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

[DrMrLordX]

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.

Um

Most B350 boards aren't suitable for the 1700, 1700x, 1800x, 2700, or 2700x. Only a few can pull it off, with some risk of VRM failure. What makes you think a hypothetical 3700x - especially one that has 12 or even 16 cores - will work with B350?

 

[TheGiant]

Um

Most B350 boards aren't suitable for the 1700, 1700x, 1800x, 2700, or 2700x. Only a few can pull it off, with some risk of VRM failure. What makes you think a hypothetical 3700x - especially one that has 12 or even 16 cores - will work with B350?

I don't. Ask the Joe buying it in the store. You will be surprised how many people will give you the answer.
https://www.asrock.com/mb/AMD/AB350 Pro4/index.asp#Specification my friend's board.

He bought it because - hey I have the easy upgrade. I hope he will..

as for the topic, I think 10C intel desktop- again a new board lol

 

[DrMrLordX]

Some of the ASRock B350 boards are ones I might trust with an octocore. Some of them. Not for significant overclocking though.

 

[PotatoWithEarsOnSide]

Overclocking is mostly redundant on Ryzen CPUs. XFR will take care of it for you.
Besides which XFR benchmarks were generally outperforming OCed benches.

 

[moinmoin]

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.

Intel still stands by the 2.7x density target. Which product will it be if not for an Atom of any kind?

 

[Topweasel]

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.



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"?

Citation needed. I am sure a board not made to work with the 1700x from the beginning (made for 65w CPU's) isn't going to work with a 16c monster. But there isn't some pandemic issue where boards are failing left and right on 1700x+-2700x CPU's. Poor overclocking sure. But even then not much of an issue (as they basically it a wall near that power envelope anyways). This comes off as more Reddit panic out of people deciding on their own that B350 boards weren't good power wise for their taste and echo chamber-ing it from, "the VRM's" suck to"VRM's near limit" to "People have to be having power issues, not me because I was smart and got a RoG".

 

[IntelUser2000]

Intel still stands by the 2.7x density target. Which product will it be if not for an Atom of any kind?

Yea I think the target came from egotism and trying to save face rather than any logical decisions. They knew after touting Moore's Law leadership for decades and especially how with 22nm they will dominate everything, they couldn't stand to fall behind. But they did.

This is how huge multinational corporations like Intel make mistakes.

 

[Arachnotronic]

Yea I think the target came from egotism and trying to save face rather than any logical decisions. They knew after touting Moore's Law leadership for decades and especially how with 22nm they will dominate everything, they couldn't stand to fall behind. But they did.

This is how huge multinational corporations like Intel make mistakes.

It's so ridiculous how as process technology got more difficult, they decided that taking bigger steps with all of those density boosters was a good idea. They lost sight of the fact that the manufacturing is supposed to serve the products and this is the result.

Their comments mocking TSMC 10nm as not being "real" 10nm look really dumb now. They're still not shipping anything in real volumes that matches TSMC 10nm in density, let alone TSMC 7nm.

 

[PotatoWithEarsOnSide]

Even their glue comments look dumb now.

 

[piesquared]

So does this confirm intel is adopting a moar cores strategy? Also, I'm not sure how it preempts Zen2 if it's slower. Sounds more like 'hey look at me over here in the corner''s Zen2.

 

[Topweasel]

So does this confirm intel is adopting a moar cores strategy? Also, I'm not sure how it preempts Zen2 if it's slower. Sounds more like 'hey look at me over here in the corner''s Zen2.

I think Intel always had a plan to up the core count. For the most part Coffee Lake 2017 was always going to exist and was 6c. I think they wanted to wait to 10nm to go higher. But this is the same company with a 28c Monolithic die. Keeping the core count down on consumers was about margins and segmentation and not about competition or design/process limitations. AMD has just barely made Intel readjust there plans an more on HEDT then consumer chips and the decision for Skylake Part 5 with 10 cores had to be made back when Ryzen bench mark rumors were only starting to come out. So the decision wouldn't be on competition from AMD but the understanding that at a minimum they were leaving a huge gap that AMD could fill if they didn't move forward.

As for what this choice has to do with Zen 2. Absolutely nothing. It has to compete with Zen 2. But Intel wouldn't have had much of an idea what Zen 2 was they started it. It won't be slower. That is an AMD Fan's wet dream. But it will be what the 9900k is, it might fill the 9900k's price range, though I think we see more of a X299 move here, the 9900k like configuration goes down in price but not a full tier and the new Top K will be priced higher then the current 9900k's pricing. On the performance end the base clock might be slightly reduced and the single core Turbo will be slightly increased. Somewhere in the middle it will run like a 220w CPU all based on Intel's specs and guidlines and people will say "that it's not supposed to be like that it's all on the motherboards", "but why care you are spending $550 on a CPU?"

 

[DrMrLordX]

Intel still stands by the 2.7x density target. Which product will it be if not for an Atom of any kind?

Now? I don't know. I think originally 10nm was aimed squarely at Phi though.

Citation needed. I am sure a board not made to work with the 1700x from the beginning (made for 65w CPU's) isn't going to work with a 16c monster. But there isn't some pandemic issue where boards are failing left and right on 1700x+-2700x CPU's. Poor overclocking sure. But even then not much of an issue (as they basically it a wall near that power envelope anyways). This comes off as more Reddit panic out of people deciding on their own that B350 boards weren't good power wise for their taste and echo chamber-ing it from, "the VRM's" suck to"VRM's near limit" to "People have to be having power issues, not me because I was smart and got a RoG".

Reddit panic? Sorry, I don't read there much anymore. I'm only commenting based on things said by folks like @The Stilt some time ago. Namely, those 4+2 configs are not what you want for a 1700, etc. despite those chips appearing on the supported CPU list.

 

[IntelUser2000]

Now? I don't know. I think originally 10nm was aimed squarely at Phi though.

Well, it can't be just Phi. It had to be about Atom in general.

See, they wanted 10nm in 2016. And these things take years to develop. Meaning maybe in 2010-2011 they would have conceptualized what they wanted in 2016. 2010-2011 is when they were revamping things to get into Smartphones. They only cancelled Broxton(Goldmont for Smartphones) in early part of 2016 and they were to come later that year. Since they started using contra revenue for 2014-2015 products, it would not be at least until that time when they realized maybe they were dead in the water for Smartphones.

Denial that they had serious, serious issues with 10nm continued until 2017(or maybe even now?!?).

The 22nm process was revised to fit Atom Smartphones. This whole sacrifice-Core-to-gain-phones mantra started with Otellini. Then they went even further with 14nm. Remember, both 22nm and 14nm were delayed. Former by about 4-5 months and latter maybe 6-8 months.

 

[DrMrLordX]

Overclocking is mostly redundant on Ryzen CPUs. XFR will take care of it for you.
Besides which XFR benchmarks were generally outperforming OCed benches.

Dunno if I agree with that. XFR doesn't do jack for me on this old 1800x. It really doesn't. If I lock the CPU in at 4 GHz, I get muchly much better performance than if I leave everything @ stock and let it try to boost its way to greatness. About the only area where it does "okay" are single-threaded workloads. On a 2700x, I might agree with you, though if you have a golden chip and can lock it in @ 4.3-4.4 GHz then you'd be crazy to do otherwise.

Well, it can't be just Phi. It had to be about Atom in general.

See, they wanted 10nm in 2016. And these things take years to develop. Meaning maybe in 2010-2011 they would have conceptualized what they wanted in 2016. 2010-2011 is when they were revamping things to get into Smartphones. They only cancelled Broxton(Goldmont for Smartphones) in early part of 2016 and they were to come later that year. Since they started using contra revenue for 2014-2015 products, it would not be at least until that time when they realized maybe they were dead in the water for Smartphones.

Denial that they had serious, serious issues with 10nm continued until 2017(or maybe even now?!?).

The 22nm process was revised to fit Atom Smartphones. This whole sacrifice-Core-to-gain-phones mantra started with Otellini. Then they went even further with 14nm. Remember, both 22nm and 14nm were delayed. Former by about 4-5 months and latter maybe 6-8 months.

It's possible, but once they shut down contra revenue and basically gave away their Cherry Trail chips, most of the noise they sent out to investors made it look like they weren't going to touch the mobile market ever again. Those margins were not going to get better just because Intel became more competitive at 10nm. To me it looked like Atom was restricted to networking devices, Phi, and the odd machines that today get Goldmont+. Now that we have things like Lakefield the picture is getting even hazier.

 

[PotatoWithEarsOnSide]

I get your point; XFR2 would probably have been a better qualification from me. There clearly are some 1st gen CPUs where overclocking is certainly viable. Going forward though, I think it'll be something that disappears from our thoughts.

 

[IntelUser2000]

Those margins were not going to get better just because Intel became more competitive at 10nm

We're talking about their original plans for 10nm. There's little doubt it was about Atom.

If they have majority marketshare, yes they can get great margins. $20 is only average smartphone chip price, and only true way back then. If you look at Tech Insights the latest Galaxy and iPhones spend $50-70 on the SoC. That would have resulted in not just great margins, but great revenue. 200 million high end phones with $50 ASP on the SoC is $10 billion.

 

[LTC8K6]

To pre-empt something, you need to beat it to market.

Intel could possibly help itself with overclocking. Bring out a chip that overclocks really well. Get the nerds back on board.

 

[Arachnotronic]

To pre-empt something, you need to beat it to market.

Intel could possibly help itself with overclocking. Bring out a chip that overclocks really well. Get the nerds back on board.

You can make a chip that overclocks well by clocking it conservatively out of the box.

 

[IntelUser2000]

You can make a chip that overclocks well by clocking it conservatively out of the box.

This.

We are not going to get much above 5GHz. I saw a headline from an article that said 5.5GHz with Coffeelake, only to find out it was using water cooling.

5.5GHz with water cooling was achievable 15 years ago.

 

[LTC8K6]

You can make a chip that overclocks well by clocking it conservatively out of the box.

That's not going to fool anyone, though.

 

[Topweasel]

Now? I don't know. I think originally 10nm was aimed squarely at Phi though.



Reddit panic? Sorry, I don't read there much anymore. I'm only commenting based on things said by folks like @The Stilt some time ago. Namely, those 4+2 configs are not what you want for a 1700, etc. despite those chips appearing on the supported CPU list.

It wasn't supposed to be an accusation on source. Just find that online communities specially on places like reddit where things fed through an echo chamber goes from something simple to "OMG, how could a company do that". Reading someone say well B350 boards won't work because they barely work now, sounds like that kind of issue. Sure It's not the best VRM config and any boards with that configuration would be a bad choice for anyone who dreams of overclocking. But I haven't read a single person have anything even close to an issue with power delivery from a run of the mill B350 with a 1600x, 1700x, 1800x, 2600x or 2700x. After the Intial run in March of last year where most hopped onto X370 boards, B350 boards quickly became the defacto go to chipset. Here, Reddit, everywhere we would have a ton of issues if this was the case.

 

[Arachnotronic]

That's not going to fool anyone, though.

My point though is that if there's a ton of headroom left in the silicon, it's smarter for Intel (and AMD) to just push it and get paid more for the better out of the box performance.

 

[maddie]

That's not going to fool anyone, though.

Don't know if serious, but the only way to have a very good overclocker is if you have conservative standard clocks. This should be considered almost a self obvious definition, or are we fantasizing here?

The word high is relative to another value, or is there an absolute value that is high?

 

[LTC8K6]

My point though is that if there's a ton of headroom left in the silicon, it's smarter for Intel (and AMD) to just push it and get paid more for the better out of the box performance.

I don't know.

Maybe you'd sell a lot more 4.0 chips that can be coaxed to 5.0 for $300, than you would 5.0 chips that can be coaxed to 5.2 for $500.

First we need more IPC though, so that clock speeds can be lower and still have better performance, and so that clock speed increases through overclocks have more effect.

 

[LTC8K6]

Don't know if serious, but the only way to have a very good overclocker is if you have conservative standard clocks. This should be considered almost a self obvious definition, or are we fantasizing here?

The word high is relative to another value, or is there an absolute value that is high?

I think in the past we had a wider variation in what each chip would do at what voltage, so clocks were conservative to cover them all, but it was not intentionally so. It was not a made up thing to create overclockable chips.