i7-9700k 8/16 core and others leaked? Is this legit?

[DrMrLordX]

my math tells me Intel would still have a 10% or more lead.

. . . which is worse than Intel's current situation. Hence my observation that the situation will not shift in Intel's favor. Things will get worse for Intel. Then we have Icelake, which faces the very real possibility of a lower Fmax and maybe a 5% increase in IPC over Coffeelake, which is performance-wise treading water for Intel. And it has to face off against Zen 2.

Intel may not get a faster mainstream desktop CPU out until Sapphire Rapids.

That is assuming of course that an 8 core CL will still overclock to close to 5 ghz.

I'm gonna give Intel the benefit of the doubt here, and say that people probably can, but as the old 9590 saying goes, "If you can cool it, you can clock it". We know what 14nm++ can do, you just gotta get rid of that heat first. Them 8c chips may get a bit toasty @ 5 GHz.

I don't think Intel is floundering much, if at all.

10nm was supposed to be out already. Intel has basically thrown in the towel on 10nm, and 10nm+ has unknown performance characteristics at this point. Though it appears that 10nm nodes will not outperform 14nm++ until 10nm++ is ready. Which is not really good news for anyone, other than Intel's competition in the foundry space.

 

[Dufus]

The long and the short of it is, if you can keep a CPU cooler, it will use less power, for the same frequency.

Are my posts really that hard to understand? Not having a go at you just genuinely surprised if that's the case and would like to know.

Higher temperatures for same clocks and workload can dissipate more power. However isn't reducing temperatures going to encourage people to overclock higher and draw even more power resulting in even higher current draw and possibly faster degradation over time?

 

[LTC8K6]

. . . which is worse than Intel's current situation. Hence my observation that the situation will not shift in Intel's favor. Things will get worse for Intel. Then we have Icelake, which faces the very real possibility of a lower Fmax and maybe a 5% increase in IPC over Coffeelake, which is performance-wise treading water for Intel. And it has to face off against Zen 2.

Intel may not get a faster mainstream desktop CPU out until Sapphire Rapids.



I'm gonna give Intel the benefit of the doubt here, and say that people probably can, but as the old 9590 saying goes, "If you can cool it, you can clock it". We know what 14nm++ can do, you just gotta get rid of that heat first. Them 8c chips may get a bit toasty @ 5 GHz.



10nm was supposed to be out already. Intel has basically thrown in the towel on 10nm, and 10nm+ has unknown performance characteristics at this point. Though it appears that 10nm nodes will not outperform 14nm++ until 10nm++ is ready. Which is not really good news for anyone, other than Intel's competition in the foundry space.

I thought I read somewhere that the problem with 10/7nm for Intel was in the IGP area, and not the CPU area?

 

[Yeroon]

Are my posts really that hard to understand? Not having a go at you just genuinely surprised if that's the case and would like to know.

VirtualLarry has the more accurate gist of it. Cooler chip temps allow better overclocks for same voltage, and it has nothing to do with thermal deltas having better heat transfer (which it seems like you are implying, though I could be mistaken).
Your statement make sense only from a thermal transfer view, but hotter silicon has higher leakage resulting in higher voltage requirements to sustain the same clock, which also results in more heat. Its a self exciting loop, one you do your best to avoid.

 

[frozentundra123456]

. . . which is worse than Intel's current situation. Hence my observation that the situation will not shift in Intel's favor. Things will get worse for Intel. Then we have Icelake, which faces the very real possibility of a lower Fmax and maybe a 5% increase in IPC over Coffeelake, which is performance-wise treading water for Intel. And it has to face off against Zen 2.

Intel may not get a faster mainstream desktop CPU out until Sapphire Rapids.



I'm gonna give Intel the benefit of the doubt here, and say that people probably can, but as the old 9590 saying goes, "If you can cool it, you can clock it". We know what 14nm++ can do, you just gotta get rid of that heat first. Them 8c chips may get a bit toasty @ 5 GHz.



10nm was supposed to be out already. Intel has basically thrown in the towel on 10nm, and 10nm+ has unknown performance characteristics at this point. Though it appears that 10nm nodes will not outperform 14nm++ until 10nm++ is ready. Which is not really good news for anyone, other than Intel's competition in the foundry space.

So everybody lauds AMD for their multithreaded performance but a 33% increase in multithreaded performance for Intel somehow "makes their situation worse"? Assuming the 8 core CL cpu does materialize, Intel will go from a 20 to 25% lead in single thread performance (probably more due to better ipc as well, especially in gaming, but I will give AMD the benefit of the doubt and assume equal ipc) and lower multithreaded performance (although an 8700k at 5 ghz is not that far from 1800x in a lot of workloads), to a smaller lead in single thread performance, *but* will in all likelihood gain the lead in multithreaded performance.

 

[Arachnotronic]

Things will get worse for Intel. Then we have Icelake, which faces the very real possibility of a lower Fmax and maybe a 5% increase in IPC over Coffeelake,

5% IPC increase after 4 years? Does this really make sense?

 

[epsilon84]

Actually, the current situation will not shift as much in Intel's favor as you might think. If it's on 14nm++ like the 8700k, I doubt that it's top clockspeed will go up by much, and all they get is a max 33% increase in MT performance assuming perfect scaling.

In contrast, AMD is going to get a 10% increase in performance across the board with Pinnacle Ridge, plus who knows what from optimized IF/IMC performance. Should be interesting to see who comes out on top.

It's not that hard to work out, really.

Right now, Intel has the IPC and clockspeed advantage, but a core disadvantage. It will achieve core parity with the 9700K, while still holding it's IPC and clockspeed advantage.

Yes, PR with higher clocks will bridge the clockspeed gap somewhat but I still expect a 9700K to clock higher and it will still maintain an IPC advantage.

So in a nutshell:
9700K = Core parity, IPC and clockspeed advantage compared to PR

The easiest way to see this is to look at a 8700K compared to an R5 1600/1600X, all 6C/12T chips. Which has the higher performance? Obviously the 8700K. Overclock the 1600 to 4GHz and you have your 10% projected gain from PR right there. A stock 8700K still outperforms an overclocked 1600. The 9700K should be in a similar position compared to Ryzen 7

 

[DooKey]

5% IPC increase after 4 years? Does this really make sense?

More like wishful thinking.

 

[IntelUser2000]

Those should be ECC-enabled versions of Coffee Lake-S and Coffee Lake-H, basically the thing that is called Xeon E3 now.

I'm having doubts what "Coffeelake" really is.

https://www.intel.com/content/dam/w.../datasheets/xeon-scalable-datasheet-vol-1.pdf

On that Datasheet, it has Xeon E as part of the Xeon Scalable Processor family. If its what they should mean, Xeon E has Skylake-SP core. Based on Anandtech's news release, Xeon E is also Coffeelake. AIDA64 news say the mobile Xeon E is called Coffeelake-H and regular Xeon E is called Coffeelake-S.

Either that, or Intel is calling every Xeon under the Xeon Scalable naming umbrella.

 

[DrMrLordX]

I thought I read somewhere that the problem with 10/7nm for Intel was in the IGP area, and not the CPU area?

Haven't seen that. All I've read is that the only products Intel will be able to ship anytime soon on the 10nm node are mobile parts with the iGPU fused off. Keep in mind that Intel was supposed to have 8c/16t Cannonlake desktop parts on the market right now (or soon). That got scrapped. So they aren't only having problems with iGPUs. Unless we all accept that Intel cancelled desktop Cannonlake 'cuz they couldn't get the iGPUs right.

So everybody lauds AMD for their multithreaded performance but a 33% increase in multithreaded performance for Intel somehow "makes their situation worse"?

33% at best. See Amdahl's law etc. And not everyone lauds AMD for their multithreaded performance. Certainly not around here.

10% more clockspeed is a win no matter what the workload.

Intel will go from a 20 to 25% lead in single thread performance (probably more due to better ipc as well, especially in gaming, but I will give AMD the benefit of the doubt and assume equal ipc) and lower multithreaded performance (although an 8700k at 5 ghz is not that far from 1800x in a lot of workloads), to a smaller lead in single thread performance, *but* will in all likelihood gain the lead in multithreaded performance.

That's what I just said. Intel's position now is comparatively stronger.

5% IPC increase after 4 years? Does this really make sense?

Anybody got any other numbers for how much higher Icelake's IPC will be compared to Coffeelake? 5% is all I've seen from the doomsayers that are predicting that Core can't really be improved, and that it'll take something radical like Sapphire Rapids to change the situation significantly.

but I still expect a 9700K to clock higher

Higher than what? Pinnacle Ridge? Of course, I expect a properly-cooled 8c Coffeelake to hit 5 GHz about as often as an 8700k. But I do not expect the 9700k to outclock an 8700k.

More like wishful thinking.

I guess there are some who think Intel won't even be able to achieve 5% improvement?

 

[Arachnotronic]

Anybody got any other numbers for how much higher Icelake's IPC will be compared to Coffeelake? 5% is all I've seen from the doomsayers that are predicting that Core can't really be improved, and that it'll take something radical like Sapphire Rapids to change the situation significantly.

The "doomsayers" are, IMHO, spreading FUD. If Intel can't improve IPC by more than 5% after four years, then they're hopelessly incompetent.

 

[jelome1989]

So then what are we looking at here? How much IPC increase will Icelake have to offer?

 

[Indus]

. A stock 8700K still outperforms an overclocked 1600. The 9700K should be in a similar position compared to Ryzen 7

But you forgot how much more that performance costs. $250 board and $380 cpu vs $70 board and $170 cpu.

Also with gaming its equal since 90% of gamers use a 60 hz display. It's like buying a F1 car for a 60 mph speed limit.

 

[TheELF]

Anybody got any other numbers for how much higher Icelake's IPC will be compared to Coffeelake? 5% is all I've seen from the doomsayers that are predicting that Core can't really be improved, and that it'll take something radical like Sapphire Rapids to change the situation significantly.

Well most people can't even agree on what IPC is...
The main bad thing is that there is no reliable way of measuring IPC,most people use cinebench or some other ancient software that hasn't been updated for years and then go around quoting that IPC hasn't improved for years while the truth is that the software is the one that hasn't improved for years and is incapable of using all the IPC of most cores,proof for that is the huge gains you get when running HTT,plenty of idle commands for hyper-threading to use up.

Well and also IPC is pretty irrelevant for most software,think about it, if even rendering(ridiculously parallel) software can't use up the IPC we have now then why even go for more IPC,why even care about it?
Overhead for bottleneck situations?Overhead for the HTT/SMT thread?
There is already enough of that if you get more then 30% out of HTT in even the heaviest situations.

 

[TheELF]

But you forgot how much more that performance costs. $250 board and $380 cpu vs $70 board and $170 cpu.

Also with gaming its equal since 90% of gamers use a 60 hz display. It's like buying a F1 car for a 60 mph speed limit.

So what's your argument here?That 90% of people will go and get the 9700K if it ever comes out?
Anybody that gets a 9700k with a 60Hz display deserves a darwin award anyway.

 

[fastamdman]

But you forgot how much more that performance costs. $250 board and $380 cpu vs $70 board and $170 cpu.

Also with gaming its equal since 90% of gamers use a 60 hz display. It's like buying a F1 car for a 60 mph speed limit.

There are many more tasks computers do than just gaming.... Price premiums will always be there for the latest and greatest technology, that isn't ever going to go away. The better and faster technology will always be more expensive than the slower cheaper technology. Just because the price to performance doesn't meet your standard, doesn't mean it doesn't for someone else. Also your 60hz display theory is kind of invalidated as it depends on the resolution as well. It's much harder to push a constant 60fps for example in 4k than it is in 1080p. Whether the game is more cpu or gpu bound doesn't matter as there are games on both sides of that fence. So if someone has a 60hz display and is playing at 1080, would XY or Z be overkill, sure. But if they are upgrading to a 4k display that is still 60hz for Christmas for example, the faster cpu will be a better performer for them. Price to Performance is measured in so many different ways that you can't simply say this cpu is the best for X amount, because you don't know EVERY task that user does. There is a difference between stock market traders, gamers, content creators, web designers, etc etc etc the list goes on for days and days. People will buy what they feel is the best value for them, or what they believe is the most future proof or is simply the newest and fastest for bragging rights. It's not gonna change regardless of what comes out next.

 

[TheELF]

It's much harder to push a constant 60fps for example in 4k than it is in 1080p. Whether the game is more cpu or gpu bound doesn't matter as there are games on both sides of that fence. So if someone has a 60hz display and is playing at 1080, would XY or Z be overkill, sure. But if they are upgrading to a 4k display that is still 60hz for Christmas for example, the faster cpu will be a better performer for them.

No,it's only harder for the GPU,the 60FPS limit stays the same so the CPU won't be needed to perform any better.

 

[Yuriman]

Well most people can't even agree on what IPC is...
The main bad thing is that there is no reliable way of measuring IPC,most people use cinebench or some other ancient software that hasn't been updated for years and then go around quoting that IPC hasn't improved for years while the truth is that the software is the one that hasn't improved for years and is incapable of using all the IPC of most cores,proof for that is the huge gains you get when running HTT,plenty of idle commands for hyper-threading to use up.

Well and also IPC is pretty irrelevant for most software,think about it, if even rendering(ridiculously parallel) software can't use up the IPC we have now then why even go for more IPC,why even care about it?
Overhead for bottleneck situations?Overhead for the HTT/SMT thread?
There is already enough of that if you get more then 30% out of HTT in even the heaviest situations.

To my knowledge, Zen needs two cycles to run AVX2 instructions, whereas Intel's CPUs need just one. So, in cases of software using newer instruction sets, Intel has made a lot of ground. It is, as you say, that software simply isn't using these newer instruction sets.

Can we blame this on Intel, for segmenting the market and leaving these instruction sets off of some processors? Or, is it that these instructions only have a narrow range of use? Or, is it software developers?

When running legacy software, Ryzen is often actually a downgrade over my 5 year old Core i5, at least according to benchmarks. Cinnebench (which is admittedly not representative of all use cases), gives a ~20% lower single core score on a 4ghz Ryzen CPU than on my 3570K @ 4.5, and most of what I do is not going to benefit at all from having more multithreaded performance. So, even 5 years later, I don't feel very compelled to upgrade.

CFL is slightly more tempting but I think I'll hold off for another generation or two.

 

[frozentundra123456]

So what's your argument here?That 90% of people will go and get the 9700K if it ever comes out?
Anybody that gets a 9700k with a 60Hz display deserves a darwin award anyway.

Or they just want the fastest cpu for other tasks, or for "future proofing", or simply for the sake of having it. By this reasoning, anyone who buys a car faster than a prius deserves a darwin award as well. Why do people buy a lexus instead of a kia, a rolex instead of a 30.00 timex, or live in a luxury house or apartment? They want more than "good enough". IMO, nothing wrong with that, and cpus is one area where it is relatively cheap to buy the best.

 

[frozentundra123456]

But you forgot how much more that performance costs. $250 board and $380 cpu vs $70 board and $170 cpu.

Also with gaming its equal since 90% of gamers use a 60 hz display. It's like buying a F1 car for a 60 mph speed limit.

As I just said, some people want more than "good enough", and 8700k is not even in the same class as a 1600. .There are plenty of other CL cpus that are faster than the 1600 which cost less, such as the 8700 non K, 8600k, and even the 8400 is faster in gaming. There are also *plenty* of motherboards way cheaper than 250.00. In fact, you can get one for about half that, and for even less when the non-overclocking motherboards come out, which will be soon I think.

 

[LTC8K6]

But you forgot how much more that performance costs. $250 board and $380 cpu vs $70 board and $170 cpu.

Also with gaming its equal since 90% of gamers use a 60 hz display. It's like buying a F1 car for a 60 mph speed limit.

Every car on the road can exceed a 60 mph speed limit. Most can easily double it. Quite a few can triple it or better.
One thing an F1 car can do is get to that 60 mph limit incredibly fast. Far faster than nearly every other car on the road.
None of us need cars that can double the speed limit, but nearly all of us own such cars, if we own a car.

 

[Indus]

As I just said, some people want more than "good enough", and 8700k is not even in the same class as a 1600. .There are plenty of other CL cpus that are faster than the 1600 which cost less, such as the 8700 non K, 8600k, and even the 8400 is faster in gaming. There are also *plenty* of motherboards way cheaper than 250.00. In fact, you can get one for about half that, and for even less when the non-overclocking motherboards come out, which will be soon I think.

That's a lie unless you game at 720p which I don't.

At the resolution I game - 1440p there's a gpu bottleneck and it'll likely stay that way till coffeelake is outdated.

Hell even a sandybridge pulls even with a coffeelake at 1440p. So why should I spend so much money on a sidegrade over say a 1080ti or maybe a 2080 soon..

 

[ozzy702]

That's a lie unless you game at 720p which I don't.

At the resolution I game - 1440p there's a gpu bottleneck and it'll likely stay that way till coffeelake is outdated.

Hell even a sandybridge pulls even with a coffeelake at 1440p. So why should I spend so much money on a sidegrade over say a 1080ti or maybe a 2080 soon..

For budget gamers you shouldn't. For people that want the very fastest there's a market. The 1080ti successor will be out in a year (far sooner than CF will be outdated) and it will absolutely smash 1440p @144hz (the resolution I game at). I current have 1080tis and wanted a CPU that could keep up. I picked up an 8700k and would have held off until the rumored 9700k but my brother wanted my 7700k box so it was a win for us both.

 

[Thunder 57]

There are many more tasks computers do than just gaming....

Thank you. I swear some people think all high performance parts are good for is gaming.

 

[DrMrLordX]

The "doomsayers" are, IMHO, spreading FUD. If Intel can't improve IPC by more than 5% after four years, then they're hopelessly incompetent.

They haven't improved IPC since Skylake.

Well and also IPC is pretty irrelevant for most software,think about it, if even rendering(ridiculously parallel) software can't use up the IPC we have now then why even go for more IPC,why even care about it?

Wat