Apollo Lake (Platform) / Goldmont (CPU core Codename) Info

[Roland00Address]

So do we have any information on Apollo Lake and specifically on the performance of Apollo Lake with CPU and/or GPU.

I ask for Intel officially announced Apollo Lake at the Intel Developer's Forum yet I have only see info that the OEMs care about with things like platform cost and nothing about performance. Apollo Lake according to Intel at IDF will be out sometime in the next 3 to 8 months during the 2nd Half of 2016.

Apollo Lake is going to be the next generation Atom platform with a new cpu architecture and Gen 9 graphics.

4 Years ago in 2011 when Intel claimed to soon become very serious about tablets they were claiming Apollo Lake was going to have 10x the peformance of the Clover Trail Platform which was at the time was the about to be released dual core atom that was effectively the same architecture but with a die shrink and better graphics that was in the 2008 original atom netbooks.

Apollo Lake was supposed to be comparable in some benchmarks as the Phenom II also as quoted by Intel.

-----



Yet I did not here any details about cpu or gpu performance at IDF even though Apollo Lake is delayed being about a year late compared to the original roadmaps for atom.

We do know that the new atom will have a gen 9 graphics aka the same generation of graphics in skylake but I once again have not seen anything about the number of calculation/execution units, thus is it a 12 EU, 18 EU, 24 EU or some other number, or the clock speed for these parts?

-----

Anyone with any info? Feel free to talk about other stuff relating to Apollo Lake but please do not turn this into an Intel the Corporation vs AMD the Corporation thread, for that would be an off topic conversation and does not belong here

 

[Abwx]

Anyone with any info? Feel free to talk about other stuff relating to Apollo Lake but please do not turn this into an Intel the Corporation vs AMD the Corporation thread, for that would be an off topic conversation and does not belong here

I would say that it s asking for one thing and its contrary since you, or rather Intel, is stating this :

Apollo Lake was supposed to be comparable in some benchmarks as the Phenom II also as quoted by Intel.

Wich is of course wishfull thoughts, in benches that matters it looks that it s still below AMD s Beema/Mullins wich is itself significantly below Phenom/AthlonX4.

 

[DrMrLordX]

Do we have any Goldmont benches yet?

 

[Nothingness]

Do we have any Goldmont benches yet?

This is discussed in the Bay Trail thread. At the moment the only public bench is a Geekbench result of an engineering sample.

 

[ShintaiDK]

The Atom line just need to stop. Tablets are dead, Core M is selected in 2in1 and such. Xeon D owns micro servers. And Smartphones started they decline and will suffer greatly the next years.

 

[sm625]

Atom x7-Z8700 has a passmark single thread score of 550. These new ones will probably score 600. There are pentium 4's that score higher. A modestly overclocked G3258 scores 2400.

 

[VirtualLarry]

Atom x7-Z8700 has a passmark single thread score of 550. These new ones will probably score 600. There are pentium 4's that score higher. A modestly overclocked G3258 scores 2400.

Yes, but the Atom in question, is what, a 4-5W TDP? Whereas the G3258 is what, a 58W TDP + overclocking factor? (Ok, realistically, when I had my G3258 OCed to 4.0Ghz @ 1.2 or 1.3V, CoreTemp reported Package Power at ~54W at full load.)

Still, that's 10X the TDP / Wattage, and the Atom has better than 1/10 the ST performance.

 

[ShintaiDK]

Production cost they could just as well make Core M parts. Sure the margins for R&D and such is another matter in that price segment. But its simply time to say bye to Atom and move Core M down where it can pay off.

 

[Nothingness]

Production cost they could just as well make Core M parts. Sure the margins for R&D and such is another matter in that price segment. But its simply time to say bye to Atom and move Core M down where it can pay off.

I definitely agree. It's time Intel let Atom die and move Core M down. But are they ready to go down to the <$200 or even $300 laptop segment?

 

[ShintaiDK]

I definitely agree. It's time Intel let Atom die and move Core M down. But are they ready to go down to the <$200 or even $300 laptop segment?

You see ASP keep increasing. People are moving up, not down.

The question is rather if the life of those 200-300$ laptops doesn't belong to the past. Just as their performance.

 

[VirtualLarry]

I definitely agree. It's time Intel let Atom die and move Core M down. But are they ready to go down to the <$200 or even $300 laptop segment?

If they do, wouldn't that just solidify Intel's performance win (with Core M in lower-end devices)? Along with Windows compatibility? I think with the current situation with Atom, that Intel is simply giving ARM "breathing room", and ARM is growing because of that.

I think that the ultimate, would be a future Core M-derived Phone SoC, with Intel / x86 app compatibility, with some sort of docking. Maybe an Ubuntu phone. That would be sweet!

 

[DrMrLordX]

This is discussed in the Bay Trail thread.

Yuck. That thread is a monster, and Goldmont is going to be neither Bay Trail or Cherry Trail so it deserves its own thread . . .

At the moment the only public bench is a Geekbench result of an engineering sample.

Okay . . . so only spotty information.

 

[Sweepr]

Yuck. That thread is a monster, and Goldmont is going to be neither Bay Trail or Cherry Trail so it deserves its own thread . . .

It is the Atom thread and will continue to be. The Geekbench and SiSoftware results for Broxton were posted there first. No problem with a separate thread for specific architecture or desktop/notebook Apollo Lake discussion though.

 

[ElFenix]

apollo lake is a terrible code name. it's like they're intentionally trying to confuse people. skylake, kaby lake, cannon lake, apollo lake

 

[Roland00Address]

I would say that it s asking for one thing and its contrary since you, or rather Intel, is stating this :



Wh ich is of course wishfull thoughts, in benches that matters it looks that it s still below AMD s Beema/Mullins wh ich is itself significantly below Phenom/AthlonX4.

One is talking about a product and its performance. The product is no longer being sold and it is the product Intel specifically compared it to in 2011. They did not compare it to nehalem or sandybridge they compared it to the Phenom II which was AMD's big product in 2011 (Phenom II was 2008 to 2012, Sandy was 2011, Nehalem was 2008).

I can't control what Intel does, but at the time of the announcement in 2011 they were announcing apollo lake which is a "small cpu core" was giving similar performance to their competitors big core.

All I am asking about is can we keep this thread about the product and its performance and stay away from talking about corporations and companies and so on. Aka it is kinda similar about staying away from talking about politics and political parties even though politics and political parties influence other things.

Wh ich is of course wishfull thoughts, in benches that matters it looks that it s still below AMD s Beema/Mullins wh ich is itself significantly below Phenom/AthlonX4.

As for your other point that I am addressing and thus the reason for this post...we all want to know does apollo lake bench. I seriously hope it does bench well and it is good in performance. Yet I have not seen any benches and I am now doubting the real world performance is not going to be the same as marketing slides and easy promises that were given 5 years ago.

Put another way this is evaluating did Intel keep its promise or did it not. We already know they are/were a year behind in launching apollo lake, lets now see if they are right or wrong on other promises such as performance, thermals, etc.

It is the Atom thread and will continue to be. The Geekbench and SiSoftware results for Broxton were posted there first. No problem with a separate thread for specific architecture or desktop/notebook Apollo Lake discussion though.

I made a separate thread for the baytrail thread while amazing is a monster with a 100+ pages. I honestly do not care who makes the thread I just wanted something that was easy to view and get information for Apollo Lake Platform / Goldmont CPU Cores is not the same as Bay Trail. In other words I am saying I do not care if my thread is the thread or someone else authors the thread. Now even though I do not care about the personal poster I do like it when the threads are organized, and are upkeeped by a person who feels personally invested in always posting great information.

Sweepr I just want to say thank you for your stewardship of the previous baytrail and cherrytrail thread. I know I personally can't be as good as steward as you could be, (especially since I no longer work in the computer sales industry), my little post here was merely meant to be a smaller post talking about apollo lake performance and whether we have seen any benchmarks or rumors and not a metapost. If you want to create a Apollo Lake meta thread I am all for it, for your previous stewardship speaks for yourself.

Let me say thank you for your past postings, aka me using words to express my gratitude to you, Sweepr.

 

[Roland00Address]

The Atom line just need to stop. Tablets are dead, Core M is selected in 2in1 and such. Xeon D owns micro servers. And Smartphones started they decline and will suffer greatly the next years.

Personally I see little future for Atom to be a "big thing" instead of a niche market as long Microsoft continues to screw up like they did with Windows 7 Phone and Windows 8 Phone. How did they screw that up, lack of apps and lack of updating the software seriously during those key years (now a days though I would argue windows is more up to date compared to Android due to framgentations and carriers not updating to something past android 4)

If Windows can't compete in the low end devices and the low end devices whether phones, tablets, etc becomes an Android only thing then it is likely we will not see Atom capture a large market share for Android can run effectively equally well on arm or x86, or dozens of other cpu architectures.

Compare this to windows, While windows with certain builds can run on x86 or ARM it is only on x86 that you can run legacy apps. The problem of this is the entry methods for these low end devices often suck with legacy apps for the legacy apps assume you have a keyboard or mouse and are not based around other styles of input such as touch, voice, and pen. Sure there are ways to get keyboard and mouse input into a tablet but it is not quite the same that these legacy software was designed to do.

 

[Abwx]

They did not compare it to nehalem or sandybridge they compared it to the Phenom II which was AMD's big product in 2011 (Phenom II was 2008 to 2012, Sandy was 2011, Nehalem was 2008).

I can't control what Intel does, but at the time of the announcement in 2011 they were announcing apollo lake which is a "small cpu core" was giving similar performance to their competitors big core.

They could had compared it to Nehalem and stated the same things, the "benches" they are talking about are the ones using specific instructions like AES, or SHA where Intel has been traditionaly weak, in wich case AMD s low power offering is also beating Nehalem in "some benches"...

Otherwise this chip is well below a Phenom for whatever doesnt use those instructions, for the record Mullins is 20% faster in Integer and 30% in FP than Baytrail, and still, it s significantly slower than the Phenom for both kind of apps, if Intel s statement was to be taken seriously then this Apollo lake would have roughly 35% better IPC than BT in Integer and something like 45-50% in FP.

 

[Shivansps]

I would say that it s asking for one thing and its contrary since you, or rather Intel, is stating this :



Wich is of course wishfull thoughts, in benches that matters it looks that it s still below AMD s Beema/Mullins wich is itself significantly below Phenom/AthlonX4.

Phenom 2 N930 is 644 st passmark that just 100 points away from z8700, so z8750 shuld be already very close.

 

[Roland00Address]

They could had compared it to Nehalem and stated the same things, the "benches" they are talking about are the ones using specific instructions like AES, or SHA where Intel has been traditionaly weak, in wich case AMD s low power offering is also beating Nehalem in "some benches"...

Key word could have

Second key word they, they meaning not me, but instead intel representatives and intel marketing.

I can't control what intel does. I was not there in 2011 and did not see the benches on the computer screen, all I can do is post marketing slides and hope the marketing slides are a good representative to the truth. I also must take the intel representative who did the comparison did a good job with his comparison and I can't choose what he compares it to such as phenom II or nehalem. Remember I am not the person who made the initial comparison and I am just channeling what he said and stated that those are not my words but his. Thus while I am not quoting him word to word I am still making it obvious since I was not there to see the benches that it was his eyes and his voice and I am repeating the spirit of what he is saying.

In other words Abwx I am only the messenger, shoot the other guy ( *wink* )

------

The point of this thread is for info like this

Otherwise this chip is well below a Phenom for whatever doesnt use those instructions, for the record Mullins is 20% faster in Integer and 30% in FP than Baytrail, and still, it s significantly slower than the Phenom for both kind of apps, if Intel s statement was to be taken seriously then this Apollo lake would have roughly 35% better IPC than BT in Integer and something like 45-50% in FP.

Hopefully we get real world benches, but if we can't get this information at least informed speculation trying to make sense of things.

 

[StrangerGuy]

I won't be remotely surprised if this isn't a direct response to OEMs who wanna ever cheaper chips to sell in their mass market craptops while branding it as Core.

 

[Abwx]

I can't control what intel does. I was not there in 2011 and did not see the benches on the computer screen, all I can do is post marketing slides and hope the marketing slides are a good representative to the truth. *wink* )

Well, marketing slides are just this, marketing dedicated, it s not by chance that they specified in "some benches", they know too well what are the real numbers, indeed even their Spec number is BS since they use Specint_rate2006 wich is indicative of bandwith but not of IPC as is Specint2006..

Phenom 2 N930 is 644 st passmark that just 100 points away from z8700, so z8750 shuld be already very close.

At 2GHz and in CB 11.5 :

Baytrail : 1.53

Mullins : 2.0

Phenom : 2.27

BT would need to be clocked at 3GHz to get the same score as the Phenom, so if Appolo lake is at 2.3 as displayed in the Geekbench link FP IPC would had to be improved by almost 30% over BT, and this for legacy code...

For Integer there s few comparison but the 3DMARK CPU score is 2800 for the Phenom 930 and 1914 for a Pentium 3700 wich is clocked over 2GHz in MT tests, so even more improvement is required in Integer over BT..

 

[IntelUser2000]

One is talking about a product and its performance. The product is no longer being sold and it is the product Intel specifically compared it to in 2011.

That was 4 years before the original Goldmont release schedule, 5 years after the delay.

It could have been that at the time the architecture did have enough potential to reach that sort of performance, but other reasons forced them to change architecture. Goldmont based mobile were being claimed as for "Hero" devices and all that nonsense. Maybe that claim was true in the beginning.

I can list a couple: 14nm sucking, too much performance means Core chips being cannibalized.

 

[Shivansps]

At 2GHz and in CB 11.5 :

Baytrail : 1.53

Mullins : 2.0

Phenom : 2.27

BT would need to be clocked at 3GHz to get the same score as the Phenom, so if Appolo lake is at 2.3 as displayed in the Geekbench link FP IPC would had to be improved by almost 30% over BT, and this for legacy code...

For Integer there s few comparison but the 3DMARK CPU score is 2800 for the Phenom 930 and 1914 for a Pentium 3700 wich is clocked over 2GHz in MT tests, so even more improvement is required in Integer over BT..

Yes 2.3Ghz as the top sku whould be a dissaster, i think we all know that, the current top CT runs at 2.56Ghz and the top Braswell is 2.64Ghz.

Still, remember that only the top Mullin, the A10 Micro 6700T, is really faster, and i dont know of any device using it, the A4 Micro-6400T is more or less the same than a 2.4Ghz CT, and now after the refresh the X5-8550 is at that level, i whould expect its replacement to be faster.

Anyway Mullin is old too.

 

[IntelUser2000]

Specint_rate2006 wich is indicative of bandwith but not of IPC as is Specint2006..

Also they are using SpecInt2000_rate not 2006.

Apollo Lake was supposed to be comparable in some benchmarks as the Phenom II also as quoted by Intel.

They claimed a lot of things. Actually, all marketing slides claim a lot of things. They were saying Ultrabook graphics performance in 2013 to be 7x 2011. The 7x performance needs 45W Iris Pro 580, and its not even released yet in 2016!

The only purpose of marketing slides is to calm investors who'd storm the company if they knew what was really going on.

 

[Abwx]

Yes 2.3Ghz as the top sku whould be a dissaster, i think we all know that, the current top CT runs at 2.56Ghz and the top Braswell is 2.64Ghz.

Still, remember that only the top Mullin, the A10 Micro 6700T, is really faster, and i dont know of any device using it, the A4 Micro-6400T is more or less the same than a 2.4Ghz CT, and now after the refresh the X5-8550 is at that level, i whould expect its replacement to be faster.

Anyway Mullin is old too.

Yes but their perf/Mhz is essentially the same as BT while their TDP is not much of an improvement, besides Hardware.fr measured the Pentium N3700 at roughly 10W+ in Luxmark CPU + GPU, a long shot from the claimed 6W..

Actually Mullins is still competitive with Braswell, about as good in CPU perf/Watt and better in GPU perfs + perf/watt, not sure that Intel s latest will bring big improvements given that it s the same node and that they are power constrained if they want to get lower than said underestimated TDP.

Also they are using SpecInt2000_rate not 2006.

They claimed a lot of things. Actually, all marketing slides claim a lot of things. They were saying Ultrabook graphics performance in 2013 to be 7x 2011. The 7x performance needs 45W Iris Pro 580, and its not even released yet in 2016!

The only purpose of marketing slides is to calm investors who'd storm the company if they knew what was really going on.

As i pointed to Shivansps the focus is power since Braswell didnt deliver accordingly, for absolute perfs one will have to wait the next iterations as there will be some because whatever the bad reputation of small CPUs by here they are out of reach when it comes to perf/Watt, i mean they have much higher IPC than what the transistors ratios would suggest when compared to huge cores.