Discussion Intel current and future Lakes & Rapids thread

[jpiniero]

The TDP increase for the Y is said to be due to the doubling of the core, but it seems that they are just planning to shift the stack.

There is going to be Comet Y according to that Dell leak so perhaps those models will be 5.2/7 W. The Icelake U models base is already pretty low, so perhaps 5.2 for Y is asking too much.

Intel said there would be fanless designs but 9 W is really pushing it.

 

[JoeRambo]

The tests you re pointing are not that arithmetic computing heavy but instead are making big usage of instructions related to data manipulation, computed flows are bigger and ther will be more RAM accesses required.

For the LZMA test there is effective heavy arithmetic computation that will slow down the flow of data required from and sent back to the RAM, indeed you can see that the ST computed data flow is 7MB/s for LZMA and about 22MB/s for HTML5 Parse, in MT the computed flows are 30 and 76MB/s respectively...

I don't think "computed data flow" speed is any way indicative of memory access patterns. Compression is very memory latency/bandwith dependent, and things like WinRAR scale real well with decreased memory latency. LZMA might have different characteristics, as it is using different algorithm, but i expect it to still be quite memory hierachy latency sensitive.


EDIT: anyway, my conclusion so far from limited GB4 benching, is that IPC in generic "integer tasks" that matter the most to me improved 10-20% if the load is not touching memory too much. But the current implementation seems to have nasty deficiencies in memory hierarchy, ones that Intel definately needs to fix before releasing this CPU to market. Zen2 with similar memory access latencies, but 32MB of L3 per chiplet can easily compete with ICL as-is.

 

[IntelUser2000]

That's the whole point. The listed Single Threaded turbo is the same, yet 8265U manages to clock higher (and even in multi-threaded workloads, where process shrink should provide the biggest uplift in sustained clocks).

It's actually too early to tell. My Dell XPS 12 doesn't clock at the highest ST Turbo that Core i7 3517U is capable of reaching. Manufacturers usually have more knobs they can adjust than a simple spec sheet suggests. I don't know why Dell does this but it happens.

And the biggest outliers are AES (upgraded fixed-function hardware that won't benefit most workloads) and SFFT, which uses AVX-512. Curiously GEMM also uses AVX-512, yet is slower.

This is why I ignore the AES, Memory Bandwidth, and the FP subsection. I just look at the Integer portion. In Geekbench, you can only look at the Integer section separately when you are not comparing results. The Acer WHL system gets 4894, while the Dell ICL gets 4945.

CNL had some horribad memory to start with, wasn't it running some laptop 2133 stuff? Still 105ns is horrible when previuos gen had 70?

Again, too early to judge anything. This is Dell's WHL system. Its also in the 100ns or so range:
https://browser.geekbench.com/v4/cpu/13233129

This is why benchmarking standards exist, and why laptops are very difficult to test. When its released though, we'll have options from different manufacturers and there will be different review sites each doing slightly different things. Then the picture will become clearer.

 

[HisEvilness]

I hope Quad channel will become a thing on high-end mainstream mobo's, for instance have it on the Maximus and Taichi boards.

 

[Markfw]

I hope Quad channel will become a thing on high-end mainstream mobo's, for instance have it on the Maximus and Taichi boards.

The CPU has to support it. The motherboard just supplies what the CPU is designed to do. For quad channel for AMD, its threadripper.

 

[IntelUser2000]

If you believe Francois(former Intel guy), he says 15% due to uarch and 3% with mitigations.

Though I generally don't trust his claims as his track record is not good, I think it makes sense in this case.

 

[mikk]

It's actually too early to tell. My Dell XPS 12 doesn't clock at the highest ST Turbo that Core i7 3517U is capable of reaching. Manufacturers usually have more knobs they can adjust than a simple spec sheet suggests. I don't know why Dell does this but it happens.

We have evidence that the showcased Dell XPS 7390 with i7-1065G7 can go higher than the 3.48 Ghz in Geekbench.

https://pics.computerbase.de/8/7/8/7/3/5-1080.558f540c.jpg

Likely 2C usage with 3.8 Ghz.

 

[IntelUser2000]

Likely 2C usage with 3.8 Ghz.

That's cool.

Yea, this is why its a waste of time talking about preliminary results. And when you are talking about user-submitted benchmarks like PCMark/3DMark/Geekbench, then there are 100s of things that can be different. You can find a system with the same CPU and configuration performing HALF the performance of the average.

 

[mikk]

https://img1.mydrivers.com/img/20190530/1e0066b500224a96a1cf3d0d7da32ec1.jpg


15W to 25W power switch in windows on the fly?

 

[Edrick]

I hope Quad channel will become a thing on high-end mainstream mobo's, for instance have it on the Maximus and Taichi boards.

The cost difference between high-end mainstream boards and mid-range enthusiast boards (which have had quad channel since 2011ish) is marginal at best.

 

[tamz_msc]

As for Geekbench, somebody on reddit ran a 6700K locked to 3.5 GHz and compared against this ICL CPU:

~9% Integer.

 

[firewolfsm]

15W to 25W power switch in windows on the fly?

This looks like the kind of feature dell would only have on their developer BIOS.

 

[Hans de Vries]

As for Geekbench, somebody on reddit ran a 6700K locked to 3.5 GHz and compared against this ICL CPU:

~9% Integer.

But Ice Lake is (very likely) running at it's boost frequency of 3.9 GHz not 3.48 GHz.
The last image shows the ~18% improvement versus Sky-Lake (Intel Core i5-6600)

When compared to a more recent Whisky Lake we get from earlier in this thread:
The first image below: Intel Core i5-8265U versus Ice Lake, both with 3.9 GHz Boost clock.

Note the correlation in many of the the sub test results. Many test have near identical results.


Whisky Lake versus Ice Lake at 3.9 GHz boost
https://ark.intel.com/content/www/u...-8265u-processor-6m-cache-up-to-3-90-ghz.html
https://browser.geekbench.com/v4/cpu/compare/12219578?baseline=13303489




Sky Lake versus Ice Lake at 3.9 GHz boost

https://ark.intel.com/content/www/u...5-6600-processor-6m-cache-up-to-3-90-ghz.html
Sky Lake average GB4 score: http://browser.geekbench.com/processors/1710
http://browser.geekbench.com/v4/cpu/compare/12915383?baseline=13303489

 

[dmens]

If you believe Francois(former Intel guy), he says 15% due to uarch and 3% with mitigations.

Pro-tip: do not believe a word from that guy.

 

[mikk]

Note the correlation in many of the the sub test results. Many test have near identical results.

Why is this a correlation for you when many tests have identical results? Because Icelake can't have an advantage at the same clock?

Sky Lake versus Ice Lake at 3.9 GHz boost
http://browser.geekbench.com/v4/cpu/compare/12915383?baseline=13303489

Just to make it clear, this Skylake device in your link didn't boost with 3.9 Ghz.

maximum 3553

At 3.9 Ghz Skylake gets better results as you can see from your Whiskey Lake link, keep in mind Whiskey Lake and Skylake both share the same CPU cores, it should be obvious both were running with a different Turbo.

 

[Hans de Vries]

Why do they compare to 2015 CPU's instead of 2018 CPU's ??

Why is this a correlation for you when many tests have identical results? Because Icelake can't have an advantage at the same clock?.

Architectural improvements, certainly if they are big, will influence different benchmarks in wildly different ways. it is extremely unlikely that they are all affected in the same way. What we see here is just the overlap in architecture.

Just to make it clear, this Skylake device in your link didn't boost with 3.9 Ghz. At 3.9 Ghz Skylake gets better results as you can see from your Whiskey Lake link, keep in mind Whiskey Lake and Skylake both share the same CPU cores, it should be obvious both were running with a different Turbo.

One can not trust the frequency measurements of GB4. The GB4 benchmarks are made in short bursts with pauses. Non architectural performance increases can be made by increasing/decreasing frequency and power faster and more accurate depending on the workload. This is why you should compare to a recent CPU and not an 2015 one.

 

[mikk]

One can not trust the frequency measurements of GB4. The GB4 benchmarks are made in short bursts with pauses. Non architectural performance increases can be made by increasing/decreasing frequency and power faster and more accurate depending on the workload. This is why you should compare to a recent CPU and not an 2015 one.

You can trust it because it is a valid ST score for any Skylake based device at ~3.5 Ghz. I know it because I just validated it with my own system.

7700K @3.5 Ghz
7700K @3.9 Ghz
(DDR4-2133 CL15)

If you are saying this 6600K device was running at 3.9 Ghz in the Single-Core test, look at this:

7700K 3.5 Ghz http://browser.geekbench.com/v4/cpu/compare/12915383?baseline=13335188
7700K 3.9 Ghz http://browser.geekbench.com/v4/cpu/compare/12915383?baseline=13335375

The Whiskey Lake sample is running slightly below 3.9 Ghz which is in-line with my 3.9 Ghz test, frequency measurement from GB4 is correct there, pretty sure it is correct for the Icelake device as well.

 

[jpiniero]

Yeah I have to agree that the Dell laptop was boosting only to 3.5.

 

[HisEvilness]

The CPU has to support it. The motherboard just supplies what the CPU is designed to do. For quad channel for AMD, its threadripper.

Yes I am aware of this but from a marketing perspective I can also understand why they would push their high-end products I have a rendering rig and a gaming rig, where my gaming rig is an Intel 8086k and my render rig has an AMD 1600X(also do NDI capture and NAS storage)

The cost difference between high-end mainstream boards and mid-range enthusiast boards (which have had quad channel since 2011ish) is marginal at best.

Yeh but the CPU's cost more I don't really need it on my gaming rig just that quad would yield more fps for 1440p, 4k you are just GPU limited.

 

[Hans de Vries]

You can trust it because it is a valid ST score for any Skylake based device at ~3.5 Ghz. I know it because I just validated it with my own system.

7700K @3.5 Ghz
7700K @3.9 Ghz
(DDR4-2133 CL15)

If you are saying this 6600K device was running at 3.9 Ghz in the Single-Core test, look at this:

7700K 3.5 Ghz http://browser.geekbench.com/v4/cpu/compare/12915383?baseline=13335188
7700K 3.9 Ghz http://browser.geekbench.com/v4/cpu/compare/12915383?baseline=13335375

The Whiskey Lake sample is running slightly below 3.9 Ghz which is in-line with my 3.9 Ghz test, frequency measurement from GB4 is correct there, pretty sure it is correct for the Icelake device as well.

A typical GB4 ST score for the current generation is around 1.25 X MHz at high frequencies and a bit more than 1.25 at lower frequencies.

How much of that is realized depends on the percentage of time that can be run on the max boost frequency, which is again improving over time by better power algorithms.
So I would agree that the 2015 i5-6600 doesn't run at its maximum boost all of the time.

If you now look at the image of the first GB4 compare assuming Ice Lake does run at 3,482 MHz

Whisky lake has 1.27 X MHz.
Ice Lake would have 1.50 X MHz

This is a lot, not impossible, but the reason that I find this unlikely is the correlation between 12 of the sub-test results. They are virtually the same for the two CPU's.
It would be highly coincidental that architectural improvements do benefit so many sub-tests in virtually the same way. All these 12 sub-tests would receive the same 12% IPC improvement (3.9/3.48) .....

I would expect much the performance increase to come from for instance AVX512 implemented in the libraries as shown for the other sub-tests.
The latest versions of ICC will without doubt be able for Spec2006 and Spec2017 and a number of other tests.

But we will see soon enough....

 

[jpiniero]

I should point out that from that leak, the FCT on the 1065G7 is indeed 3.5.

As to why it didn't turbo beyond that for the ST test, I wouldn't be surprised that it's rare enough to get a fully enabled die that the preproduction models were just limited to 3.5 boost for testing purposes.

 

[jpiniero]

Also, Intel mentioned that there are 11 models. My guess is that along with the 6 15 W U models mentioned in the leak, there's 4 Y models (one i7, two i5 and an i3) and two 28 W U models.

 

[mikk]

I should point out that from that leak, the FCT on the 1065G7 is indeed 3.5.

As to why it didn't turbo beyond that for the ST test, I wouldn't be surprised that it's rare enough to get a fully enabled die that the preproduction models were just limited to 3.5 boost for testing purposes.

But not for final SKUs, this must be a software thing, i mean the Bios and drivers are Pre-final at this point.

 

[birdie]

At 3.9 Ghz Skylake gets better results as you can see from your Whiskey Lake link, keep in mind Whiskey Lake and Skylake both share the same CPU cores, it should be obvious both were running with a different Turbo.

I'm so tired of hearing this over and over again. Geekbench has not been updated to properly support Ice Lake (it might be possible that such support is also required from Windows as well), so it's not known whether the turbo frequency that it reports for Ice Lake based systems is correct. I'm inclined to believe that the tested system properly boosted to 3.9GHz and the results are indicative of Ice Lake's real performance (which is far from what Intel led us to believe). I see no reason or justification as to why the tested system was limited to 3.5GHz. Still we're just a few weeks from Ice Lake broad availability, so maybe we should stop speculating about its performance for now and wait for official reviews. Some websites actually test CPUs under the same frequency to show a difference in IPC, so let's hope we'll see such tests as well.

Also, it would be great to see the tests of various CPU generations under patched and unpatched Windows versions (the ones which had been released prior to the whole speculative execution fiasco).

 

[jpiniero]

Seems that the Dell is going to be the only Icelake model on the market for a period of time.

But not for final SKUs, this must be a software thing, i mean the Bios and drivers are Pre-final at this point.

That's a good point, it could be a bug.