14nm 6th Time Over: Intel Readies 10-core "Comet Lake" Die to Preempt "Zen 2" AM4

[Thunder 57]

Rumour: Ryzen 3000/x570 to have 16c/32t

I hope not. There is no need for a core war right now. We need better IPC/clocks on 8 cores. Anyone needing 16+ can use HEDT. Amdahl's Law is a thing, and we are seeing diminishing returns in most software already.

 

[dlerious]

I hope not. There is no need for a core war right now. We need better IPC/clocks on 8 cores. Anyone needing 16+ can use HEDT. Amdahl's Law is a thing, and we are seeing diminishing returns in most software already.

He talked about AMD not wanting to lose, for instance, CS:GO users. It's not core count they're concerned about, iit's clock speeds.

 

[Kaloi48]

These would be nice to see. I would also love some actual clarification on the 210w PL2. Because the Intel basically states in its sheet that it's the default target. But if Intel "suggests" or requires that the default be 210w it isn't in violation of their datasheet. What is the PL2 set to when some plugs in a 8700k? If it's ~113PL2, wouldn't this imply that Intel set the default for the 9900K and not every single mobo manufacturer ignoring the data sheet to illegally overclock your CPU for you?

210W PL2 is some sort of design guidance or recommendation from Intel. A 2017 leak of 8C Coffee Lake revealed the PL2 value:
https://www.ptt.cc/bbs/PC_Shopping/M.1512391678.A.EC6.html

 

[Kaloi48]

https://videocardz.com/newz/amd-x570-chipset-with-pcie-4-0-could-launch-at-computex

Appears that Comet Lake won't be getting an updated chipset (although that doesn't necessarily mean it won't be on a different socket) and that Intel is going to release IGP-less models of the 9th series soon.

It probably means Comet Lake-S won't come out in 2019.
The Comet Lake-S will use a new LGA 1200 socket. I'll be surprised if Intel doesn't rebrand the chipset to 4xx, even though it is the same KBL/CNL chipset.

The original leak said the 10C Comet Lake-S is planned to be released in Q1 2021. But since the information was shared at an Intel IoTG partner event, the plan for consumer desktop segment may be somewhat different, especially the K processors, which are likely to be available several months earlier.

 

[sandorski]

I hope not. There is no need for a core war right now. We need better IPC/clocks on 8 cores. Anyone needing 16+ can use HEDT. Amdahl's Law is a thing, and we are seeing diminishing returns in most software already.

Nah, more Cores, more better. AMD needs to continue pushing Intel hard, both for their own benefit and the Consumers benefit.

 

[Zapetu]

I don't see them competing with their HEDT chips. They've also promised support for AM4 til 2020.

We don't know if this is true or not but the video suggested that AMD and (some) MB vendors have different opinions whatever current AM4 MBs can handle 16 cores (possibly overclocked) or not.

I hope not. There is no need for a core war right now. We need better IPC/clocks on 8 cores. Anyone needing 16+ can use HEDT. Amdahl's Law is a thing, and we are seeing diminishing returns in most software already.

He talked about AMD not wanting to lose, for instance, CS:GO users. It's not core count they're concerned about, iit's clock speeds.

If this rumour turns out to be true then it seems to me that AMD is pushing IPC, clock speeds and core counts all at the same time really hard and there area some concerns (from the MB vendors) whatever they should be holding back a little. My guess is that unlike Intel in the past, AMD doesn't really want to hold back at this point and if they can bring 12/16 cores to AM4, they will. It's all about market share now.

I will still wait for this rumour to turn out to be true or not but high core counts with high yields would be one of the big benefits of the possible chiplet design on AM4. Because of the WSA, it still might be beneficial for AMD to keep selling their 12nm products for now and only add some higher end products at the top (for AM4 and TR4). Still they could have taken a different (more safe/traditional) approach than suggested by the video but there has already been many unexpected surprises lately, so you never know. We all might be quite dumbfounded later on for what AMD has done this time.

 

[Thunder 57]

Nah, more Cores, more better. AMD needs to continue pushing Intel hard, both for their own benefit and the Consumers benefit.

If they do have higher core variants they better charge a premium for them. They need all the money they can get.

I think they would do better with higher IPC/clocks than cores though. I think most people would take a high IPC, well clocked 8 core Zen 2 over a 10 core Coffee Lake Refresh of a refresh.

 

[PotatoWithEarsOnSide]

If there's going to be PCIe4 support, I guess that many buyers won't be overly bothered about backwards compatibility with previous boards; they'll be wanting PCIe4 boards.
Of course, many does not equal all, but AMD may consider that they'll bring over many new to AMD customers to their platform. Admittedly, some of the value goes out of the platform if a new board is required, especially for those that bought into AM4 on the back of the compatibility guarantee that AMD outlined.

 

[ozzy702]

Let me just leave this here....

View attachment 1120
View attachment 1121

Looks good to me.

If there's going to be PCIe4 support, I guess that many buyers won't be overly bothered about backwards compatibility with previous boards; they'll be wanting PCIe4 boards.
Of course, many does not equal all, but AMD may consider that they'll bring over many new to AMD customers to their platform. Admittedly, some of the value goes out of the platform if a new board is required, especially for those that bought into AM4 on the back of the compatibility guarantee that AMD outlined.

Can't they release CPUs with PCIe4 support that require new mobos but that also work in current AM4 mobos at PCIe3. That would be the best of both worlds.

 

[Vattila]

We all might be quite dumbfounded later on for what AMD has done this time.

Yeah. I'm scratching my head. I still think AMD can use the Threadripper platform to compete above 8 cores (especially with Intel's pricing), and that an iGPU is needed in the mainstream (for OEM traction), but rumours seem to point to a core-count increase. I've started a speculation thread with a poll to gauge the current expectations.

https://forums.anandtech.com/threads/speculation-ryzen-3000-series.2558009/

 

[epsilon84]

He talked about AMD not wanting to lose, for instance, CS:GO users. It's not core count they're concerned about, iit's clock speeds.

Seriously, CS:GO?! That game gets hundreds of fps on modern hardware. Yes, technically Intel is 'faster' at this game but the only people that would even care are professional CS:GO gamers that can tell the difference between 100fps vs 130fps 0.1% lows: (at 4:33 https://youtu.be/9LqaGnuCPsg )

Perhaps AMD would be better to direct their attention to optimising for AAA titles where currently Zen+ is lagging behind (in an actual noticeable way) like the Assassins Creed, Battlefield and Far Cry series.

 

[epsilon84]

If there's going to be PCIe4 support, I guess that many buyers won't be overly bothered about backwards compatibility with previous boards; they'll be wanting PCIe4 boards.
Of course, many does not equal all, but AMD may consider that they'll bring over many new to AMD customers to their platform. Admittedly, some of the value goes out of the platform if a new board is required, especially for those that bought into AM4 on the back of the compatibility guarantee that AMD outlined.

That's the downside to guaranteeing support for 4 years, new standards come out and you end up with situations like this.

I'm also sceptical about how budget B350 and B450 mobos with lower end VRMs will support a hypothetical 16C Zen 3000 chip. Many of those boards actually already have suboptimal VRMs for 8C Zen+ like the 2700X, so I understand the mobo makers concerns about doubling the core count, even on a more efficient 7nm process.

 

[mikk]

It probably means Comet Lake-S won't come out in 2019.
The Comet Lake-S will use a new LGA 1200 socket. I'll be surprised if Intel doesn't rebrand the chipset to 4xx, even though it is the same KBL/CNL chipset.

The original leak said the 10C Comet Lake-S is planned to be released in Q1 2021. But since the information was shared at an Intel IoTG partner event, the plan for consumer desktop segment may be somewhat different, especially the K processors, which are likely to be available several months earlier.

There was nothing in the original leak about a 2021 date or any other release date indication, it means your entire posting seems fake.

 

[scannall]

I hope not. There is no need for a core war right now. We need better IPC/clocks on 8 cores. Anyone needing 16+ can use HEDT. Amdahl's Law is a thing, and we are seeing diminishing returns in most software already.

I'd have to agree. While we were certainly short on cores pre-Ryzen, stuck at 4/8 for consumers it's time to let software catch up a bit. For a basic mom box, a 4/4 2200G is plenty. For a new gamer build I'd set the baseline at 6/12. Move up to 8/8 or 8/16 for a 'premium' gaming rig.

But more cores also = more heat, which means lower clocks. 7nm allows a lot more transistors per die. But I don't think blowing the budget on more cores is always the answer. Fatter and wider cores with more throughput would be a better use of that budget. Not to mention larger caches. Core counts are fine for now. Time to pump up the volume on throughput.

 

[Kaloi48]

There was nothing in the original leak about a 2021 date or any other release date indication, it means your entire posting seems fake.

The original leak is earlier than Chrisdar's Nov. 8 post, you can check the source link I PMed you.

 

[TheGiant]

That's the downside to guaranteeing support for 4 years, new standards come out and you end up with situations like this.

I'm also sceptical about how budget B350 and B450 mobos with lower end VRMs will support a hypothetical 16C Zen 3000 chip. Many of those boards actually already have suboptimal VRMs for 8C Zen+ like the 2700X, so I understand the mobo makers concerns about doubling the core count, even on a more efficient 7nm process.

exactly...
I am waiting for the ryzen 3000 series review with b350 middle budget board (not total low end)
My prediction is it won't work to the max CPU capabilities (XFR..) and if it does the memory for r3000 will have a major impact on performance and it won't work at high speeds
And to be rude, I am expecting the review sites won't do any reviews with that b350 middle level boards, but with the top end x570 boards with the reason to get max performance but also not to show the performance deficit with the budget b350 board...
I am really waiting to see if AMD and the MB makers deliver the real AM4 compatibility or it is just the same story as before- to get the max of your new CPU buy a new board, except the case where you have top level board for 200+ EUR +, where the MB makers are risking nerd rage

So to see if the result is the same as Intel's (where you need to change boards because of chipsets and ...and with the nerd rage again Intel su...s again new board) or ..

 

[beginner99]

We don't know if this is true or not but the video suggested that AMD and (some) MB vendors have different opinions whatever current AM4 MBs can handle 16 cores (possibly overclocked) or not.

I'm also sceptical about how budget B350 and B450 mobos with lower end VRMs will support a hypothetical 16C Zen 3000 chip. Many of those boards actually already have suboptimal VRMs for 8C Zen+ like the 2700X, so I understand the mobo makers concerns about doubling the core count, even on a more efficient 7nm process.

it matters what the actual defintion of "support" is. AMD can just release an AM4+ or AM5 socket and only these support the >8 core versions. The 8 core and lower ones would also be supported on AM4 and obviously AM4 would be limited to pcie3 while new socket and chipset would support pcie4.

One could also imagine that they simply use the current dies for lower end of stack (rebrand like with GPUs). So if 3800x is 16-core , 3700x 12-core, then a 3600x would be the current 2700x and so forth.

 

[DrMrLordX]

Intel doesn't disclose transisitor counts. My guess is that it would be at least 140 mm2 if it was on 14 nm, and probably a bit more.

So let's assume 150mm2 to be on the safe side, for a 14nm i3-8121 (as an added bonus the iGPU might work!). That's a 2.14x increase in density, which is not 2.7x as advertised by Intel.

I'm also sceptical about how budget B350 and B450 mobos with lower end VRMs will support a hypothetical 16C Zen 3000 chip. Many of those boards actually already have suboptimal VRMs for 8C Zen+ like the 2700X, so I understand the mobo makers concerns about doubling the core count, even on a more efficient 7nm process.

Those low-end VRMs won't work. 7nm will yield greater efficiency, but higher core counts = more of the power consumption biasing towards current rather than voltage. That hits minimalist VRM configurations where it hurts. You want a lot of phases when dealing with low voltage/high current (look at the 9900k on Z370 vs Z390).

 

[maddie]

exactly...
I am waiting for the ryzen 3000 series review with b350 middle budget board (not total low end)
My prediction is it won't work to the max CPU capabilities (XFR..) and if it does the memory for r3000 will have a major impact on performance and it won't work at high speeds
And to be rude, I am expecting the review sites won't do any reviews with that b350 middle level boards, but with the top end x570 boards with the reason to get max performance but also not to show the performance deficit with the budget b350 board...
I am really waiting to see if AMD and the MB makers deliver the real AM4 compatibility or it is just the same story as before- to get the max of your new CPU buy a new board, except the case where you have top level board for 200+ EUR +, where the MB makers are risking nerd rage

So to see if the result is the same as Intel's (where you need to change boards because of chipsets and ...and with the nerd rage again Intel su...s again new board) or ..

How is choice ever a bad thing? You actually prefer being forced to buy a new board with each gen?

Want the best & latest tech? Buy a new motherboard with that new CPU
Want a cheap increase? Buy the new CPU and stick it in your present board.

Nobody is forcing anyone to keep their motherboard, and these new tech features like PCIe4, XFR 3, better memory support, keeps the motherboard manufacturers happy as it generates new sales.

Best of both worlds.

 

[mikk]

The original leak is earlier than Chrisdar's Nov. 8 post, you can check the source link I PMed you.

LGA 1200 and a release date in Q1 2021 could be an indicator for a different architecture, LPDDR4x support as well because Skylake IMC doesn't support this. Possibly Intel gave up on ICL-H/S @10nm and porting it to 14nm, even though 14nm in 2021 looks really ugly, it could mean 10nm/7nm not before 2022 for desktop. Possibly ICL-Y/U for 10nm and ICL-H/S for 14nm.

 

[IntelUser2000]

So let's assume 150mm2 to be on the safe side, for a 14nm i3-8121 (as an added bonus the iGPU might work!). That's a 2.14x increase in density, which is not 2.7x as advertised by Intel.

The 2.7x is technically not a lie. But its only true for small cores like Atoms. Their Core was always 2x. Higher performing transistors have to be larger, higher performing circuits may be more complex(thus take more space) and higher thermal density may require spreading them out more.

But here we start to see the idiocy of their decisions. They risked killing the future of the company to enter new markets. 4 year delay on 10nm for what? To make chips that lost 4 billion/year for Tablets and sell 5 million Smartphone chips to Asus.

 

[jpiniero]

It probably means Comet Lake-S won't come out in 2019.
The Comet Lake-S will use a new LGA 1200 socket. I'll be surprised if Intel doesn't rebrand the chipset to 4xx, even though it is the same KBL/CNL chipset.

They do need to release something in the late 2019/early 2020 range. If it's not Comet, then I'm not sure what else it could be. The idea that they would release yet another 14 nm desktop part in 20/21 doesn't seem farfetched however.

 

[Topweasel]

So let's assume 150mm2 to be on the safe side, for a 14nm i3-8121 (as an added bonus the iGPU might work!). That's a 2.14x increase in density, which is not 2.7x as advertised by Intel.



Those low-end VRMs won't work. 7nm will yield greater efficiency, but higher core counts = more of the power consumption biasing towards current rather than voltage. That hits minimalist VRM configurations where it hurts. You want a lot of phases when dealing with low voltage/high current (look at the 9900k on Z370 vs Z390).

AMD has implied that AMD can double performance for nearly free. I expect something like a qualifcation list for the the highest X chips if they go higher than lets say 100w rated TDP. But really I don't know why we thing AMD couldn't offer more cores at competitive clocks in the same power range. Core clock wise we don't have a good idea how offloading the IO is going to affect the ability to clock the chiplets either.

 

[DrMrLordX]

The 2.7x is technically not a lie. But its only true for small cores like Atoms. Their Core was always 2x. Higher performing transistors have to be larger, higher performing circuits may be more complex(thus take more space) and higher thermal density may require spreading them out more.

But here we start to see the idiocy of their decisions. They risked killing the future of the company to enter new markets. 4 year delay on 10nm for what? To make chips that lost 4 billion/year for Tablets and sell 5 million Smartphone chips to Asus.

Intel wasn't going to sell Tremont (or any other 10nm Atom chip) to cellphone/tablet manufacturers anyway. Or at least, I don't think they were. What they wanted to do with newer Atom iterations was continue to strengthen their Phi offerings. Phi - as we knew it - is now dead. We have Cascade Lake-AP, we have Loihi, we have . . . unspecified dGPUs coming in 2020. Intel has a lot of divergent "massively parallel" compute options hitting the market for various sectors. Phi, which was meant to serve them all, is no longer a part of the picture.

AMD has implied that AMD can double performance for nearly free. I expect something like a qualifcation list for the the highest X chips if they go higher than lets say 100w rated TDP. But really I don't know why we thing AMD couldn't offer more cores at competitive clocks in the same power range. Core clock wise we don't have a good idea how offloading the IO is going to affect the ability to clock the chiplets either.

I have no doubt that they can double performance in the same power envelope. But remember that power is based on voltage and current. If you maintain power by:

1). Increasing transistor count (notably, core count)
2). Reducing threshold voltage
3). Increasing current

then you are going to put even more stress on 4+2 VRM configs common to B350 motherboards that already struggle with every octocore Zen/Zen+ chip ever released. There's more in play than, "can it handle a 95W TDP processor"?

 

[TheGiant]

How is choice ever a bad thing? You actually prefer being forced to buy a new board with each gen?

Want the best & latest tech? Buy a new motherboard with that new CPU
Want a cheap increase? Buy the new CPU and stick it in your present board.

Nobody is forcing anyone to keep their motherboard, and these new tech features like PCIe4, XFR 3, better memory support, keeps the motherboard manufacturers happy as it generates new sales.

Best of both worlds.

The choice is ofc good thing. THe bad is the presentation. All around the web tells us AM4 B350 is ok.
But I think buying a new r7-3700X in medium b350 board will lock me to low or no overcklocking (or even reaching default turbo clocks because of VRMs) and low speed DDR4. Those signal you wont get on the internet as a non technical user.
Joe simply open "tech site", see the review and expect the same results but he doesn't get them.
Then he is disappointed.
I hope I am wrong and if AMD delivers that my friends asrock b350 pro4 (currently with R1600 non oced) can run the 3700X even at default turbo clocks without throttling, then salute to AMD.
So the next gen reviews should display the boards from the 300 series which can run the the new R3XXX without problems with the 3466MHz low latency DDR4 it deserves. Otherwise, display the performance of what you can get with the old board. That will be a correct signal. We are not getting that signal now. Only tech staff knows it.