News Intel Teases Rocket Lake: Double-Digit IPC Gains, Cypress Cove Architecture

[shady28]

It isn't worth it to argue with Shady28. S/he cannot accept the fact that on and even playing field there is evidence that shows even Intel's newest (Tiger Lake) cannot beat Zen 3 in IPC. I would assume the Intel reference laptop is going to use RAM that will put the Tiger Lake in the best light so therefore it had LPDDR4-4266MHz RAM. What is ironic is Shady28 was fighting for Comet Lake's numbers to be higher even though we know Comet Lake has a lower IPC than Tiger Lake.

IDK what you are on about, the last few posts have been about Ryzen 5000 (Zen 3) and the memory settings that were used at various benchmark sites.

Has nothing to do with laptop LPDDR4 ratings, although JEDEC does have a standard for that too and it's been around for like 5 years :

" LPDDR4 launches with an I/O data rate of 3200 MT/s and a target speed of 4266 MT/s, "

Ref:

JEDEC Releases LPDDR4 Standard for Low Power Memory Devices

 

[InvalidError]

"While meeting all other specs is still within specs" - that's what I'm talking about. It is not in spec, because it's not in spec, period.

If fuel efficiency standards for your state require 30MPG and your car does 60MPG, your car is in-spec because it handily beats the required spec. 3200-16-16-28 handily beats the JEDEC 3200-22-whatever spec. and therefore is within specs.

 

[shady28]

If fuel efficiency standards for your state require 30MPG and your car does 60MPG, your car is in-spec because it handily beets the required spec. 3200-16-16-28 handily beats the JEDEC 3200-22-whatever spec. and therefore is within specs.

Napkin math?

That is not a JEDEC standard. That is Intel XMP and/or AMD A-XMP, neither of which is a standard. Definitions of the terms you're using below - note where it says 'lets you overclock RAM and compatible DDR3/4 memory.." :

This is how the two are differentiated by a manufacturer :

And with DDR4-3200:

 

[InvalidError]

That is not a JEDEC standard.

There is no such thing as "JEDEC standard" latency, only the DEFAULT latency which is based on the WORST memory JEDEC could conceive being sold to consumers so people can throw parts together, turn the system on and reasonably expect to get boot so they can set the RAM timings to manufacturer spec.

Also, lower latency isn't really overclocking since clocks are unchanged and all of the DRAM's internal processes still happen at the exact same speed - as fast as the DRAM chips are capable of - regardless of what you set timings at. The only thing lower latency changes is how the host schedules commands and data transfers.

 

[shady28]

There is no such thing as "JEDEC standard" latency, only the DEFAULT latency which is based on the WORST memory JEDEC could conceive being sold to consumers so people can throw parts together, turn the system on and reasonably expect to get boot so they can set the RAM timings to manufacturer spec.

Also, lower latency isn't really overclocking since clocks are unchanged and all of the DRAM's internal processes still happen at the exact same speed - as fast as the DRAM chips are capable of - regardless of what you set timings at. The only thing lower latency changes is how the host schedules commands and data transfers.

It's XMP or A-XMP, it's overclocking, stated right from the source of XMP (Intel). You are just making up stuff now.

 

[Rogue Leader]

All,

Those whom are in glass houses should not be throwing stones. Lets keep this cordial.

You know who you are.

 

[spongiemaster]

Coffee lake memory scaling: https://www.tomshardware.com/reviews/best-ram-speed,5951-6.html
For comparison, Pinnacle Ridge memory scaling: https://www.tomshardware.com/review...le-ridge,6064-6.html?region-switch=1604949757 and Matisse: https://www.tomshardware.com/reviews/amd-ryzen-3000-best-memory-timings,6310.html

Looks like Intel benefits as much (often more) from going from 2800 to 3200 MHz (and faster memory in general) as AMD does. Although in either case the improvement going from 2933 to 3200 obviously isn't going to be huge.

The conclusion of the article says pretty much exactly what I said.

For the average computer user that engages in everyday tasks, memory speed doesn't hold as much importance on the Z390 platform. The aggregate difference between the fastest configuration (DDR4-3000 C14) and the slowest configuration (DDR4-2133 C15) in office applications is less than 4 percent. However, if you're a professional that constantly works with the Adobe suite of programs, you could benefit from running DDR4-3200 C14 memory as it delivers a performance gain of around 15.64 percent over the baseline DDR4-2133 C15 configuration.

And they're using 2133 as a baseline, which no one should be buying for a system today. 2133 is practically extinct today, and 2666 should be the baseline starting point.

 

[InvalidError]

It's XMP or A-XMP, it's overclocking, stated right from the source of XMP (Intel). You are just making up stuff now.

As far as timing goes, XMP is absolutely irrelevant: 20+ years ago, manufacturer timings were punched into BIOS by reading timing information off the DIMMs' packaging. Nobody called running low latency RAM at its rated lower latency overclocking back then and it makes no sense to start calling it overclocking now, nothing is being overclocked.

If Intel is your gold standard for what overclocking means, then explain this: even on H410 boards using a locked CPU which should mean no overclocking whatsoever allowed, you can still lower DRAM timings. Intel does not consider the ability to run RAM at its intended lower latency as overclocking.

 

[shady28]

As far as timing goes, XMP is absolutely irrelevant: 20+ years ago, manufacturer timings were punched into BIOS by reading timing information off the DIMMs' packaging. Nobody called running low latency RAM at its rated lower latency overclocking back then and it makes no sense to start calling it overclocking now, nothing is being overclocked.

If Intel is your gold standard for what overclocking means, then explain this: even on H410 boards using a locked CPU which should mean no overclocking whatsoever allowed, you can still lower DRAM timings. Intel does not consider the ability to run RAM at its intended lower latency as overclocking.

Oh yes it does. In order to run over JEDEC standard on H470 and B460 you have to enable XMP (or do it manually). It will just lower the speed to 2666 or 2933 XMP. They all (not just Intel) consider that overclocking. Refer back to images from MSI on the B450 motherboard (JEDEC VS A-XMP) and the one from Kingston (JEDEC vs XMP).

Heres' another source about what that is :

What are XMP profiles and how do I use them?

XMP profiles automatically set your system RAM to higher frequencies. But should you?

www.pcgamer.com

"XMP profiles essentially allow high-performance RAM, which run above industry DDR specifications, to be appropriately set for your system. "

 

[spongiemaster]

In Ananadtech's Ryzen 5000 series review, they have tested the IPC across different CPUs in SPEC and then normalized the performance by GHz. It included Ice Lake and Tiger Lake in the comparison. In SPEC2017 the Ryzen has a 13.2% higher INT IPC and a 7.77% higher FP IPC that the Ice Lake/Tiger Lake CPUs.

Ice Lake is beating Tiger Lake in the overall IPC tests which makes the results rather suspect. With IPC in particular, you can't pick one application to accurately represent an architecture's performance unless that's the only application you run on your computer.

 

[jeremyj_83]

Ice Lake is beating Tiger Lake in the overall IPC tests which makes the results rather suspect. With IPC in particular, you can't pick one application to accurately represent an architecture's performance unless that's the only application you run on your computer.

Ice Lake was beating Tiger Lake in FP but not INT. Realistically Tiger Like is a refined Ice Lake to allow for higher clocks. It would be like what Kaby Lake was to Skylake.

 

[spongiemaster]

Ice Lake was beating Tiger Lake in FP but not INT. Realistically Tiger Like is a refined Ice Lake to allow for higher clocks. It would be like what Kaby Lake was to Skylake.

It's not like Kaby Lake. Cache structure was changed significantly for Tiger Lake from Ice Lake, including 150% more L2 cache which is now non-inclusive of the L1 and 50% more L3 per core. That should result is better IPC across the board at the application level.

 

[jeremyj_83]

It's not like Kaby Lake. Cache structure was changed significantly for Tiger Lake from Ice Lake, including 150% more L2 cache which is now non-inclusive of the L1 and 50% more L3 per core. That should result is better IPC across the board at the application level.

Places that have reviewed both CPUs have said the majority of performance increase between Tiger Lake and Ice Lake is due to pure frequency. The added cache might add a bit to IPC like Zen > Zen+ but that's it.

 

[jeremyj_83]

"
Overall, in the 2017 scores, Tiger Lake actually comes in as the leading CPU microarchitecture if you account both the integer and float-point scores together.

Although the design’s absolute performance here is exemplary, I feel a bit disappointed that in general the majority of the performance gains seen today were due to the higher clock frequencies of the new design.

IPC improvements of Willow Cove are quite mixed. In some rare workloads which can fully take advantage of the cache increases we’re seeing 9-10% improvements, but these are more of an exception rather than the rule. In other workloads we saw some quite odd performance regressions, especially in tests with high memory pressure where the design saw ~5-12% regressions. As a geometric mean across all the SPEC workloads and normalised for frequency, Tiger Lake showed 97% of the performance per clock of Ice Lake.

In a competitive landscape where AMD is set to make regular +15% generational IPC improvements and Arm now has an aggressive roadmap with yearly +30% IPC upgrades, Intel’s Willow Cove, although it does deliver great performance, seems to be a rather uninspiring microarchitecture."
https://www.anandtech.com/show/16084/intel-tiger-lake-review-deep-dive-core-11th-gen/8

 

[spongiemaster]

Places that have reviewed both CPUs have said the majority of performance increase between Tiger Lake and Ice Lake is due to pure frequency. The added cache might add a bit to IPC like Zen > Zen+ but that's it.

That's all great, except none of this has anything to do with my original point, that Willow Cove should not have a lower IPC than Sunny Cove and that any application that shows that should not be used as a basis for architecture comparisons. Those are not typical results. I have said nothing about AMD, so I don't see why you keep bringing them up.

 

[jeremyj_83]

That's all great, except none of this has anything to do with my original point, that Willow Cove should not have a lower IPC than Sunny Cove and that any application that shows that should not be used as a basis for architecture comparisons. Those are not typical results. I have said nothing about AMD, so I don't see why you keep bringing them up.

So you say Spec2006 and Spec2017 shouldn't be used to compare IPC across architectures? While I am not the biggest fan of Spec, especially 2006 since it is so old, it is an industry standard for testing across architectures.