Discussion Intel current and future Lakes & Rapids thread

Page 10 - 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
Sapphire Rapid is server only, which wouldn't make sense because they'd do all platforms for a completely new design.

Server first doesn't mean Server only.

Icelake is not a real CPU update? Are you saying it is another Skylake refresh update like Kabylake?

I think the focus for Ice/Tiger is EMIB and power draw at the lower end. There'll be changes of course but IPC improvements are going to take a back seat. Maybe they'll throw in a token 5% IPC improvement but Icelake should clock worse than Coffee at the top end. Mobile should get a nice base clock speed boost.
 

TheF34RChannel

Senior member
May 18, 2017
786
309
136
It was my understanding that Sapphire Rapid will not come to consumers, and that the 'revolutionary new core design' that Intel is now hiring for comes before its release.
 

CatMerc

Golden Member
Jul 16, 2016
1,114
1,149
136
So what's the deal with Sapphire Rapid? The title of this thread is the first time I'm hearing about this. How do we know about it, and what information is known?
 

mikk

Diamond Member
May 15, 2012
4,112
2,106
136
I think the focus for Ice/Tiger is EMIB and power draw at the lower end. There'll be changes of course but IPC improvements are going to take a back seat. Maybe they'll throw in a token 5% IPC improvement but Icelake should clock worse than Coffee at the top end. Mobile should get a nice base clock speed boost.

This is based on what? We don't have any infos about Icelake or Sapphire Rapid.


It was my understanding that Sapphire Rapid will not come to consumers, and that the 'revolutionary new core design' that Intel is now hiring for comes before its release.

Who told this?
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
It was my understanding that Sapphire Rapid will not come to consumers, and that the 'revolutionary new core design' that Intel is now hiring for comes before its release.

What would Intel do if they don't use SR for client? Keep milking the Lakes? PC sales are down another 4% this quarter; and it's only going to get worse. Seems like it would be easier to build a server core and scale that down for client than to keep milking the Lakes for eternity.

I suppose it's possible that the new core design could be used as part of a 10 nm Phi as a trial run of sorts.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
My prediction on the clocks for the Coffee i7 K:

Base: 3.6 (could be 3.5 because of marketing reasons/the 7800X)
FCT: 4
4CT: 4.4
SCT: 4.6
 

Eddward

Member
Apr 10, 2012
56
19
81
When is the next real cpu update supposed to be? It looks like 14nm through 14++ is the same cpu core arch. What's going on? Even 10nm doesn't look to be much when it comes to core arch changes.
It's pretty easy. Yes 14nm/+/++ cpu uarchs are the same. After 14nm Skylake was always planned its 10nm die shrink Cannon Lake, so everything between is a big improvisation and we are still in this phase since 10nm is not out yet. Improvisation, because you can not make any significant arch changes from one day to another especially when you make a big mistake in terms of underestimating development of 10 nm process. Put simply Intel had nothing more in inventory for 14nm, everything after Skylake was designed only for 10nm.
Ice Lake is supposed to be first significant architecture upgrade since Skylake.
 

CakeMonster

Golden Member
Nov 22, 2012
1,384
482
136
Intel had yearly architecture updates until Skylake. Not always huge, but incremental, and over time significant. And then suddenly stopped, like they had hit a brick wall. I'm not buying that this was planned. Something must have happened rather quickly that made them stop manufacturing silicon with new design. Saving money on not designing, testing, verifying the new architecture seems obvious but I'm sure they are working in architecture upgrades still, but who knows when they'll release it and when they changed their strategy and what the new strategy is.
 

ehume

Golden Member
Nov 6, 2009
1,511
73
91
I found a link for Sapphire Rapids technology here. Seen to be a "rapid recovery tool" for lost ID's when Intel uses the term.

Q - when is Coffee Lake due for release? And does anyone know what wattage to expect?
 

TheF34RChannel

Senior member
May 18, 2017
786
309
136
I found a link for Sapphire Rapids technology here. Seen to be a "rapid recovery tool" for lost ID's when Intel uses the term.

Q - when is Coffee Lake due for release? And does anyone know what wattage to expect?

CFL-S: as per listed in this thread come October for the highest wattage parts (95W).

Sapphire Rapid (7nm?): 2020. I heard it was server only in this thread, however, I hope it'll be readied for consumers as well, because what else will we get after TGL?

9gxhy79y.png
 
Last edited:

mikk

Diamond Member
May 15, 2012
4,112
2,106
136
CFL-S TDP is 65W and 95W, Xeon will also come with 80W. Clock speed is very good I'm hearing, at least the Turbo frequency. I expect some leaks from that very soon.
 

TheF34RChannel

Senior member
May 18, 2017
786
309
136
CFL-S TDP is 65W and 95W, Xeon will also come with 80W. Clock speed is very good I'm hearing, at least the Turbo frequency. I expect some leaks from that very soon.

Adding to that, for @ehume, the 65W parts will release early next year. So basically we'll see 95W (4C? and) 6C with and without hyperthreading K parts first come October.
 

mikk

Diamond Member
May 15, 2012
4,112
2,106
136
According to my infos 65W will also come this year, but don't expect much SKUs.
 

TheF34RChannel

Senior member
May 18, 2017
786
309
136
According to my infos 65W will also come this year, but don't expect much SKUs.

Now that you mention it, I do recall something like that on that announcement slide. It's going to be confusing as hell to the average user.

Do you have sources outside of this thread?
 

ehume

Golden Member
Nov 6, 2009
1,511
73
91
I was hoping for something earlier and hotter. Now I have to wait for months and I have to OC to heat up the CPU enough to stress heatsinks. Oh well, can't be cooler then Devil's Canyon.

Thanks all.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
I suppose it's possible that the new core design could be used as part of a 10 nm Phi as a trial run of sorts.

10nm Xeon Phi, also known as Knights Hill, uses the Atom Goldmont core as its base.

Intel heavily underestimated Nvidia. Their much touted Nervana accelerators were supposed to beat GPUs in DL by 3-5x. This year's Volta GPU not only likely beats 2019's Knights Hill in DP FP performance, the tensor accelerators inside Volta beats yet to be released Nervana accelerators by 2x. So with a single powerful solution, it does better than two seperate ones and is earlier to market.

Then again, if 10nm wasn't delayed we should have been having Knights Hill this year.

Intel had yearly architecture updates until Skylake. Not always huge, but incremental, and over time significant. And then suddenly stopped, like they had hit a brick wall.

Their execution faltered because of shoddy and inconsistent management, and process got delayed. So they had to change plans. Actually end of Moore's Law scaling is the real brick wall. Unfortunately for Intel it seems they have chosen a real bad time to screw up on execution.

It's much needed though, a revolutionary new design that brings forward the innovation we will need for the foreseeable future.

Revolutionary designs seem only revolutionary because the real hard work is done behind closed doors though. Also, tech advancements don't last forever. Computer chips until recently has been the only area in technology where the improvements were breathtaking. All other areas of tech has reached its diminishing returns. The sole reason for computer chips having exponential gains were single handedly due to Moore's Law scaling. Now, its practically at its end.

How's this for a sobering thought? Imagine if diminishing returns regarding new CPUs were at the rate of battery tech advancement. :)
 
Last edited:

TheF34RChannel

Senior member
May 18, 2017
786
309
136
10nm Xeon Phi, also known as Knights Hill, uses the Atom Goldmont core as its base.

Intel heavily underestimated Nvidia. Their much touted Nervana accelerators were supposed to beat GPUs in DL by 3-5x. This year's Volta GPU not only likely beats 2019's Knights Hill in DP FP performance, the tensor accelerators inside Volta beats yet to be released Nervana accelerators by 2x. So with a single powerful solution, it does better than two seperate ones and is earlier to market.

Then again, if 10nm wasn't delayed we should have been having Knights Hill this year.



Their execution faltered because of shoddy and inconsistent management, and process got delayed. So they had to change plans. Actually end of Moore's Law scaling is the real brick wall. Unfortunately for Intel it seems they have chosen a real bad time to screw up on execution.



Revolutionary designs seem only revolutionary because the real hard work is done behind closed doors though. Also, tech advancements don't last forever. Computer chips until recently has been the only area in technology where the improvements were breathtaking. All other areas of tech has reached its diminishing returns. The sole reason for computer chips having exponential gains were single handedly due to Moore's Law scaling. Now, its practically at its end.

How's this for a sobering thought? Imagine if diminishing returns regarding new CPUs were at the rate of battery tech advancement. :)

Even so, something needs to happen in CPU advancement because software will continue to advance, as will other types of hardware and if the CPU cannot keep up we're in a world of hurt.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Even so, something needs to happen in CPU advancement because software will continue to advance, as will other types of hardware and if the CPU cannot keep up we're in a world of hurt.

Ok you'll need to point out which hardware is advancing faster. GPUs? That has slowed to 20-30% per generation. And that's only because they got the bonus of going from 28nm to 16nm. It's like with FinFET/Trigate. They say "its da bomb" or whatever and they claim 30% faster in corner case scenarios but 10 years ago just a shrink not even adding new materials we got 50-60% gains for all chips.

Storage? Well I found out SSDs are situationally useful. While it helps for consistency, my boot and loading times are still quite significant. NVM RAM and storage combined with significant rearchitecting from the near ground-up will bring the benefits we were looking for with SSDs in the first place.

Useless IOPS and MB/s in marketing bullet points mean we get this: http://techreport.com/review/31901/corsair-force-series-mp500-240gb-nvme-ssd-reviewed/5

The gains we get with SSDs are even worse than the percentage gains we have with CPUs.

No, I don't think we'll be truly hurt, there will be people that believe we will though. I think many will be extremely surprised and disappointed that computer gains are just as bad as in all other areas of tech.
 

tential

Diamond Member
May 13, 2008
7,355
642
121
CFL-S: as per listed in this thread come October for the highest wattage parts (95W).

Sapphire Rapid (7nm?): 2020. I heard it was server only in this thread, however, I hope it'll be readied for consumers as well, because what else will we get after TGL?

9gxhy79y.png
So this is delayed from the August release? I really want CFL but may go Ryzen to avoid getting a quad-core in 2017.
 
Mar 10, 2006
11,715
2,012
126
I think the focus for Ice/Tiger is EMIB and power draw at the lower end. There'll be changes of course but IPC improvements are going to take a back seat. Maybe they'll throw in a token 5% IPC improvement but Icelake should clock worse than Coffee at the top end. Mobile should get a nice base clock speed boost.

"Token 5% ipc" -- are you serious? Not a chance, if you are comparing SKL/KBL/CFL to ICL.
 
  • Like
Reactions: Sweepr
Mar 10, 2006
11,715
2,012
126
Intel had yearly architecture updates until Skylake. Not always huge, but incremental, and over time significant. And then suddenly stopped, like they had hit a brick wall. I'm not buying that this was planned. Something must have happened rather quickly that made them stop manufacturing silicon with new design. Saving money on not designing, testing, verifying the new architecture seems obvious but I'm sure they are working in architecture upgrades still, but who knows when they'll release it and when they changed their strategy and what the new strategy is.

What happened is actually very simple. Intel was planning to continue to do annual architecture updates (Cannon Lake was to follow Skylake, Ice Lake to follow Cannon Lake), but it became clear -- perhaps too late -- that 10nm was not in great shape. Intel's customers wanted new products for annual refreshes, so Intel decided to continue to enhance the performance of its 14nm node for another couple of generations while doing low risk products using those enhanced nodes (e.g. Kaby Lake, Coffee Lake).

Such a solution was not ideal, but it was better than the alternative (selling Skylake for another two years).
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
"Token 5% ipc" -- are you serious? Not a chance, if you are comparing SKL/KBL/CFL to ICL.

Do you think I am being too optimistic? The Geekbench leak suggested they are going to increase the L1 data cache, so they will likely do something. But as I said I think the focus is lower power and EMIB.
 
Mar 10, 2006
11,715
2,012
126
Do you think I am being too optimistic? The Geekbench leak suggested they are going to increase the L1 data cache, so they will likely do something. But as I said I think the focus is lower power and EMIB.

I doubt EMIB will debut with Ice Lake. It'll arrive with the 7nm gen, IMHO.