Discussion Intel current and future Lakes & Rapids thread

Page 237 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

itsmydamnation

Platinum Member
Feb 6, 2011
2,743
3,075
136
Here's a comparison of Sunny cove and Zen 2 architecture for people who care about these things...

I read it , it has many subtle but important things wrong about both mirco uarches, not worth the time.
 

Elfear

Diamond Member
May 30, 2004
7,096
640
126
The 4.7 GHz 3900XT vs the 4.7 GHz 1165G7 in userbenchmark:




Looks like a 25% lead IPC for Willow Cove. That's similar to Ice Lake vs Renoir.


Isn't Userbenchmark the site that turned out to be the incognito marketing arm for Intel? IIRC, they keep changing the benchmark to better suit Intels diminishing list of strengths.
 

Bouowmx

Golden Member
Nov 13, 2016
1,138
550
146
Isn't Userbenchmark the site that turned out to be the incognito marketing arm for Intel? IIRC, they keep changing the benchmark to better suit Intels diminishing list of strengths.
The uproar is over the aggregate result (how the 1-, 2-, 4-, etc. core results are combined into a index relative to some top processor, ex. Intel i9-9900K or 10900K) and the contributed processor descriptions. The underlying individual results still have some merit (though I can't find the benchmark's method of measurement).
 
  • Like
Reactions: Elfear

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,629
136
Rocketlake rumor:
Basically beating 10900K because cores don't scale linearly. 5% improvement and rest in gaming is anemic though.

One of the reddit commenters said that the translation is wrong and that this is just a forum member speculating on Rocketlake and is not a leak. I don't know who's correct but just some additional info from someone who supposedly can ready the referenced post.
 

majord

Senior member
Jul 26, 2015
433
523
136
Lol
These leaks are getting wilder by the day. What's next , 40% higher IPC 6ghz?
 
Last edited:

SAAA

Senior member
May 14, 2014
541
126
116
More than a leak it's just speculation and it isn't even that wild. Like the cores themselves have at least icelake's 18% IPC, then 5GHz+ for an 8 core doesn't sound impossible with the TDPs they are going for those days, also look at leaked Tiger lake doing 4.7 turbo speed on mobile skus with supposedly broken 10 nm... 14 nm +++ should get those clocks without blinking. 5% faster at conventional programs and up to 15% in games would be expected results for a 11900k vs 10900k, consider it's 8 vs 10 cores too. What I'd like to see is how Intel marketing spins 8 being better than 10 to the mainstream buyer, not to talk about competition. They have to make IPC the next "GHz" thing, only avoid mentioning Apple like the plague then xD
 
  • Like
Reactions: spursindonesia

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
More than a leak it's just speculation and it isn't even that wild. Like the cores themselves have at least icelake's 18% IPC, then 5GHz+ for an 8 core doesn't sound impossible with the TDPs they are going for those days, also look at leaked Tiger lake doing 4.7 turbo speed on mobile skus with supposedly broken 10 nm... 14 nm +++ should get those clocks without blinking. 5% faster at conventional programs and up to 15% in games would be expected results for a 11900k vs 10900k, consider it's 8 vs 10 cores too. What I'd like to see is how Intel marketing spins 8 being better than 10 to the mainstream buyer, not to talk about competition. They have to make IPC the next "GHz" thing, only avoid mentioning Apple like the plague then xD
How is a speculation more than a leak? You lost me after the first words.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
25% higher IPC is really high (RKL has smaller cache than TGL), only +5% SSE performance sounds underwhelming and I doubt it. I don't believe in a 25% IPC uplift either.

Probably means +5% in multi-threaded workloads.
 

FriedMoose

Member
Dec 14, 2019
48
28
51
Isn't Userbenchmark the site that turned out to be the incognito marketing arm for Intel? IIRC, they keep changing the benchmark to better suit Intels diminishing list of strengths.
Sunny Cove beats renoir by 20% in Integer and 27.5% in floating point when you run SPEC2017 1T. Not sure how Willow Cove being 25% ahead of desktop Zen 2 is unfathomable for some people here.
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,573
136
Sunny Cove beats renoir by 20% in Integer and 27.5% in floating point when you run SPEC2017 1T. Not sure how Willow Cove being 25% ahead of desktop Zen 2 is unfathomable for some people here.
It isn't unfathomable in itself, it's just that Userbench itself is highly suspect, as they have continuously modified their scoring system to intel's benefit ever since Zen 1 was released.

The latest most obvious change was when they added the memory-latency subsection that heavily contributes to the final score just when 3300X was launched.

I mean, just compare 3300X to 7600K. Supposedly 7600K has a 2% better "effective speed" only because of memory latency, despite having 13% worse score. That's despite the fact that this latency advantage doesn't even matter in games, as 3300X has vastly bigger caches. 3300X is actually highly competitive even with a (non-overclocked) 7700K, let alone 7600K, beating it in FPS minimums quite often.

Anandtech review:
1593602898820.png

Tech-spot review
Average.png

And this is versus a 7700K. According to UserBench even a 7600K with only 4 threads an a 4.20 Ghz turbo (vs 4.5 of 7700K) is "effectively a better processor" than a 3300X.

Meanwhile, there isn't a single game where the 7600K wins vs 3300X in Anandtechs bench comparison. In productivity apps it's 10-80% slower, but according to Userbench it's "effectively better". And that is only the case because they added a new score-metric that didn't even exist a little more than a month ago just to make the 3300X look bad.
 
Last edited:

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,574
146
Sunny Cove beats renoir by 20% in Integer and 27.5% in floating point when you run SPEC2017 1T. Not sure how Willow Cove being 25% ahead of desktop Zen 2 is unfathomable for some people here.
There was some related discussion about this yesterday on Twitter, it's worth noting the Renoir system tested (the Acer Swift SF314-42) was using DDR4-3200cl32 memory, and I don't think this is a particularily fair comparison given how far behind JEDEC spec this memory considering unlike LPDDR4X-4266 memory such horrible latency does not come with the additional benefit of improved bandwidth for the CPU.

JEDEC spec for DDR4-3200 is CL20 or CL22 (both are allowed), and so I can't justify repeating these numbers as IPC figures when not taking that into consideration.
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,573
136
There was some related discussion about this yesterday on Twitter, it's worth noting the Renoir system tested (the Acer Swift SF314-42) was using DDR4-3200cl32 memory, and I don't think this is a particularily fair comparison given how far behind JEDEC spec this memory considering unlike LPDDR4X-4266 memory such horrible latency does not come with the additional benefit of improved bandwidth for the CPU.

JEDEC spec for DDR4-3200 is CL20 or CL22 (both are allowed), and so I can't justify repeating these numbers as IPC figures when not taking that into consideration.
Besides, it was 17% faster in SPECint results (17.5% if you account for frequency) not 20%. The SPECfp comparison was correct.
 
  • Like
Reactions: spursindonesia

itsmydamnation

Platinum Member
Feb 6, 2011
2,743
3,075
136
Sunny Cove beats renoir by 20% in Integer and 27.5% in floating point when you run SPEC2017 1T. Not sure how Willow Cove being 25% ahead of desktop Zen 2 is unfathomable for some people here.
how much power does sunny cover use while beating it? on my 4700u i cant make a core us above 6 watts while doing AVX2 1t stress tests. If AMD let a single core hit ~13watts it would probably be up at ~4.7 ghz like a 3900X and then Sunny would be under 10% better if not equal.
 
  • Like
Reactions: lightmanek

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Besides, it was 17% faster in SPECint results (17.5% if you account for frequency) not 20%. The SPECfp comparison was correct.

How did you get 17%?

It's 5.33 for the 1065G7 and 4.33 for the 4700U. That's 20%. Not to mention 3.9GHz Turbo for 1065G7 and 4.1GHz Turbo for 4700U. That would bring perf/clock close to 25%.

and I don't think this is a particularily fair comparison given how far behind JEDEC spec this memory considering unlike LPDDR4X-4266 memory such horrible latency does not come with the additional benefit of improved bandwidth for the CPU.

The Int test isn't very bandwidth sensitive, especially 1T. That's why you see it being further ahead in SpecFP.

how much power does sunny cover use while beating it? on my 4700u i cant make a core us above 6 watts while doing AVX2 1t stress tests.

The thing about the Ryzen mobile CPUs(both 3000 and 4000), is that they are able to aggressively clock down even during relatively heavy load. Intel CPUs just run at a constant clock speed.
1593614395444.png
1593614407788.png
 
Last edited:

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,574
146
The Int test isn't very bandwidth sensitive, especially 1T. That's why you see it being further ahead in SpecFP.
To be honest, after thinking about it again my point about bandwidth doesn't make sense anyway because Renoir would run the memory at 2666mhz on purpose to maximise clocks and reduce latency in that test.

The main problem is that the memory would still be running at CL32... which is not ideal for memory latency. With how Renoir-U handles FCLK/UCLK/MCLK you won't ever be able to compare IPC accurately at JEDEC memory timings without using DDR4-2666mhz memory in the first place to ensure that the timings are according to JEDEC spec (but then you run into this weird gray area where while it's technically not running the memory at what the chip can support on paper but at the same time it's running the best memory the chip can support in practice) or to test RNR-H with DDR4-3200 to get a proper IPC value at JEDEC timings and with the maximum supported bandwidth.
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,573
136
How did you get 17%?

It's 5.33 for the 1065G7 and 4.33 for the 4700U. That's 20%. Not to mention 3.9GHz Turbo for 1065G7 and 4.1GHz Turbo for 4700U. That would bring perf/clock close to 25%.

113960.png


It's 4.44. but of course you're correct, my bad. I accidentally calculated it as 17% slower (4.44/5.33 = 0.8330)
5.33 / 4.44 * (4.1/3.9) = 1.262 which is 26% (when accounting for frequency as well)
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
can support in practice) or to test RNR-H with DDR4-3200 to get a proper IPC value at JEDEC timings and with the maximum supported bandwidth.

The H and the desktop versions will be several percent faster per clock due to that and other reasons. Same will be true for a hypothetical desktop version of Icelake.

That's why I don't like them doing "perf/clock" comparisons of laptops versus desktops.

Not sure if Intel will be able to net an easy power saving win by being able to finely adjust clocks as Picasso/Renoir does. This is probably why Intel devices get peak scores in the first few mins but downgrade while AMD laptops don't. I don't know if its because they believe it'll be better for user experience(eliminating any case of stuttering) or it can't do it.