Intel Skylake / Kaby Lake

[Raftina]

First # on the left is the place indicator; the second # from left is bandwidth in GB/sec; the third # from left is DDR RAM speed; fourth # is the most important RAM timing, CAS latency; fifth # is total or true latency.

DDR3:
10. PC3-14,900: 1,866 DDR @ CL 9= 09.64 ns

DDR4:
04. PC4-17,000: 2,133 DDR @ CL 16= 15.00 ns
09. PC4-21,300: 2,666 DDR @ CL 13= 09.75 ns

lmfao, so now you're willing to admit, 6 or 7 pages later, that what I pointed out in my first post was correct, that DDR3 1,866 CL 9 and DDR4 2,133 CL 16 are equivalents, making what I said to begin with, that some site using DDR4 2,666 was giving an advantage to the CPU that used DDR4.

According to the table you provided:
DDR3 1866 @ CL9: 09.64 ns
DDR4 2133 @ CL16: 15.00 ns

I fail to see how these are equivalents, since their supposed true latency, according to you, is massively in favor of the DDR3.

Furthermore:
DDR4 2666 @ CL13: 09.75

The fastest DDR4 2666 in your table gives a true latency that is worse than the DDR3 used, though the advantage is minor here.

It seems to me that the test that used DDR3 1866 CL9 vs DDR4 2666 CL13 is the one that compared equivalent RAM, while the one that used DDR3 1866 CL9 vs DDR4 2133 CL 16 gave a massive advantage to DDR3 in terms of RAM latency.

 

[Stoneburner]

So is broadwell betterfor HTPC purposes?

 

[Abwx]

Intel® Architecture, Code Name Skylake Deep Dive: A New Architecture to Manage Power Performance and Energy Efficiency

New presentation: https://hubb.blob.core.windows.net/e5888822-986f-45f5-b1d7-08f96e618a7b-published/9aa257d4-09ab-44f5-8055-d3dd46e3331e/ARCS001%20-%20SF15_ARCS001_102f.pdf?sv=2014-02-14&sr=c&sig=7gstEjY8NfxvBC30cVU1zMgygj8BwIXCBi%2BmxwYQnTQ%3D&se=2015-08-19T01%3A40%3A05Z&sp=rwd

Edit:

Nice paper, although i do not agree with their starting point to legitimize extended TDPs, the formulae page 7 is too much stretched :

P~V2.f.Cdynn + leakage(V) ~ f3

They are saying that power scale as a cube at any frequency, wich is contradicted by measurements, they had to stretch the equation otherwise the optimum point would be at much lower powers, leakage is no that high.

Real optimum is when the compute power is equal to the rest of system power, that s a classical case of optimisation under constraint as learned in economics or enginering schools.

Page 35 they display what i have explained ad nauseam to point the difficulty to make accurate power measurements, that is, that the SoC can drain additional power from the battery to not overload the laptops DC input circuitries and PSUs..

 

[jj109]

If f ~ V plugging that into the power estimation we get P ~ f^3 at fmax.

What's your beef with that?

Real optimum is when the compute power is equal to the rest of system power, that s a classical case of optimisation under constraint as learned in economics or enginering schools.

Lol okay.

 

[Arachnotronic]

My Skylake parts just got here

- Core i5 6600K
- EVGA GTX 970
- ASUS Z170M-Plus
- Coolermaster Hyper 212+

I also have a 16GB kit of Crucial Ballistix DDR4-2400, and the parts will be stuffed into an SG09 case, replacing my i5 4590 + H81 setup.

Looking forward to seeing how far I can push this sucker.

 

[LTC8K6]

So is broadwell betterfor HTPC purposes?

I don't think we can say that, unless we are talking about using the igp?

 

[myocardia]

According to the table you provided:
DDR3 1866 @ CL9: 09.64 ns
DDR4 2133 @ CL16: 15.00 ns

I fail to see how these are equivalents, since their supposed true latency, according to you, is massively in favor of the DDR3.

I've been saying for more than 10 pages now that latency is a small part of the equation, when comparing one type of DRAM to another. See this post for more details, including charts from Anandtech's main page, proving it: http://forums.anandtech.com/showpost.php?p=37632136&postcount=3688

Furthermore:
DDR4 2666 @ CL13: 09.75

When comparing two sticks of the same type of DRAM, true latency falls as speed rises, unless of course you use a high speed stick with extremely low CAS latency, like your example, which makes it plummet.

The fastest DDR4 2666 in your table gives a true latency that is worse than the DDR3 used, though the advantage is minor here.

Yet again, bandwidth is what you want to look for, when comparing two disparate types of DRAM, not latency. Higher bandwidth improves both bandwidth and latency, instead of just latency. Find out more here: http://forums.anandtech.com/showpost.php?p=37632136&postcount=3688

It seems to me that the test that used DDR3 1866 CL9 vs DDR4 2666 CL13 is the one that compared equivalent RAM, while the one that used DDR3 1866 CL9 vs DDR4 2133 CL 16 gave a massive advantage to DDR3 in terms of RAM latency.

That's because you weren't aware that DDR4 2,666 CL 13 provides more than a 30% speed advantage, where speed equals total bandwidth and total latency combined. See here for more details: http://forums.anandtech.com/showpost.php?p=37632136&postcount=3688

 

[ioni]

Anyone know what's up with this 6700k/mobo combo on newegg?
http://www.newegg.com/Product/Produ...oiKoEQMFc0ORGE_6QFuAMaAiQF8P8HAQ&gclsrc=aw.ds

 

[Burpo]

Generic? lol

 

[cmdrdredd]

Nice find lol. What the heck mobo is generic?

 

[richierich1212]

Anyone know what's up with this 6700k/mobo combo on newegg?
http://www.newegg.com/Product/Produ...oiKoEQMFc0ORGE_6QFuAMaAiQF8P8HAQ&gclsrc=aw.ds

That's been there for a few days now. Not sure what's up with that combo.

 

[DrMrLordX]

Generic huh, sounds like a winner to me! A fierce competitor to Bungbungame?

 

[Burpo]

It's worse than that.. When you try to add it to your cart, you get this..

We’re sorry, the item you selected is a promotional item that is not available for individual sale.
You can have this item by purchasing one of the following bundles:

http://www.newegg.com/Shopping/FailedAddtoCart.aspx?errorType=BOM&item=N82E16813011002

Forcing you to the next page where you can buy an i5 6600K CyberPower PC for $847.. (not i7 6700K)
http://www.newegg.com/Product/BOM.aspx?Item=83-230-087

 

[gothuevos]

Any reviews of Skylake and SLI together?

I know it's super early, but more than a few posts on different forums (even one guy here) having serious performance issues with SLI on Skylake builds.

 

[Shivansps]

AMD Radeon R5 230 1GB GDDR5 16X PCIe Video Card
+$9.00] NVIDIA GeForce GT 720 VC-247-101 1GB Video Card

Now thats is what is called a DOWNGRADE.

 

[Nothingness]

My Skylake parts just got here

- Core i5 6600K
- EVGA GTX 970
- ASUS Z170M-Plus
- Coolermaster Hyper 212+

I also have a 16GB kit of Crucial Ballistix DDR4-2400, and the parts will be stuffed into an SG09 case, replacing my i5 4590 + H81 setup.

Looking forward to seeing how far I can push this sucker.

Only 2400 for the RAM? Reading this thread I had come to the conclusion that one should pick 2666 at the very least. Did I misunderstand?

Anyway that should be a nice machine

 

[Arachnotronic]

Only 2400 for the RAM? Reading this thread I had come to the conclusion that one should pick 2666 at the very least. Did I misunderstand?

Anyway that should be a nice machine

I bought the 16GB kit for my X99 system but wound up going with more memory for it. I had intended to return this kit, but I stupidly forgot to do so within Newegg's 30 day return window.

If I didn't already have this (very expensive, sadly) DDR4-2400, you can bet I'd have bought faster memory.

 

[Absolute0]

My Skylake parts just got here

- Core i5 6600K
- EVGA GTX 970
- ASUS Z170M-Plus
- Coolermaster Hyper 212+

I also have a 16GB kit of Crucial Ballistix DDR4-2400, and the parts will be stuffed into an SG09 case, replacing my i5 4590 + H81 setup.

Looking forward to seeing how far I can push this sucker.

Very cool. Any pictures when you're done?

Good luck with the clocking and hope you got nice Si

 

[Arachnotronic]

Very cool. Any pictures when you're done?

Sure! For full transparency, though, I'm going to hold off a week before building it. If the 6700K miraculously shows up in stock within that time, I'll return the 6600K and pick up a 6700K. If not, I'll go ahead with the system as-is

Good luck with the clocking and hope you got nice Si

Thanks!

 

[witeken]

The first article about the Skylake architecture presentation is up. From EETimes. The thing about the FIVR, it confuses me.

"In an interesting side note, Intel removed the integrated voltage regulator from the new SoC to hit tablet power consumption levels. “We didn’t have time to be flexible enough to have it in higher SKUs -- we were limited in time,” said Mandelblat."

Does the mean FIVR is in Core m and Core U, but not in "higher (TDP) SKUs"?

http://www.eetimes.com/document.asp?doc_id=1327450

 

[Dresdenboy]

Intel® Architecture, Code Name Skylake Deep Dive: A New Architecture to Manage Power Performance and Energy Efficiency

New presentation: https://hubb.blob.core.windows.net/...XCBi+mxwYQnTQ=&se=2015-08-19T01:40:05Z&sp=rwd

File seems to be down (authentification timeout). Luckily I downloaded it on my smartphone ^^.

 

[IntelUser2000]

File seems to be down (authentification timeout). Luckily I downloaded it on my smartphone ^^.

It will end up being permanently there when the official presentation is over.

witeken: The only way it makes sense if that's said right is that Core M doesn't have it(to reach "Tablet power levels"), but U-series does because it has same TDP as the previous generation, and of course desktop models don't have it.

Core M TDP goes from 4.5W to 4W. Perhaps it was a stretch to have good performance at 4.5W with FiVR? Maybe that's why it needed higher TDP to be acceptable? How did they overcome the whole thickness and space issue? It's quite substantial for a Tablet.

eDRAM changes: I find the eDRAM changes interesting. If the thread in RWT is true, "reduce the average latency of memory accesses and increase the memory’s effective bandwidth." This is probably a perfect fit for upcoming, or even further future XPoint pairings. Abandon the DRAM completely and use eDRAM as a cache. This isn't exactly far-fetched, because someone pointed that all future exascale systems use that configuration: eDRAM/HBM/HMC as cache to NVRAM, with DRAM non-existent.

That is what would make the computer worth upgrading from a Sandy Bridge, not measily 30% improvement. SNB is way more than enough for vast, vast majority of people.

Heck, even a 5% improvement from my 2600K would be ok if a optimized software with NVRAM that replaces normal DRAM, plus "fast DRAM" like eDRAM exists. That would change the paradigm of computing! Near-instant boot that completely forgoes stupid things like sleep mode and hibernate, and uses truly zero power when not in use, and hundred+ gigabytes of memory on a laptop, near instant loading of applications, game level loading, virus scanning etc.

 

[ClockHound]

So...IDF is here. New Intel TV show is here. Lotsa spider bots are here. Can Skylake-S supply in the Americas be far behind?

Or maybe they're in post-production for another TV series: Skylake-S - Overclocking for Xmas.

 

[ShintaiDK]

So is broadwell betterfor HTPC purposes?

HTPC as in home theater PC with 4K movie playback? No, Skylake is better.

The only better Broadwell desktop model is the C model in terms of IGP performance.

 

[phillyman36]

Just out of curiosity are there any EU stores that ship to the US that have i7 6700k in stock? Any link?