Intel Skylake / Kaby Lake

[Dresdenboy]

Yep. This graph that ShintaiDK posts in every thread is also complete BS, it only applies to specific server applications utilizing AVX/AVX2 etc etc.

AVX in Dothan?

BTW if you count AVX instruction throughput, then doing 8 float ops with one AVX instruction has the same IPC as doing 4 with SSE. This btw is a problem with using Cinebench numbers.

Cinebench - while being cache friendly and thus somewhat independent of memory subsystem - is a bad way of trying to get a clue of IPC. That it's a "singlethreaded" benchmark doesn't qualify it for that task. It measures the efficiency of SIMD calculations, loop predictors and buffers, and some stuff along the path through the FPUs.

 

[ShintaiDK]

The graph deals with IPC, not performance.

FWIW I went from i920 to i4770K (both not OC) and saw some of my programs getting twice faster, but these are very specific (using GMP lib which uses new instructions).

And?

Point is people felt Nehalem to SB was a big upgrade because they also went up in frequency, both stock and OC.

The IPC advancements since Dothan have been stable. So if one uarch feels much more than another its not really due to IPC. But due to frequency.

 

[Fjodor2001]

And?

...And the graph is thus incorrect, since IPC has not improved at the rate that the graph indicates.

 

[JM Popaleetus]

7-14% over Broadwell? That means it's a sizable jump over the previous desktop chip consumers could get - which was Haswell. Right?

No, over Haswell.

 

[ShintaiDK]

...And the graph is thus incorrect, since IPC has not improved at the rate that the graph indicates.

I dont see your numbers.

 

[AtenRa]

I've heard from a little bird that some chips may be able to operate at 5GHz stable (running applications) on water.

 

[ShintaiDK]

Sure.

That graph shows Haswell being a 10% IPC increase over Ivy Bridge.

Lets look at the official Anandtech review for Haswell, from June 2013.

Same clockspeeds of the 3770k (Ivy Bridge) and 4770k (Haswell)

Thats 5% faster IPC, not the 10% from ShintaiDK's BS graph.

Here's the link to the official Anandtech Haswell review, if you want to compare other benchmarks:

http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6

If you want to measure the actual IPC you want to remove as many outside factors as possible.

CB ST its 7.2% Povray ST its 14%, PCmark 7 is 7.6%, Truecrypt is 18.9% and VS2012 compile is 13% arcording to your link.

 

[Fjodor2001]

...And the graph is thus incorrect, since IPC has not improved at the rate that the graph indicates.

I dont see your numbers.

Sorry, I checked your graph again and the number appear to be quite accurate. I was too quick and assumed the numbers from "dark zero" were extracted from your graph, which they weren't. The IPC increase numbers that "dark zero" specified seem incorrect though.

 

[Fjodor2001]

N/A

 

[Fjodor2001]

Just wondering, how have you calculated at the "cumulative line"? IPC_YEAR_1 + [...] + IPC_YEAR_N, or IPC_YEAR_1 * [...] * IPC_YEAR_N?

Looks like it's the latter, since e.g. Westmere -> Ivy Bridge indicates a jump from 1.3->1.5 in cumulative IPC but SB increased 10% and IB 5%. I.e. if you had just added them up it would be 1.3 + 0.1 + 0.05 = 1.45.

So assuming it's the latter, shouldn't the cumulative line be exponential, if yearly IPC increase remains constant over time?

 

[ShintaiDK]

Just wondering, how have you calculated at the "cumulative line"? IPC_YEAR_1 + [...] + IPC_YEAR_N, or IPC_YEAR_1 * [...] * IPC_YEAR_N?

Looks like it's the latter, since e.g. Westmere -> Ivy Bridge indicates a jump from 1.3->1.5 in cumulative IPC but SB increased 10% and IB 5%. I.e. if you had just added them up it would be 1.3 + 1.1 + 1.05 = 1.45.

So assuming it's the latter, shouldn't the cumulative line be exponential, if yearly IPC increase remains constant over time?

You can see the percentage increases yourself and do a doublecheck.

The 15% Merom is about as steep as the 11% Haswell.

 

[Sweepr]

Again. Skylake is going to shine in the mobile market. But I predict, for desktops it's going to be another 7-14% release.

Definitely. If this score represents the average Core i5 6200U score then we're looking at +20% CPU performance bump for Skylake-U.

Now remmember a certain person (doing damage control in other thread) told us Carrizo would be faster at 15W TDP, let's see how this lie holds up when the first independent reviews come out.

I've heard from a little bird that some chips may be able to operate at 5GHz stable (running applications) on water.

Interesting, I have heard similar things too. Let's hope it's true.

 

[Sweepr]

Intel preparing to release notebook Skylake in October

Following the upcoming release of desktop Skylake processors at Gamescom in Germany in August, Intel's Skylake processors for notebooks will start showing up in October. The CPU giant is planning to launch a total of 18 notebook Skylake processors in the fourth quarter, looking to help boost notebook demand. However, because many notebook brand vendors are still suffering from inventory issues, they are continuing to adjust their schedules for Skylake products to minimize the impact.

The sources pointed out that many notebook brand vendors saw their inventories building up quickly during the first half because their sales in Europe were stagnated and emerging markets such as China were performing weaker than expected. These vendors are seeing their notebook inventories digesting in a rather slow pace, which is affecting their plans for new products and orders.

Intel is set to release its K series processors and Z170 chipsets for desktops initially at Gamescom and will release eight more mid-range and high-end processors later.

The notebook supply chain has already prepared to start mass-producing Skylake-based notebooks in September and Hewlett-Packard (HP) is currently ahead of others in the preparation, while most of the rest are still busy trying to clear up their inventories.

As for Windows 10, the sources are skeptical about the operating system's ability to give a major boost to notebook demand and the third quarter is unlikely to see a usual seasonal uptick in demand. These vendors are expected to see some recovery in the fourth quarter after clearing their inventory.

Several notebook brand vendors also plan to release Skylake-based Chromebooks in October and will not release the entry-level Celeron-based models until the first quarter of 2016.

www.digitimes.com/news/a20150722PD213.html

It's happening.

 

[Brunnis]

Core i5 6200U (Skylake-U, 2.4GHz base clock) Geekbench Score
Single-thread: 2928
Multi-thread: 6195

http://browser.primatelabs.com/geekbench3/3017270

To put things in perspective, the first leaked score is already faster than the fastest Core i5 5200U score listed.
Average Core i5 5200U (Broadwell-U, 2.2GHz base clock, 2.7GHz Turbo) Geekbench score is ~2500/5000.
Average Core i5 4200U (Haswell-U, 1.6GHz base clock, 2.6GHz Turbo) Geekbench score is ~2500/4800.

This will be the main processor of choice of many Windows 10 notebooks, including (probably) the Surface Pro 4.

Here's a comparison with my Dell laptop with an i5-5300U. Doesn't look too promising, to be honest, but I suspect that there are too many variables to really draw any sweeping conclusions. There are some strange differences in the results, particularly when it comes to the 5300U's large win in some of the FP tests...

http://browser.primatelabs.com/geekbench3/compare/3017270?baseline=2434031

 

[Sweepr]

Here's a comparison with my Dell laptop with an i5-5300U. Doesn't look too promising, to be honest, but I suspect that there are too many variables to really draw any sweeping conclusions. There are some strange differences in the results, particularly when it comes to the 5300U's large win in some of the FP tests...

http://browser.primatelabs.com/geekbench3/compare/3017270?baseline=2434031

Except your score is the fastest Core i5 5300U out there.
Average Core i5 5300U scores lower than that, ~2600/5400.
Also there will be a faster Core i5 6300U SKU, Core i5 6200U succeeds Core i5 5200U.

 

[Fjodor2001]

Intel preparing to release notebook Skylake in October

www.digitimes.com/news/a20150722PD213.html

It's happening.

So Intel will release them in October, but actual product availability may be later due to inventory issues (sales channels and stores already full of "old stuff")?

I wonder if this includes both Skylake U and Y. The article doesn't say...

 

[Sweepr]

So Intel will release them in October, but actual product availability may be later due to inventory issues (sales channels and stores already full of "old stuff")?

I wonder if this includes both Skylake U and Y. The article doesn't say...

No, there will be actual notebooks in October.
Surface Pro 4 is rumoured to launch by then, DigiTimes mentioned Chromebooks too and at least some OEMs (HP rumoured to be ahead of others, ASUS was expecting Skylake notebooks in Q3, etc.) might have Windows notebooks on the market. The Macbook Air also needs a refresh badly, expect new design + new screen soon after Skylake-U launch.
I think Skylake-U arrives first this time, or maybe simultaneously with Skylake-Y.

 

[ShintaiDK]

Asus for example will release right away.

Its all up to the individual OEMs.

 

[Fjodor2001]

Asus for example will release right away.

Its all up to the individual OEMs.

Yes, I know. It's just that the article said:

"However, because many notebook brand vendors are still suffering from inventory issues, they are continuing to adjust their schedules for Skylake products to minimize the impact. [...] The sources pointed out that many notebook brand vendors saw their inventories building up quickly during the first half because their sales in Europe were stagnated and emerging markets such as China were performing weaker than expected. These vendors are seeing their notebook inventories digesting in a rather slow pace, which is affecting their plans for new products and orders."

So I assume some OEMs will not release Skylake based laptops until later, when they have cleared their sales channels of old inventory.

Good that there at least are some OEMs that will have products available at release though.

 

[Brunnis]

Except your score is the fastest Core i5 5300U out there.
Average Core i5 5300U scores lower than that, ~2600/5400.

Didn't know it was the fastest one... No idea why either. It was a clean Windows 7 installation, but no other "special treatment". But that's great, I guess. And perhaps puts the Skylake score in a little better light.

However, it sort of proves that it's pretty pointless to compare until you have a decent samples size.

 

[Essence_of_War]

Just wondering, how have you calculated at the "cumulative line"? IPC_YEAR_1 + [...] + IPC_YEAR_N, or IPC_YEAR_1 * [...] * IPC_YEAR_N?

Looks like it's the latter, since e.g. Westmere -> Ivy Bridge indicates a jump from 1.3->1.5 in cumulative IPC but SB increased 10% and IB 5%. I.e. if you had just added them up it would be 1.3 + 0.1 + 0.05 = 1.45.

So assuming it's the latter, shouldn't the cumulative line be exponential, if yearly IPC increase remains constant over time?

You can see the percentage increases yourself and do a doublecheck.

The 15% Merom is about as steep as the 11% Haswell.

I'll do Fjodor's homework for them:

From the chart

D-M
15% YoY ; 1.15 overall

M-P
~2% YoY; ~1.17 overall

P-N
~12% YoY; ~1.32 overall

N-W
0% YoY; ~1.32 overall

W-S
10% YoY; ~1.45 overall

S-I
~5% YoY; ~1.49 overall

I-H
~10% YoY; ~1.68 overall

H-B
~5% YoY; ~1.75 overall

So in vector form the YoY increases are:

[1.15 1.02 1.12 1.00 1.10 1.05 1.10 1.05]

and a cumulative product gives to two decimal places:

[1.15 1.17 1.31 1.31 1.44 1.52 1.67 1.75]

which lines up pretty closesly with my eyeballing of the graph. So I'd say it looks like the graph is done "correctly" in that it counts the accumulation of YoY percentage gains as a product.

 

[crashtech]

I though I saw a chart that indicated the US availability of Skylake-K will occur in mid-September. Do we have any confirmation of that?

 

[ShintaiDK]

I though I saw a chart that indicated the US availability of Skylake-K will occur in mid-September. Do we have any confirmation of that?

K model is august 5th. Non K models is late september.

 

[Nothingness]

The IPC advancements since Dothan have been stable.

I wouldn't qualify 15% vs 10% for tocks (30% less improvements), and 2% vs 5% for ticks (>100% more improvements) as stable OTOH if you combine tock and tick it looks much more stable (which makes sense: given that Intel doesn't seem to increase frequency any more with new process, so they have to improve the micro-arch to provide more performance).

So if one uarch feels much more than another its not really due to IPC. But due to frequency.

It's also due to ISA improvements as I mentioned.

 

[Nothingness]

Core i5 6200U (Skylake-U, 2.4GHz base clock) Geekbench Score
Single-thread: 2928
Multi-thread: 6195

http://browser.primatelabs.com/geekbench3/3017270

To put things in perspective, the first leaked score is already faster than the fastest Core i5 5200U score listed.
Average Core i5 5200U (Broadwell-U, 2.2GHz base clock, 2.7GHz Turbo) Geekbench score is ~2500/5000.
Average Core i5 4200U (Haswell-U, 1.6GHz base clock, 2.6GHz Turbo) Geekbench score is ~2500/4800.

The 6200U score is for the 64-bit version of Geekbench, while the 2500/5000 you quote is for 32-bit versions. 64-bit scores for 5200U are about 2800/5500.

It should also be noted the improvement seems to mostly come from memory.

See this for instance: http://browser.primatelabs.com/geekbench3/compare/2999148?baseline=3017270