Discussion Intel current and future Lakes & Rapids thread

[jpiniero]

Probably higher because all these samples are based on early 0000 samples.

Might have to be a little conservative on base clocks for quality yield reasons, even if functional yield is close to 100%. I imagine the 8C Turbo will be pretty high.

Remember that there isn't going to be any i5 models yielded from the 8C die unless Intel changes their mind.

 

[Bouowmx]

Well, I've never even entertained the idea that there would be an iGPU on this chip.

8+2 (internal designation) is precisely the unique selling proposition: 8 cores + integrated GPU.

Xeon with disabled integrated GPU: behavior follows directly from past models (E3-1200 series).

 

[piesquared]

For those of us who do not game the "anemic" Intel iGPU is plenty good enough for looking at pictures and videos.

Oh I'm sure there will be a small niche that will only require 8 CPU cores and a display out, but most interested in an 8 core CPU are going to be adding a discrete card. For those people that anemic igp is just wasted die space.

 

[piesquared]

8+2 (internal designation) is precisely the unique selling proposition: 8 cores + integrated GPU.

Xeon with disabled integrated GPU: behavior follows directly from past models (E3-1200 series).

It might be unique but i doubt it will be a motivating factor for anyone to buy one.

"im not looking for an 8 core CPU"
"but it has an integrated intel igp"
"sold!"

I dont see that happening very often.

 

[jpiniero]

Oh I'm sure there will be a small niche that will only require 8 CPU cores and a display out, but most interested in an 8 core CPU are going to be adding a discrete card. For those people that anemic igp is just wasted die space.

OEMs do want the IGP though. Plus it'd be helpful for power saving if Intel does do a 8C mobile part.

 

[Dave2150]

Well, I've never even entertained the idea that there would be an iGPU on this chip. I don't imagine anyone is expecting that. I'm not sure why intel would consider side saddling their anemic gpu on this, it seems utterly pointless.
The only way putting a GPU next to this thing makes any sense is if they use AMD's GPU again, at least it would be useful.

But, i don't think there is any chance of this containing a GPU on chip or on package. I'm not entirely convinced this mythical CPU even exists, with or without GPU.

-They would essentially be building a single CPU with really only one usable SKU. How would they position the harvested chips with failed cores and how can they can compete on price with AMD's low cost and high yielding modular 8 core chips.

-How long ago would have they had to add this to the roadmap in order to design and release it this year(?). Probably at least 3 years, at the time when they would have been designing 10nm chips. I suppose there is a glimmer of a possibility that they ported everything back to 14nm when they realized 10nm is a disaster. What would that do to their business model though? Moving it back to 14nm would balloon the die size, TDP and/or wreak havok on clocks, and likely hurt yields, compared to their 10nm plans.

-Another glimmer of a possibility is if this has an intel 'GPU' on board and was ported back to 14nm from the originally intended 10nm when they realized 10nm was a disaster, and first caught wind of what Ryzen was going to be. But that only makes sense if they replace the entire CL stack with harvested whateverLake dies. This would surely drop their margins dramatically due to the Ryzen factor.

Having an iGPU is extremely handy/useful, as it allows testing/troubleshooting if there is a problem with your GPU.

I wish there was a small iGPU on Ryzen's mainstream for example

 

[Beemster]

P ~ 1/2 f CV^2 for each core

.........just look at the core size comparison of (NEW design) Ryzen and (OLD design) Skylake cores..........forget about anything else.....would it not be obvious that for very similar IPC and at the same operating voltage ......that AMD would have an advantage from a yield point of view even if one had to stay with a monolithic multi core die?........ and Intel would have a power/core disadvantage based on C alone .....which will blow up total power with high core count server designs at high f where all cores get pretty highly loaded?..........remember....Intel uses core frequency to define TDP.......and it has come down sharply as cores have been added............ they have some in house design application the loads all the cores up to a worse case server like load and then they see what freq they can maintain at the desired TDP.....................so where does AMD attack FIRST?.............high core count high freq design server chips...........where does Intel need to defend the most?.........high core count high freq design server chips...........BUT THEY CAN NOT until they get to EMIB and beyond 14nm

https://www.youtube.com/watch?v=XAx9G5PqBzM

 

[dacostafilipe]

...........

Don't take this wrong, but it's really difficult for me to read your posts with the lack of new lines and all those dots. I force myself to read the content, because it's good, but it's not easy.

 

[Eug]

Don't take this wrong, but it's really difficult for me to read your posts with the lack of new lines and all those dots. I force myself to read the content, because it's good, but it's not easy.

I just skip those posts.

 

[dullard]

Oh I'm sure there will be a small niche that will only require 8 CPU cores and a display out, but most interested in an 8 core CPU are going to be adding a discrete card. For those people that anemic igp is just wasted die space.

Do you have any numbers backing that up, or are you projecting your own wishes to the rest of the population?

I can imagine MANY uses for a powerful CPU with an IGP. Business professionals for example. Accountants; engineers; programmers; scientists; multimedia editing; anyone who does number crunching in simulations, models, etc. A large number of use cases don't need any GPU power at all, but no one wants their $100K+/year employees sitting on their hands waiting for their computer to finish its task. Sure some of those uses may also need a GPU (if the multimedia editing goes into video encoding for example), but most do not need any GPU when the IGP is plenty powerful enough.

 

[Eug]

Do you have any numbers backing that up, or are you projecting your own wishes to the rest of the population?

I can imagine MANY uses for a powerful CPU with an IGP. Business professionals for example. Accountants, engineers, programmers, scientists, multimedia editing, etc. A large number of use cases don't need any GPU power at all, but no one wants their $100K+/year employees sitting on their hands waiting for their computer to finish its task.

Probably musicians too. (Are you including that in multimedia editing?) Multicore, with fast single-core and dual-core Turbo speed and lots of RAM. GPU basically irrelevant as long as it can display everything on-screen properly (2D) on multiple monitors.

 

[LTC8K6]

Don't take this wrong, but it's really difficult for me to read your posts with the lack of new lines and all those dots. I force myself to read the content, because it's good, but it's not easy.

Someone already told him that, but the moderator misinterpreted the post as an attack on the poster, rather than an attack on the dots.

https://forums.anandtech.com/thread...re-rapid-thread.2509080/page-63#post-39432295

 

[dullard]

Probably musicians too. (Are you including that in multimedia editing?) Multicore, with fast single-core and dual-core Turbo speed and lots of RAM. GPU basically irrelevant as long as it can display everything on-screen properly (2D) on multiple monitors.

Good example. Musicians are about as far from my line of work as possible, so I didn't think about them. I hate the word "multimedia", but I used it to be very general as editing of many different types all require intensive CPU power.

 

[scannall]

I wish there was a small iGPU on Ryzen's mainstream for example

Even a very basic one in the chipset.

 

[Beemster]

Don't take this wrong, but it's really difficult for me to read your posts with the lack of new lines and all those dots. I force myself to read the content, because it's good, but it's not easy.

Sorry. You are right. I've been told as much by others before and elsewhere. I will make an effort to comply. It's just that I guess I think in a conversational tone and I post my ideas that way. Is this better? No dots!

 

[Beemster]

Someone already told him that, but the moderator misinterpreted the post as an attack on the poster, rather than an attack on the dots.

https://forums.anandtech.com/thread...re-rapid-thread.2509080/page-63#post-39432295

Really? For the record, I took it as a personal attack as well.

 

[Beemster]

Don't take this wrong, but it's really difficult for me to read your posts with the lack of new lines and all those dots. I force myself to read the content, because it's good, but it's not easy.

Thanks. FWIW, here's some more stuff to chew on. Remember, Intel uses base frequency to define their TDP. Let's assume they stay consistent with what ever program they use to tax all the cores and define core frequency at a given TDP. They had to drop the core frequency from 4.2GHz to 3.7GHz in going from 4 to 6 cores. Now there is at least some indication it might go down to 2.7 GHz at 8 cores. AMD supports a 3.7GHz core frequency at 8 cores. Presumably, they define their core frequency in some such similar matter as Intel. I am thinking 3.0 GHz myself for Intel's core frequency at 8 core. So, for all other things equal, this might indicate about a 20%-25% AMD advantage in power/core for their new Ryzen core vs the old Core core. FWIW, if those core sizes (11mm^2 for Ryzen and 17mm^2 for Skylake are correct), Skylake core is 54% larger in AREA. Does anyone know if Intel requires unidirectional gates? If yes, does anyone know which direction in relation to the Skylake core design? Is it along the long axis or short axis of the core design. It just seems likely that a core which is 54% larger in AREA would have the C term in the Power ~ 1/2 f CV^2 relation that could be 23% larger if all other things are equal. The square root of 1.54 is 1.23. This large drop in core frequency in going from 4 to 6 to 8 core might be attributed to C increase alone and just illustrates Intel also(among other things) needs a NEW compact core design relative to AMD. So, in total, Intel needs a new more compact core, needs to get past 14nm to a smaller node, and needs EMIB to yield a very high core count server offering to compete with AMD at 7nm. AMD already has a compact core. Also, the new ZEN 2 design might come with IPC improvements to the NEW Ryzen core. Intel can not squeeze out any more IPC gains from the OLD Core core. AMD will be at 7nm early next year. AMD already has Infinity Fabric. AMD had Jim Keller when they desperately needed him. AMD developed Infinity Fabric when they desperately needed it. AMD finally gets to an advance node when they desperately need to. How about that there AMD!..............and finally, AMD has Lisa Su. Who Intel got???

Only one set of dots. Better?

 

[jpiniero]

I think Skylake Server is 17, Skylake Client is like 8ish.

 

[Eug]

Only one set of dots. Better?

Heh.

I don't want to sound too negative, but I think people have issue with your lack of formatting. It's easier to read well-organized thoughts in paragraphs and lines as opposed to a single long stream of consciousness.

 

[Arachnotronic]

Oh I'm sure there will be a small niche that will only require 8 CPU cores and a display out, but most interested in an 8 core CPU are going to be adding a discrete card. For those people that anemic igp is just wasted die space.

The iGPU will be useful in business PCs, which make up a large chunk of the desktop market. Don't underestimate the value of the iGPU for people who will benefit from more CPU grunt and only need minimal GPU/media capabilities.

 

[LTC8K6]

Really? For the record, I took it as a personal attack as well.

He only said your dots are annoying as hell, but he typed ...... instead of the word "dots".

 

[LTC8K6]

Thanks. FWIW, here's some more stuff to chew on. Remember, Intel uses base frequency to define their TDP. Let's assume they stay consistent with what ever program they use to tax all the cores and define core frequency at a given TDP. They had to drop the core frequency from 4.2GHz to 3.7GHz in going from 4 to 6 cores. Now there is at least some indication it might go down to 2.7 GHz at 8 cores. AMD supports a 3.7GHz core frequency at 8 cores. Presumably, they define their core frequency in some such similar matter as Intel. I am thinking 3.0 GHz myself for Intel's core frequency at 8 core. So, for all other things equal, this might indicate about a 20%-25% AMD advantage in power/core for their new Ryzen core vs the old Core core. FWIW, if those core sizes (11mm^2 for Ryzen and 17mm^2 for Skylake are correct), Skylake core is 54% larger in AREA. Does anyone know if Intel requires unidirectional gates? If yes, does anyone know which direction in relation to the Skylake core design? Is it along the long axis or short axis of the core design. It just seems likely that a core which is 54% larger in AREA would have the C term in the Power ~ 1/2 f CV^2 relation that could be 23% larger if all other things are equal. The square root of 1.54 is 1.23. This large drop in core frequency in going from 4 to 6 to 8 core might be attributed to C increase alone and just illustrates Intel also(among other things) needs a NEW compact core design relative to AMD. So, in total, Intel needs a new more compact core, needs to get past 14nm to a smaller node, and needs EMIB to yield a very high core count server offering to compete with AMD at 7nm. AMD already has a compact core. Also, the new ZEN 2 design might come with IPC improvements to the NEW Ryzen core. Intel can not squeeze out any more IPC gains from the OLD Core core. AMD will be at 7nm early next year. AMD already has Infinity Fabric. AMD had Jim Keller when they desperately needed him. AMD developed Infinity Fabric when they desperately needed it. AMD finally gets to an advance node when they desperately need to. How about that there AMD!..............and finally, AMD has Lisa Su. Who Intel got???

Only one set of dots. Better?

So how come chips with the same # of cores, and the same base frequency, and the same architecture, can have different TDP ratings?
If you look around at Intel chips, you'll find TDP inconsistencies.

 

[wahdangun]

This is simply not true.

It is, now what's constitutes a stepping ?
It's basically same architecture, with some tweak, that allow higher clock.

 

[dullard]

Thanks. FWIW, here's some more stuff to chew on. Remember, Intel uses base frequency to define their TDP. Let's assume they stay consistent with what ever program they use to tax all the cores and define core frequency at a given TDP. They had to drop the core frequency from 4.2GHz to 3.7GHz in going from 4 to 6 cores.

Now there is at least some indication it might go down to 2.7 GHz at 8 cores. AMD supports a 3.7GHz core frequency at 8 cores. Presumably, they define their core frequency in some such similar matter as Intel. I am thinking 3.0 GHz myself for Intel's core frequency at 8 core.

So, for all other things equal, this might indicate about a 20%-25% AMD advantage in power/core for their new Ryzen core vs the old Core core. FWIW, if those core sizes (11mm^2 for Ryzen and 17mm^2 for Skylake are correct), Skylake core is 54% larger in AREA. Does anyone know if Intel requires unidirectional gates? If yes, does anyone know which direction in relation to the Skylake core design? Is it along the long axis or short axis of the core design. It just seems likely that a core which is 54% larger in AREA would have the C term in the Power ~ 1/2 f CV^2 relation that could be 23% larger if all other things are equal. The square root of 1.54 is 1.23. This large drop in core frequency in going from 4 to 6 to 8 core might be attributed to C increase alone and just illustrates Intel also(among other things) needs a NEW compact core design relative to AMD.

So, in total, Intel needs a new more compact core, needs to get past 14nm to a smaller node, and needs EMIB to yield a very high core count server offering to compete with AMD at 7nm. AMD already has a compact core. Also, the new ZEN 2 design might come with IPC improvements to the NEW Ryzen core. Intel can not squeeze out any more IPC gains from the OLD Core core. AMD will be at 7nm early next year. AMD already has Infinity Fabric. AMD had Jim Keller when they desperately needed him. AMD developed Infinity Fabric when they desperately needed it. AMD finally gets to an advance node when they desperately need to. How about that there AMD!..............and finally, AMD has Lisa Su. Who Intel got???

Only one set of dots. Better?

1) Intel has a consistent way to test TDP. The processor is heated to its maximum case temperature, then a CPU intensive (but not viral-load) software is ran, and the power consumed is measured.

2) With two pre-release engineering sample 8-core chips, why do you focus only on the lower frequency one and not the 3.1 GHz chip?

3) Engineering sample chips are not the same as the chips that will be sold. The engineering sample of the Kaby Lake 7700K was at 3.7 GHz (http://ranker.sisoftware.net/show_r...d5e3d0e2dbe2d7f183be8ea8cda895a583f0cdf5&l=en) but the actual 7700K was at 4.2 GHz. The engineering sample of the Coffee Lake 8700K was at 3.5 GHz (http://ranker.sisoftware.net/show_r...d5e3dbecdfe6d4f280bd8dabceab96a680f3cef6&l=en) but the actual 8700K was at 3.7 GHz.

4) The 6-core Coffee lake is ~151 mm^2 and the 4-core Kaby lake is ~126 mm^2. Thus two cores are 25 mm^2 at most (there were other added features in Coffee Lake that may take up some of that area). One core is then 12.5 mm^2 at most. Not 17 mm^2. You are thinking server chips, not consumer chips.

5) The die area from 6-core Coffee Lake to the unannounced 8-core chip would go up about 15.6%. Thus, if the die design was uniform across the chip (it isn't exactly uniform but it is the best estimate that I have), the we could expect capacitance to go up about 15.6%. Assuming all things equal, using the P = Pother + 1/2*C/V^2 formula, the speed would drop to 3.2 GHz. But that ignores any benefits from 14nm++ that didn't make it into the Coffee Lake chip and it ignores all other Power consumption such as leakage current.

6) You asked if we liked it better. Yes, I like it better without the dots. But, I edited your post in the quote above to make it even more readable online. See how paragraphs are used? I appreciate your effort to improve readability. Now please just add paragraphs and we can all be happy reading your posts. Feel free to write in a conversational tone. That isn't the problem. The problem is that we can't read large blocks of text without standard looking breaks.

 

[wahdangun]

So how come chips with the same # of cores, and the same base frequency, and the same architecture, can have different TDP ratings?
If you look around at Intel chips, you'll find TDP inconsistencies.

Because you can't compare the TDP between them, and it's worst since they introduce speculative turbo, even in the same chip you can have different performance variation.