Intel's Future Chips: News, Rumours & Reviews

[juanrga]

This is an Intel thread. So I will be brief:

* There are two 7nm. AMD will use 7SOC. IBM will use 7HPC. The 5GHz claim is for 7HPC.

* Similar situation with 14nm. AMD uses 14LPP; IBM uses 14HP. IBM CPUs like the Z-14 hit 5.2GHz on 14nm. AMD 14nm CPUs cannot.

 

[-Fran-]

Are there 2 flavours of "7nm" in GloFo? Is AMD going to keep on using the efficient one (hence, the 55%+ efficiency one?) or put some in the high performance node (if there is one)?

Plus, like I said, I have no idea what they call "performance". The obvious implication is "hertz", but there have to be caveats, since they're not explicitly calling it out.

 

[YoAndy]

Id really love for AMD to hit 5GHz because that will bring better competition to the table but that seems to good to be true.

 

[gamerk316]

Id really love for AMD to hit 5GHz because that will bring better competition to the table but that seems to good to be true.

Unlikely. AMDs been going the more efficient route for a while now, and has historically shied away from pure speed (Bulldozer being the notable exception). Nevermind the baseline cooling requirements that drives up platform cost.

There's also the side effect that competing with Intel at the high end forces Intel to release an even bigger monster; AMD is better of leaving that segment strictly to Intel and competing in the low/mid-range markets.

 

[YoAndy]

Id really love for AMD to hit 5GHz because that will bring better competition to the table but that seems to good to be true.

Unlikely. AMDs been going the more efficient route for a while now, and has historically shied away from pure speed (Bulldozer being the notable exception). Nevermind the baseline cooling requirements that drives up platform cost.

There's also the side effect that competing with Intel at the high end forces Intel to release an even bigger monster; AMD is better of leaving that segment strictly to Intel and competing in the low/mid-range markets.

Yeah I feel the same way, I don't think 7mm Ryzen will be able to pass 4.5GHz.. According to GLOFO their 7nm FinFET will be the highest performing technology offering from the pure play foundry to date. Delivering double the density of today’s 14nm & 16nm FinFET technologies and a 30% performance improvement. So we could expect 30% max performance improvement over the current 14nm.

@goldstone77 30% performance improvement won't get you to 5GHz..

 

[goldstone77]

HPC= high performance computing
SOC= system on a chip

HPC will be your desktop/server processors
SOC will be laptops/notebooks/mobile devices etc

 

[aldaia]

@goldstone77 30% performance improvement won't get you to 5GHz..

1800X freq is 3.6 GHz base - 4 GHz turbo.
Not that i expect those frequencies (specially base), but .... 30% extra means 4.7 GHz base - 5.2 GHz turbo

 

[gamerk316]

@goldstone77 30% performance improvement won't get you to 5GHz..

1800X freq is 3.6 GHz base - 4 GHz turbo.
Not that i expect those frequencies (specially base), but .... 30% extra means 4.7 GHz base - 5.2 GHz turbo

You are confusing "performance" with "clockspeed".

Performance = IPC * Clock * Number of Cores

The ways AMD can gain 30% performance improvements range from increasing Clock, increasing singe core performance though IPC gains, adding more cores, or a combination of the three. It's very unlikely any performance gains will be 100% clock.

Case in point, moving from 8 cores base to 10 cores base gets you 25% gains right there. So it wouldn't be terribly hard to reach 30% performance gains, best case, without even touching the clock.

In other news: Better then expected results leads Intel to a 4.5% stock gain. And Intel plans to have chips that address Specter/Meltdown out in Q4 2018.

https://www.slashgear.com/intel-spectre-and-meltdown-proof-cpus-coming-this-year-25517122/

 

[juanrga]

Once more time, Taking Glofo graphs for 7nm to get clocks of Zen2 chips is so useless as taking Glofo graphs for 14nm and claim that Ryzen can hit 6GHz because Glofo said the 14nm node has 55% higher performance than 28nm:

And Glofo did claim that 14LPP gives 55% higher performance than 28HPP

Does this mean that 14LPP is a 6GHz node because 28HPP was capable of 4GHz? Obviously no.

We already discussed that 7nm Glofo graph in some AMD thread, and it was demonstrated then that the graph was derived from a model with idealized transistors, not the transistors one can find in real CPUs.

Now back on topic.

Intel earning reports confirms what I have been saying during last months: CofeeLake and Skylake are selling very well. Indeed, Intel announced record sales of i7s and a huge 20% gain in the datacenter thanks to the new Xeons.

Intel also announced 10nm processors started shipping the last year, and will continue to ship this year, with volume production for second half.

Intel also announced Meltdown and Spectre will be handled at silicon-level this year. I interpret this as the new CPUs that will be released this year will fix this security flaw.

 

[-Fran-]

Once more time, Taking Glofo graphs for 7nm to get clocks of Zen2 chips is so useless as taking Glofo graphs for 14nm and claim that Ryzen can hit 6GHz because Glofo said the 14nm node has 55% higher performance than 28nm:

And Glofo did claim that 14LPP gives 55% higher performance than 28HPP

Does this mean that 14LPP is a 6GHz node because 28HPP was capable of 4GHz? Obviously no.

We already discussed that 7nm Glofo graph in some AMD thread, and it was demonstrated then that the graph was derived from a model with idealized transistors, not the transistors one can find in real CPUs.

Right... 28nm in a totally different architecture is a very good apples to apples comparison of the different nodes... Can you claim if AMD made Pilediver in 14nm (current node), they wouldn't have been able to make the FX9K into a 125W CPU keeping the clocks? Can you confirm or deny that in any way shape or form with data (that doesn't come from a dark place)?

 

[YoAndy]

Once more time, Taking Glofo graphs for 7nm to get clocks of Zen2 chips is so useless as taking Glofo graphs for 14nm and claim that Ryzen can hit 6GHz because Glofo said the 14nm node has 55% higher performance than 28nm:

And Glofo did claim that 14LPP gives 55% higher performance than 28HPP

Does this mean that 14LPP is a 6GHz node because 28HPP was capable of 4GHz? Obviously no.

We already discussed that 7nm Glofo graph in some AMD thread, and it was demonstrated then that the graph was derived from a model with idealized transistors, not the transistors one can find in real CPUs.

Right... 28nm in a totally different architecture is a very good apples to apples comparison of the different nodes... Can you claim if AMD made Pilediver in 14nm (current node), they wouldn't have been able to make the FX9K into a 125W CPU keeping the clocks? Can you confirm or deny that in any way shape or form with data (that doesn't come from a dark place)?

If you think that Zen 2019 will be a 5GHz processor, im gonna have to laugh.

 

[-Fran-]

Once more time, Taking Glofo graphs for 7nm to get clocks of Zen2 chips is so useless as taking Glofo graphs for 14nm and claim that Ryzen can hit 6GHz because Glofo said the 14nm node has 55% higher performance than 28nm:

And Glofo did claim that 14LPP gives 55% higher performance than 28HPP

Does this mean that 14LPP is a 6GHz node because 28HPP was capable of 4GHz? Obviously no.

We already discussed that 7nm Glofo graph in some AMD thread, and it was demonstrated then that the graph was derived from a model with idealized transistors, not the transistors one can find in real CPUs.

Right... 28nm in a totally different architecture is a very good apples to apples comparison of the different nodes... Can you claim if AMD made Pilediver in 14nm (current node), they wouldn't have been able to make the FX9K into a 125W CPU keeping the clocks? Can you confirm or deny that in any way shape or form with data (that doesn't come from a dark place)?

If you think that Zen 2019 will be a 5GHz processor, im gonna have to laugh.

Let me put some context for you, that you might find interesting:

Bristol Ridge: A12-9800E - 4 cores (2M/4T) - 3.1Ghz / 3.8GHz - Radeon R7 8CUs / 900MHz - 35W.
Rave Ridge: Ryzen 5 2400G - 4C/8T - 3.6Ghz/3.9Ghz - Vega 11 11CUs / 1250Mhz - 65 W (configurable 45–65 W).

Do you see what I'm getting at? Have you seen it yet?

This is by no means proof that it will happen, but keep in mind Bulldozer is a "speed demon" uArch (by Juan's words) that are optimized to get as much speed as possible. So, keep that in mind when you read the above information.

 

[goldstone77]

GlobalFoundries is claiming they are making 5GHz operating devices. How else do you expect performance to increase when you shrink down the process? Faster moving transistors! How do you make them operate faster? Design them to function faster! What do we know? GlobalFoundries said they are making 5GHz operating Server, Data Center, ASICS parts! GlobalFoundries have made massive beneficial changes to the design of their transistors! It's not rocket science it's deductive reasoning based on the information provided.
Here is the claim of 5GHz operating devices:

Here is the transistor performance at 7nm compared to 14nm

Here is what the transistor looks like at 14nm vs. what they look like at 7nm. Notice the drastic change in shape of the transistors!

We know that a long rectangular slightly rounded edges provides the best characteristics for high performance transistors! GlobalFoundries 14nm transistors look to be closer to the 22-nm on the chart below, while the 7nm looks closer to the 10nm. While being a much more refined smooth design as well.

Here is Intel's 10nm transistor design. It's looks really similar to GlobalFoundries 7nm.

Here are the details of different cells or IP blocks that can be implemented of which include 7nm HPC. It's as simple as saying okay build on this IP foundation for my 7nm chips and they select 9T libraries for the process:

 

[gamerk316]

GlobalFoundries is claiming they are making 5GHz operating devices. How else do you expect performance to increase when you shrink down the process? Faster moving transistors! How do you make them operate faster? Design them to function faster! What do we know? GlobalFoundries said they are making 5GHz operating Server, Data Center, ASICS parts! GlobalFoundries have made massive beneficial changes to the design of their transistors! It's not rocket science it's deductive reasoning based on the information provided.
Here is the claim of 5GHz operating devices:

Here is the transistor performance at 7nm compared to 14nm

Here is what the transistor looks like at 14nm vs. what they look like at 7nm. Notice the drastic change in shape of the transistors!

We know that a long rectangular slightly rounded edges provides the best characteristics for high performance transistors! GlobalFoundries 14nm transistors look to be closer to the 22-nm on the chart below, while the 7nm looks closer to the 10nm. While being a much more refined smooth design as well.

Here is Intel's 10nm transistor design. It's looks really similar to GlobalFoundries 7nm.

Here are the details of different cells or IP blocks that can be implemented of which include 7nm HPC. It's as simple as saying okay build on this IP foundation for my 7nm chips and they select 9T libraries for the process:

Don't confuse the following:

1: "We can allow you to make a single core special purpose CPU that operates in the 5GHz range at some power envelope"

and

2 "We can allow you to make a multi-core general purpose CPU that operates in the same power envelope as the above, but properly will run 20-30% slower as a result"

Seriously people, I still remember when people were claiming Intels 45nm would allow them to clock up the Core architecture up to the 10GHz they promised for Netburst. We go through this EVERY power node; please stop.

 

[Isaiah4110]

In other news: Better then expected results leads Intel to a 4.5% stock gain. And Intel plans to have chips that address Specter/Meltdown out in Q4 2018.

https://www.slashgear.com/intel-spectre-and-meltdown-proof-cpus-coming-this-year-25517122/

Address Meltdown/Spectre by fixing the 20+ year old hardware flaw that quickly? Now THAT is some CPU news!

 

[goldstone77]

Spoiler

GlobalFoundries is claiming they are making 5GHz operating devices. How else do you expect performance to increase when you shrink down the process? Faster moving transistors! How do you make them operate faster? Design them to function faster! What do we know? GlobalFoundries said they are making 5GHz operating Server, Data Center, ASICS parts! GlobalFoundries have made massive beneficial changes to the design of their transistors! It's not rocket science it's deductive reasoning based on the information provided.
Here is the claim of 5GHz operating devices:

Here is the transistor performance at 7nm compared to 14nm

Here is what the transistor looks like at 14nm vs. what they look like at 7nm. Notice the drastic change in shape of the transistors!

We know that a long rectangular slightly rounded edges provides the best characteristics for high performance transistors! GlobalFoundries 14nm transistors look to be closer to the 22-nm on the chart below, while the 7nm looks closer to the 10nm. While being a much more refined smooth design as well.

Here is Intel's 10nm transistor design. It's looks really similar to GlobalFoundries 7nm.

Here are the details of different cells or IP blocks that can be implemented of which include 7nm HPC. It's as simple as saying okay build on this IP foundation for my 7nm chips and they select 9T libraries for the process:

Don't confuse the following:

1: "We can allow you to make a single core special purpose CPU that operates in the 5GHz range at some power envelope"

and

2 "We can allow you to make a multi-core general purpose CPU that operates in the same power envelope as the above, but properly will run 20-30% slower as a result"

Seriously people, I still remember when people were claiming Intels 45nm would allow them to clock up the Core architecture up to the 10GHz they promised for Netburst. We go through this EVERY power node; please stop.

Of course design plays it's roll in frequency. What did we learn from GlobalFoundries licensing Samsungs process? That process/IP will restrict frequency. Improved process and improved transistor designs while shrinking this process will fundamentally lead to higher operating frequencies. Compare the technological improvements being made, not Intel PR.

 

[-Fran-]

Spoiler

GlobalFoundries is claiming they are making 5GHz operating devices. How else do you expect performance to increase when you shrink down the process? Faster moving transistors! How do you make them operate faster? Design them to function faster! What do we know? GlobalFoundries said they are making 5GHz operating Server, Data Center, ASICS parts! GlobalFoundries have made massive beneficial changes to the design of their transistors! It's not rocket science it's deductive reasoning based on the information provided.
Here is the claim of 5GHz operating devices:

Here is the transistor performance at 7nm compared to 14nm

Here is what the transistor looks like at 14nm vs. what they look like at 7nm. Notice the drastic change in shape of the transistors!

We know that a long rectangular slightly rounded edges provides the best characteristics for high performance transistors! GlobalFoundries 14nm transistors look to be closer to the 22-nm on the chart below, while the 7nm looks closer to the 10nm. While being a much more refined smooth design as well.

Here is Intel's 10nm transistor design. It's looks really similar to GlobalFoundries 7nm.

Here are the details of different cells or IP blocks that can be implemented of which include 7nm HPC. It's as simple as saying okay build on this IP foundation for my 7nm chips and they select 9T libraries for the process:

Don't confuse the following:

1: "We can allow you to make a single core special purpose CPU that operates in the 5GHz range at some power envelope"

and

2 "We can allow you to make a multi-core general purpose CPU that operates in the same power envelope as the above, but properly will run 20-30% slower as a result"

Seriously people, I still remember when people were claiming Intels 45nm would allow them to clock up the Core architecture up to the 10GHz they promised for Netburst. We go through this EVERY power node; please stop.

Of course design plays it's roll in frequency. What did we learn from GlobalFoundries licensing Samsungs process? That process/IP will restrict frequency. Improved process and improved transistor designs while shrinking this process will fundamentally lead to higher operating frequencies. Compare the technological improvements being made, not Intel PR.

I'd say that is a non-trivial point to take.

For better or for worse, whenever Intel talks about die shrinks, everyone listens carefully and takes notes. No matter the reason, they're still leading it (by how much, not really worth discussing for this point), so all current "common" knowledge comes from whatever Intel has been doing and tweaking for themselves. That includes the "reasoning" why. So, it is not a trivial point to take.

Cheers!

 

[8350rocks]

Spoiler

GlobalFoundries is claiming they are making 5GHz operating devices. How else do you expect performance to increase when you shrink down the process? Faster moving transistors! How do you make them operate faster? Design them to function faster! What do we know? GlobalFoundries said they are making 5GHz operating Server, Data Center, ASICS parts! GlobalFoundries have made massive beneficial changes to the design of their transistors! It's not rocket science it's deductive reasoning based on the information provided.
Here is the claim of 5GHz operating devices:

Here is the transistor performance at 7nm compared to 14nm

Here is what the transistor looks like at 14nm vs. what they look like at 7nm. Notice the drastic change in shape of the transistors!

We know that a long rectangular slightly rounded edges provides the best characteristics for high performance transistors! GlobalFoundries 14nm transistors look to be closer to the 22-nm on the chart below, while the 7nm looks closer to the 10nm. While being a much more refined smooth design as well.

Here is Intel's 10nm transistor design. It's looks really similar to GlobalFoundries 7nm.

Here are the details of different cells or IP blocks that can be implemented of which include 7nm HPC. It's as simple as saying okay build on this IP foundation for my 7nm chips and they select 9T libraries for the process:

Don't confuse the following:

1: "We can allow you to make a single core special purpose CPU that operates in the 5GHz range at some power envelope"

and

2 "We can allow you to make a multi-core general purpose CPU that operates in the same power envelope as the above, but properly will run 20-30% slower as a result"

Seriously people, I still remember when people were claiming Intels 45nm would allow them to clock up the Core architecture up to the 10GHz they promised for Netburst. We go through this EVERY power node; please stop.

Of course design plays it's roll in frequency. What did we learn from GlobalFoundries licensing Samsungs process? That process/IP will restrict frequency. Improved process and improved transistor designs while shrinking this process will fundamentally lead to higher operating frequencies. Compare the technological improvements being made, not Intel PR.

I'd say that is a non-trivial point to take.

For better or for worse, whenever Intel talks about die shrinks, everyone listens carefully and takes notes. No matter the reason, they're still leading it (by how much, not really worth discussing for this point), so all current "common" knowledge comes from whatever Intel has been doing and tweaking for themselves. That includes the "reasoning" why. So, it is not a trivial point to take.

Cheers!

I recall when Intel was behind on process and no one cared or listened to what they were doing...

 

[YoAndy]

Spoiler

GlobalFoundries is claiming they are making 5GHz operating devices. How else do you expect performance to increase when you shrink down the process? Faster moving transistors! How do you make them operate faster? Design them to function faster! What do we know? GlobalFoundries said they are making 5GHz operating Server, Data Center, ASICS parts! GlobalFoundries have made massive beneficial changes to the design of their transistors! It's not rocket science it's deductive reasoning based on the information provided.
Here is the claim of 5GHz operating devices:

Here is the transistor performance at 7nm compared to 14nm

Here is what the transistor looks like at 14nm vs. what they look like at 7nm. Notice the drastic change in shape of the transistors!

We know that a long rectangular slightly rounded edges provides the best characteristics for high performance transistors! GlobalFoundries 14nm transistors look to be closer to the 22-nm on the chart below, while the 7nm looks closer to the 10nm. While being a much more refined smooth design as well.

Here is Intel's 10nm transistor design. It's looks really similar to GlobalFoundries 7nm.

Here are the details of different cells or IP blocks that can be implemented of which include 7nm HPC. It's as simple as saying okay build on this IP foundation for my 7nm chips and they select 9T libraries for the process:

Don't confuse the following:

1: "We can allow you to make a single core special purpose CPU that operates in the 5GHz range at some power envelope"

and

2 "We can allow you to make a multi-core general purpose CPU that operates in the same power envelope as the above, but properly will run 20-30% slower as a result"

Seriously people, I still remember when people were claiming Intels 45nm would allow them to clock up the Core architecture up to the 10GHz they promised for Netburst. We go through this EVERY power node; please stop.

Of course design plays it's roll in frequency. What did we learn from GlobalFoundries licensing Samsungs process? That process/IP will restrict frequency. Improved process and improved transistor designs while shrinking this process will fundamentally lead to higher operating frequencies. Compare the technological improvements being made, not Intel PR.

I'd say that is a non-trivial point to take.

For better or for worse, whenever Intel talks about die shrinks, everyone listens carefully and takes notes. No matter the reason, they're still leading it (by how much, not really worth discussing for this point), so all current "common" knowledge comes from whatever Intel has been doing and tweaking for themselves. That includes the "reasoning" why. So, it is not a trivial point to take.

Cheers!

I recall when Intel was behind on process and no one cared or listened to what they were doing...

Is not that Easy to recall when Intel has been ahead since their launch of their Core 2 Duo processors back in 2006. Your name actually cracks me up, AMD took a misstep in its Bulldozer / Piledriver architecture (FX-series) been the worse modern processor architecture ever created by AMD.

 

[juanrga]

Is not that Easy to recall when Intel has been ahead since their launch of their Core 2 Duo processors back in 2006. Your name actually cracks me up, AMD took a misstep in its Bulldozer / Piledriver architecture (FX-series) been the worse modern processor architecture ever created by AMD.

Even former AMD CEO named that architecture "a mitigated disaster"... 😉

 

[goldstone77]

Second half of 2018 production 10nm chip
Intel Ice Lake processor as soon as possible in 2019 market

As for the Ultrabooks, Intel is planning to introduce Whiskey Lake as a replacement for another 15W quad core Kaby Lake Refresh processor. In addition to the quad-core quad-core GT2 graphics unit designed in 4 + 2, it will also feature the 4 + 3 design, later known as "Coffee Lake U," which uses the GT3 graphics unit, the more powerful Shader Cores, EDRAM As a fast on-package memory, Whiskey Lake U and Coffee Lake U will feature 14 + + nm process technology.

For higher performance, Coffee Lake H plans to use Hexacore chips with TPD thermal power of 45W and will be four cores in the same footprint as the Kaby Lake H, with the reported clocking of up to 4.8 GHz and top models or will likely Is an open multiplier Core i9-8950HK. To date, Intel has introduced the Core i9 with only the Core i9-7980XE desktop processor with 18 CPU cores.

 

[goldstone77]

Look at the time table in the title:

Second half of 2018 production 10nm chip
Intel Ice Lake processor as soon as possible in 2019 market

Also:

Coffee Lake H plans to use Hexacore chips with TPD thermal power of 45W and will be four cores in the same footprint as the Kaby Lake H, with the reported clocking of up to 4.8 GHz and top models or will likely Is an open multiplier Core i9-8950HK.

 

[gamerk316]

Coffee Lake H plans to use Hexacore chips with TPD thermal power of 45W and will be four cores in the same footprint as the Kaby Lake H, with the reported clocking of up to 4.8 GHz and top models or will likely Is an open multiplier Core i9-8950HK.

Source? Looks like a badly translated google translate to me. I also doubt the 4.8 GHz top end speed @ 45W; seems about 1GHz too high.

 

[juanrga]

Icelake is 10nm+, not 10nm.
"Intel has introduced the Core i9 with only the Core i9-7980XE desktop processor with 18 CPU cores" LOL No. What happened to the i9-7960X, i9-7940X, i9-7920X, and i9-7900X introduced earlier?

https://www.notebookcheck.net/Intel-Core-i9-8950HK-SoC.279270.0.html

I see no problem with 4.8GHz. Current Kabylake R chips hit 4.2GHz

 

[goldstone77]

The source is on the translated page because I can't read the native language lol. I found it yesterday when searching for new information on up coming chips. That definitely looks like a ridiculous competitive slide Intel would put out! I think it's legit, but only time will tell if I'm right or not. http://www.hkepc.com/16278
All that aside, since it's an i9 I wonder if that means it will have quad channel ram?