is G.Skill DDR4 3600 CL15-15-15-35 or 3200 CL14-14-14-34 (depending on your budget) with the fastest latency they sell.

 

[TheGiant]

Are you really sure? The L3 bandwitch went up significantly but its still way below BDW-E.

I think the SKL-X core gaming perf is bottlenecked by the cache performance...

 

[jj109]

L3 is a big pool of eviction caches. HW prefetcher fetches straight into L2 now.

Can't really compare 1:1

 

[JoeRambo]

Not Skylake-S, but Haswell(-S?) @ 4.8/4.4... Wow, 76 clocks to L2<->L2. I guess it does have to go up-and-down the L3 at 36 cycles/ea.
Command line parameters: --c2c_latency -c0 -w1
Latency = 11.8 core clocks (2.9 ns)

Command line parameters: --c2c_latency -c0 -w2
Latency = 75.7 core clocks (18.9 ns)

Wow core-to-core is pretty slow. According to the manual running without arguments does a full suite of tests

.....

Local Socket L2->L2 HIT latency 17.1
Local Socket L2->L2 HITM latency 18.9

Yeah, that was what i was looking for, old stuff with inclusive L3 had 80 and up depending on ring size in that test. Intel now has some real nice latency for some subtle thread synch/communication cases.

So gaming problems are most likely to be memory latency and uncore speed/latency related.

I'd love to see 3200 uncore and CL14 3200 mem run compared to default settings in gaming tests that "suffer".

 

[TheF34RChannel]

Are you really sure? The L3 bandwitch went up significantly but its still way below BDW-E.

I think the SKL-X core gaming perf is bottlenecked by the cache performance...

Looks that way. If you're on Broadwell-E and just game stick with it (to the above poster).

 

[RichUK]

All you need is active cooling on the VRMs.

My Z170 Asrock Pro4s throttled my OC'd 7700k - a fan over the VRM heat sink sorted it.

 

[Edrick]

Are you really sure? The L3 bandwitch went up significantly but its still way below BDW-E.

I think the SKL-X core gaming perf is bottlenecked by the cache performance...

All the reviews of skylake-x I have seen have been tested with stock 2400mhz northbridge. I have shown that just upping that to 3200mhz shows a measurable gain. I wonder when we will see reviews with this value overclocked.