Discussion Intel current and future Lakes & Rapids thread

Page 298 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Maybe LGA1700 is quad channel? It is both a marketing trick and what is needed to win the IGP wars on DDR5.

Quad Channel DDR5 would be intriguing but I would have to say no. The mainstream parts have been dual since forever.

And it wouldn't be for the IGP, if anything it looks like desktop will continue to be 32 EUs max.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
@inf64 Actually it shows 4% according to the Computerbase.de data.

The 10900K is underperforming. The 9900KS and the 9900K are both at 5GHz. The 9900KS does slightly better.

Also the 10900K gets 1.528 Geomean, while 9900K gets 1.542. If you divide by clocks, Zen 2's advantage is 7.13% against the 9900K, but 8.1% against 10900K. Why? Because Spec doesn't scale linearly with clocks like any real world application.

If you only look at SpecInt, then the advantage over the 9900K is only 4.88%. SpecInt is more representative of consumer workloads.

And that's just SpecInt_2017. In SpecInt_2006, the 9900K is actually 1.87% faster per clock than Zen 2.

Addendum: Zen 2 also shows advantage in memory bandwidth. So while latencies are higher, it'll benefit applications that need more bandwidth. Like, SpecFP, which it does better relative to Skylake.
 
Last edited:

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
@inf64 Actually it shows 4% according to the Computerbase.de data.

The 10900K is underperforming. The 9900KS and the 9900K are both at 5GHz. The 9900KS does slightly better.

Also the 10900K gets 1.528 Geomean, while 9900K gets 1.542. If you divide by clocks, Zen 2's advantage is 7.13% against the 9900K, but 8.1% against 10900K. Why? Because Spec doesn't scale linearly with clocks like any real world application.

If you only look at SpecInt, then the advantage over the 9900K is only 4.88%. SpecInt is more representative of consumer workloads.

And that's just SpecInt_2017. In SpecInt_2006, the 9900K is actually 1.87% faster per clock than Zen 2.


Did they test the CPUs with the same RAM speed or according to spec? If according to spec the difference could be even smaller when all have the same RAM speeds.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
@mikk AMD may be slower on single thread for memory, but its better on memory level parallelism.

It's with the same memory. Latency-wise Intel wins undoubtly. Not so for bandwidth.
 

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
@inf64
In SpecInt_2006, the 9900K is actually 1.87% faster per clock than Zen 2.

So restricted to SSE2 and only for integer related instructions...

You have another one like this..?.

Besides AT tested the intel set up with overclocked RAM in respect of official specs and with turbo set on.

FTR in Cinebench R15, that is up to SSE4.2, the delta is 9% in ST and 13% in MT.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
@Abwx I don't know where you pulled that from, but AT used official maximum RAM for the 9900K, that is DDR4-2666.

Zen 2 on that review is using DDR4-3200.

Also, Cinebench is a single application, versus Spec which has many subtests. Not to mention the Computerbase.de review.

Unless the reviewer sets the frequency to fixed amount, these "IPC" comparisons are all a joke. For further info, they should do another one with SMT disabled.
 
  • Like
Reactions: mikk

SAAA

Senior member
May 14, 2014
541
126
116
Putting aside the discussion on exact IPC numbers (single percents can change for ram, actual clocks or even different cache on the same uArch) I wouldn't count Rocket Lake out of the picture if it does held 5GHz with better than Skylake's IPC. A meager +10% in games would put it easily ahead of both Comet Lake and Zen 3, but we don't really know how Ice/Tiger changes affected that scenario.
I don't see Zen 3 pulling much higher than 4.5GHz all core clocks so that's already a 11% crawled back for Rocket, plus another 10% minimum for IPC.
They might be neck to neck in single thread, while the aging 14 nm will make consumption really bad.
Multicore there's no discussion that Zen 3 and 4 will keep their lead, even against Alder Lake with Golden Cove cores (and Meteor Lake too if that stays below 12 core).
 

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
So Intel has listed some Tigerlake based Celerons and Pentiums on ARK.

The most significant thing to me is the fact that they now support AVX256, no AVX512 however.

 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
For gaming, all-core frequency does not matter as much. Gaming workload, even a seemingly well threaded one will be less intense per core than a multi-threaded content creation workload.

Also, I don't know why we are assuming 10% as a defacto number for Cypress Cove. That's a substantial deficit over Sunny Cove, and big enough of a difference that it can no longer be considered as same core. It's possible it clocks less and the gain is 10%.
 
  • Like
Reactions: Tlh97 and Gideon

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
@Abwx I don't know where you pulled that from, but AT used official maximum RAM for the 9900K, that is DDR4-2666.

Zen 2 on that review is using DDR4-3200.


3200MHz is official spec for Zen 2, why should it be slowed down to match Intel s 2660Mhz..?.

IPC as stated, and guaranted, by the manufacturer include of course the official (and guaranted) RAM frequency since the efforts around the controler are part of the designs improvement, arent they.?
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
So Intel has listed some Tigerlake based Celerons and Pentiums on ARK.

The most significant thing to me is the fact that they now support AVX256, no AVX512 however.


It also supports DL Boost. They should have at least supported original AVX. That kind of fragmentation is bad.

3200MHz is official spec for Zen 2, why should it slowed down to match Intel s 2660Mhz..?.

Are you doing this on purpose? Cause its annoying. Who among us enthusiasts will buy a 10900K or a 3950X to pair it with official spec? That's crazy talk. It costs like $10 more to buy an XMP-enabled RAM and easily get DDR4-3200.

And we're talking about uarch comparisons. You are either deliberately being obtuse or were always like this. If you just read few posts above you, you'd have seen that. I myself at least skim through the previous page to get context. Just stop it. It's not like we're talking face to face and you just entered the conversation. 298 pages of previous history exist.
 

SAAA

Senior member
May 14, 2014
541
126
116
For gaming, all-core frequency does not matter as much. Gaming workload, even a seemingly well threaded one will be less intense per core than a multi-threaded content creation workload.

Also, I don't know why we are assuming 10% as a defacto number for Cypress Cove. That's a substantial deficit over Sunny Cove, and big enough of a difference that it can no longer be considered as same core. It's possible it clocks less and the gain is 10%.

Gaming does have more benefits from memory or other aspects indeed. As for the 10% the only hint I have is this:


That score is more or less 10% better than Comet Lake if it's running at 5 GHz, consider Tiger Lake at 4.8 GHz achieves a few results around 1580-1620.
 
  • Love
Reactions: spursindonesia

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
So Intel has listed some Tigerlake based Celerons and Pentiums on ARK.

The most significant thing to me is the fact that they now support AVX256, no AVX512 however.

It also has turbo boost enabled which mobile Pentium U usually doesn't. That's actually a pretty big improvement over the current Pentium Gold U.

It also supports DL Boost. They should have at least supported original AVX. That kind of fragmentation is bad.

I'm sure it does support AVX if it does support AVX2.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
That score is more or less 10% better than Comet Lake if it's running at 5 GHz, consider Tiger Lake at 4.8 GHz achieves a few results around 1580-1620.

Sorry but that doesn't tell me much. You can find Tigerlake results with 1200 score running at 4.6GHz too.

Userbenchmarks like Geekbench are notoriously hard to trust. That's why I wait for release, and just compare the highest scores.

And by the way, the Tigerlake 1165G7 just made another record:

ST: 1716

@jpiniero I know. What I mean is Celerons and Pentiums should have always had at least AVX support, not wait until 2020 to do it.
 
  • Like
Reactions: Grazick

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
@jpiniero I know. What I mean is Celerons and Pentiums should have always had at least AVX support, not wait until 2020 to do it.

Including AVX increases the power consumption. With products that typically have no turbo, the power consumption becomes an issue with yield. I guess they decided that since Gracemont is coming with AVX2, they need to enable it on the Pentiums and Celerons, and enabled Turbo Boost on the Pentium.
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Quad Channel DDR5 would be intriguing but I would have to say no. The mainstream parts have been dual since forever.

And it wouldn't be for the IGP, if anything it looks like desktop will continue to be 32 EUs max.

It is hard to explain that increment in pins whiout some large i/o change. Maybe tri-channel? i dont know.

So Intel has listed some Tigerlake based Celerons and Pentiums on ARK.

The most significant thing to me is the fact that they now support AVX256, no AVX512 however.


2/2 in 2020... 2/2 needs to stop existing ASAP.
 

SAAA

Senior member
May 14, 2014
541
126
116
Sorry but that doesn't tell me much. You can find Tigerlake results with 1200 score running at 4.6GHz too.

Userbenchmarks like Geekbench are notoriously hard to trust. That's why I wait for release, and just compare the highest scores.

And by the way, the Tigerlake 1165G7 just made another record:

ST: 1716

@jpiniero I know. What I mean is Celerons and Pentiums should have always had at least AVX support, not wait until 2020 to do it.

1716! It made the top 10… also that's under Linux, higher than the best windows one with a locked 4.7 GHz 1165G at 1620.

For some reason I don't think it will be hard to see 2000 on good desktops in 2 years, give or take the usual node delays…
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
It is hard to explain that increment in pins whiout some large i/o change. Maybe tri-channel? i dont know.

It just doesn't matter. Actually Prescott went from 478 pins to 775 pins, likely due to increased power delivery requirements.

1716! It made the top 10… also that's under Linux, higher than the best windows one with a locked 4.7 GHz 1165G at 1620.

For some reason I don't think it will be hard to see 2000 on good desktops in 2 years, give or take the usual node delays…

Nevermind 2 years. I expect 2000+ with Alderlake in 1165G7's TDP range. Although I don't expect desktops to do a whole lot better.

It's an unannounced goal for the lowest power chips to get nearly identical performance in single thread. Higher TDP is for cores and vector units.
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
@Abwx

Unless the reviewer sets the frequency to fixed amount, these "IPC" comparisons are all a joke. For further info, they should do another one with SMT disabled.


Agree on this. Any decent reviewer is using a fixed frequency for a reason if possible. Obviously this isn't always doable like for mobile variants but on desktop a fixed frequency is the way to go.
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
Why is everyone sleeping on this guy, when it comes to Matisse's ipc. He tests a broad range of codes and he does a more thorough job than any reviewer yet, in my humble opinion. Of course, he was ostracized on this forum because his results indicated virtual ipc parity, with Zen 2 ever so slightly ahead of skylake, and this is even with mitigations applied.

Edit: He had a fall out with certain individuals over Skylake retaining a slight edge over Zen(+) yet after it was all said and done, Skylake was indeed faster. I mean, if Zen 2 is 18% faster than Zen? and only ~3-5% faster than Skylake, then what was all that argument about? Yet, the vitriol he was subjected to from a section of the population here ultimately led to his departure.
 
Last edited:
  • Like
Reactions: yuri69 and mikk

inf64

Diamond Member
Mar 11, 2011
3,685
3,957
136
Well we do have real world benchmarks at computerbase and indexes they created represent real world performance. We'll see on the 5th where Zen3 lands but I have a pretty good idea how that will pan out.
 
  • Love
Reactions: spursindonesia

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
Why is everyone sleeping on this guy, when it comes to Matisse's ipc. He tests a broad range of codes and he does a more thorough job than any reviewer yet, in my humble opinion. Of course, he was ostracized on this forum because his results indicated virtual ipc parity, with Zen 2 ever so slightly ahead of skylake, and this is even with mitigations applied.

Edit: He had a fall out with certain individuals over Skylake retaining a slight edge over Zen(+) yet after it was all said and done, Skylake was indeed faster. I mean, if Zen 2 is 18% faster than Zen? and only ~3-5% faster than Skylake, then what was all that argument about? Yet, the vitriol he was subjected to from a section of the population here ultimately led to his departure.

I don't think I was posting at the time but I remember his power numbers being much higher than everybody else. If I had to guess it probably had more to do with that.
 

dsc106

Senior member
May 31, 2012
320
10
81
I'm wondering if y'all think Alder Lake-S will be worth waiting for vs Ryzen 5950x.

Pros: Something tells me Intel will take the performance crown away from AMD, not just today, but will best what AMD has next. True next gen platform with pci4.0 and DDR5. Start of a new era/socket, vs Ryzen 5000, end of a socket/era.

Cons: Will Intel *actually* get it out in H2 2021, and will any be available? Will DDR5 be super expensive comparatively if I want 128gb of RAM, inflating the cost substantially? Or will this be a longer waiting game (especially after any early supply issues), followed by much higher prices?

I am also curious if the new Atom small cores are going to be something really utilized in software right away, or how that will compare to a 5950x with 16 big cores - I mean, could they even be smaller?

Half of me says, just wait and get the biggest leap in performance in many years. Intel will pull through! The other half says, good luck seeing this in your system until Q1-Q2 2021, and expect to shell out at least an extra $500 for the platform thanks to DDR5, and expect perhaps a 15% performance increase over Ryzen 5950x. If that's the case, I'd rather wait until Alder Lake is more mature and DDR5 is cheaper.
 
  • Haha
Reactions: spursindonesia

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
Cons: Will Intel *actually* get it out in H2 2021, and will any be available? Will DDR5 be super expensive comparatively if I want 128gb of RAM, inflating the cost substantially? Or will this be a longer waiting game (especially after any early supply issues), followed by much higher prices?

Alder Lake supports both DDR4 and DDR5, in fact the first Alder Lake Sisoft entries did use DDR4.
 

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
I'm wondering if y'all think Alder Lake-S will be worth waiting for vs Ryzen 5950x.

Pros: Something tells me Intel will take the performance crown away from AMD, not just today, but will best what AMD has next. True next gen platform with pci4.0 and DDR5. Start of a new era/socket, vs Ryzen 5000, end of a socket/era.

Cons: Will Intel *actually* get it out in H2 2021, and will any be available? Will DDR5 be super expensive comparatively if I want 128gb of RAM, inflating the cost substantially? Or will this be a longer waiting game (especially after any early supply issues), followed by much higher prices?

I am also curious if the new Atom small cores are going to be something really utilized in software right away, or how that will compare to a 5950x with 16 big cores - I mean, could they even be smaller?

Half of me says, just wait and get the biggest leap in performance in many years. Intel will pull through! The other half says, good luck seeing this in your system until Q1-Q2 2021, and expect to shell out at least an extra $500 for the platform thanks to DDR5, and expect perhaps a 15% performance increase over Ryzen 5950x. If that's the case, I'd rather wait until Alder Lake is more mature and DDR5 is cheaper.

I don't see how 8 big cores and 8 little cores beats 16 big cores. Could happen, I guess. I wouldn't count on it.
 
  • Like
Reactions: lightmanek