Confirmed - i9 9900k will have soldered IHS, no more toothpaste TIM

Page 9 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

skaertus

Senior member
Mar 20, 2010
217
28
91
Really now? You want Intel to destroy AMD? How would that be helpful?

Talk about short term thinking.

As a market strategy for Intel, I meant. For us consumers it is good that AMD is alive and well, and delivers processors that push Intel to improve.
 

StinkyPinky

Diamond Member
Jul 6, 2002
6,761
777
126
Intel will release the 9900k at a pretty hefty price imo, but will drop it once Zen 2 hits the market. As that is what they are using this CPU to compete against. They need a CPU in 2019 to compete on the high end and the 9900k will be it.

Hard to tell how the AMD 3700X will do, but I'm thinking probably 10 cores at 4.5 so 8 cores at 5 should be within scope for most use cases.
 
  • Like
Reactions: Lodix and IEC

alcoholbob

Diamond Member
May 24, 2005
6,271
323
126
....
This site explains it in more detail then me...
Maybe links and picture should of been given with all the intel bashing this forum has lately...

https://overclocking.guide/the-truth-about-cpu-soldering/



So yeah, after the many thermal cycles, or so eventually it leads to micro cracks.



My honest take on it...

Solder is nice as it did improve thermal performance.
Lets be realistic here, the Thermal Conductivity for metal is in a factor of at least 10x greater then that of any oil or paste..
But to a non overclocker, i dont think that thermal conductivity is important.
Infact to most non overclocker, which tends to be either basic consumer machines, or office machines, i think longivity and more robust engineering is probably preferred.

To an overclocker tho, a moderate one at the least, that thermal conductivity can spell anywhere near 15-20% headroom.
So, i honestly feel, all "K"'s should of been soldered. A overclocker is not looking to keep the chip long term, and would rather want that thermal performance.

However putting back solder on the cpu to me is honestly mixed as we got easy methods of deliding a cpu. There are kits now to make it easy, and the razor with silicon wedges like how i did my old Opty's back in the days are no longer needed.

If i was going to overclock the chip to hell, i would most likely delid it, skip putting the IHS back on completely, slap a flat waterblock on it with a direct die kit like these:





Because the IHS is to me another thermal barrier which i would want to get rid of.
And a Soldered CPU would make it more difficult to do so....

But *shrug* that is just me....

Hmm, actually now that more and more builds are mini itx/cube style, I think thermal conductivity is more important than ever.

I used to have a 5820k in a mini itx build in the Silverstone SG13, which is a soldered chip. I don't overclock, and use a low profile Noctua L9i cooler. It would run around 65-70C on load with stock clocks. But my system died due to a power surge.

So I replaced the 5820k with a 8600k, which is a not a soldered chip, and threw it into the same case with the same cooler. Even downclocking the CPU to 3.2GHz it still hits 85-90C on load during gaming (mostly DOTA/Civ 6). And even though the 8600k is built on a smaller process with and uses less power, it generates a ton of more heat due to the TIM vs solder. So for my build, I have to go with either a soldered chip or a delidded chip in order to keep temps reasonable.
 
Aug 11, 2008
10,451
642
126
Intel will release the 9900k at a pretty hefty price imo, but will drop it once Zen 2 hits the market. As that is what they are using this CPU to compete against. They need a CPU in 2019 to compete on the high end and the 9900k will be it.

Hard to tell how t14he AMD 3700X will do, but I'm thinking probably 10 cores at 4.5 so 8 cores at 5 should be within scope for most use cases.
Dont want to get sidetracked to zen again, but since it has taken multiple refinements of 14 nm to reach even a reasonable chance of all core 5ghz, i am still skeptical that zen 2 will reach it just like that, right out of the box.
 

RichUK

Lifer
Feb 14, 2005
10,320
672
126
450-500 bucks for the 9900k would be a sweet deal.

Just think, Intel has gone from releasing the 7700k quad core to a higher frequency octa core on the mainstream platform in less than 2 years.
 
  • Like
Reactions: Arachnotronic

Alpha One Seven

Golden Member
Sep 11, 2017
1,098
124
66
They moved away from solder not because it was cheaper, which is the soap box myth most people tout.
It was because as the waffer got thinner, and the node process smaller, the chip became more fragile.
They noticed the metal solder had a higher instance of damaging the die from the constant phase change stress (going from liquid state to solid ... on and off on a PC), and hence changed to paste.

Although it came at a cost of performance, to intel's eyes, the performance prices was enough merit over the longevity.
This is why they changed to paste.

And now i will see a lot of people start ranting because they can no longer delid the cpu, and run the die naked.
You can still delid, just desolder first.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
I hope at least the 8-core die has partial hardware fixes for meltdown and spectre like whiskey lake. I mean they did put in the work and the 8-core is a new die. 6-core and lower will most likely just be rebrands of the existing dies.
 

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
While none of these processors have launched so far, I can see how the future line-up is going to be, based, of course, on the current knowledge (https://wccftech.com/intel-9th-gen-core-i9-9900k-core-i7-9700k-8-core-cpu-preorder/).

  • The 9th gen seems like a beefed-up version of the 8th gen (so much that it is called Coffee Lake Refresh). It is the 4th generation of Intel chips using the same architecture of Skylake (released in 2015) and using the same 14nm process of Broadwell (released in 2014). This is a significant setback from what we have seen all the years before. At least Intel made it up adding more cores in the 8th generation, but that is due to competition from AMD.
  • The i9-9900K is the only processor in the line up so far that uses HyperThreading. Intel owns patents for HT since the days of Pentium 4, some 18 years ago. It is just sad that Intel is limiting this old tech to its most expensive processor (the i9), while AMD allows everyone to have access to it.
  • The i9-9900K seems to be really powerful, but at a cost. It reached 10,719 in 3DMark, according to sources (https://www.guru3d.com/news-story/i...rk-leaks-roughly-25-faster-than-i7-8700k.html), which is 35% more than the i7-8700K. However, it will also be 25% more expensive than the i7-8700K (in the range of USD 450, while the i7-8700K launched for USD 359), so they do not compete in the same category.
  • The i9-9900K is only 17% faster than the Ryzen 2700X, according to those benchmarks, but it is 36% more expensive than its price at launch (USD 450 versus USD 329).
  • On Geekbench (https://wccftech.com/intel-core-i9-9900k-core-i7-9700k-core-i5-9600k-cpu-performance-leak/), the i9-9900K performed 6248 in single core, and 33037 in multicore. The i7-9700K shows more modest results. Compared to the previous generation i7-8700K (https://hexus.net/tech/news/cpu/121925-heres-intel-core-i7-9700k-scores-geekbench/), the i7-9700K will be 4.7% faster in single core, and 9.8% faster on multi-core. Intel clearly chose to limit speed gains by adding two cores but removing HT, to give consumers no too much. Both chips will cost the same at launch, so expect gains of less than 10% in performance.
  • The i5-9600K has a base clock of only 100 MHz faster than the i5-8600K. The i5-9600 and the i5-8500 will have the same base clock of i5-8600 and i5-8500, and only a slight increase in the boost clock. The Core i5-9400 will have an increase of only 100 MHz. And the and the Core i-8300 seems to have simply rebranded i3-9100.
So, once again, I see no real gains in terms of speed from one generation to the other. Of course there are improvements, but only if the price is increased as well. The expected improvement from one generation to the other at the same price point is close to zero. Very disappointing.
So you don't really know what you're talking about then? It's just your opinion mixed with sketchy articles and questionable benchmarks.
 

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
I hope at least the 8-core die has partial hardware fixes for meltdown and spectre like whiskey lake. I mean they did put in the work and the 8-core is a new die. 6-core and lower will most likely just be rebrands of the existing dies.
It's Coffee Lake, it's not new. It has no such fixes. It's a desktop chip, and typical desktop users shouldn't be concerned with Meltdown or Spectre.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
It's Coffee Lake, it's not new. It has no such fixes. It's a desktop chip, and typical desktop users shouldn't be concerned with Meltdown or Spectre

It's a new 8-core die so for sure it needed a new layout / design and hence fixing meltdown like in whiskey lake isn't out of the question since well they already did the work with whiskey lake. A hardware fix would be a compelling reason to go with the 8-core instead of the 6-core.

While I agree that Spectre is overblown especially fro consumers, meltdown software fix does have a large penalty on ssd performance so yeah this actually does matter since AFAIK you can't opt-out of the software fix.
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
I hope at least the 8-core die has partial hardware fixes for meltdown and spectre like whiskey lake. I mean they did put in the work and the 8-core is a new die. 6-core and lower will most likely just be rebrands of the existing dies.


Even the 6 core could have hardware fixes because it's a new stepping apparently, at least from the i5-9600K we have a valid reason.

Intel Core i5-9600K
GenuineIntel Family 6 Model 158 Stepping 12
https://browser.geekbench.com/v4/cpu/9754549


8th Gen Coffeelake is based on Stepping 10 and therefore 6C i5-9600k is not just a renaming.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Even the 6 core could have hardware fixes because it's a new stepping apparently, at least from the i5-9600K we have a valid reason.

Intel Core i5-9600K
GenuineIntel Family 6 Model 158 Stepping 12
https://browser.geekbench.com/v4/cpu/9754549


8th Gen Coffeelake is based on Stepping 10 and therefore 6C i5-9600k is not just a renaming.

that would be at least some good thing about this 9th series at least if this is the hardware fix that solves the ssd performance issues.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
What kind of lead do you see the 9900K having then? At 5GHz that is

Against a bone stock 2700X, ~ 37% in MT and > 31% in ST.

Based on the average IPC difference of ~14% and the known average frequencies of the 2700X (=< 4.35GHz ST, ~ 4.05GHz MT).
Naturally if the IPC or the SMT yield has changed in CFL-S8 (which is unlikely), then that will throw off the figures.

Practically the ST difference will most likely be even larger, because a chip which can run at 5GHz on all eight cores should have no issues in reaching 5.2GHz+ on 1-2 cores.
 
  • Like
Reactions: Drazick

epsilon84

Golden Member
Aug 29, 2010
1,142
927
136
Against a bone stock 2700X, ~ 37% in MT and > 31% in ST.

Based on the average IPC difference of ~14% and the known average frequencies of the 2700X (=< 4.35GHz ST, ~ 4.05GHz MT).
Naturally if the IPC or the SMT yield has changed in CFL-S8 (which is unlikely), then that will throw off the figures.

Practically the ST difference will most likely be even larger, because a chip which can run at 5GHz on all eight cores should have no issues in reaching 5.2GHz+ on 1-2 cores.

That is actually a very big lead if true, I don't believe Intel has had this kind of performance advantage since perhaps Skylake vs Excavator?

Gotta give Intel credit for squeezing every last iota of performance out of the ageing 14nm process / architecture, it actually will be interesting to see if AMD can exceed this level of performance with Zen 2, especially if they are still topping out at 8C for the Ryzen 7 SKUs. Of course more cores will even the ledger (or exceed the 9900K) in terms of MT performance, but I have my doubts that the ST performance gap can be totally bridged as that will require a very significant IPC / fmax uplift, very exciting times ahead either way though for PC enthusiasts, just need Intel to sort out their 10nm mess now so we can have continued competition beyond 2020
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
This is the point where the manufacturing process really starts hurting AMD.
Previously AMD has been able to exceed or match in MT performance by having more cores, but now when that advantage is gone it will be pretty much suffering until the 7nm parts arrive.
 
  • Like
Reactions: Drazick

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
That is actually a very big lead if true, I don't believe Intel has had this kind of performance advantage since perhaps Skylake vs Excavator?

Gotta give Intel credit for squeezing every last iota of performance out of the ageing 14nm process / architecture, it actually will be interesting to see if AMD can exceed this level of performance with Zen 2, especially if they are still topping out at 8C for the Ryzen 7 SKUs. Of course more cores will even the ledger (or exceed the 9900K) in terms of MT performance, but I have my doubts that the ST performance gap can be totally bridged as that will require a very significant IPC / fmax uplift, very exciting times ahead either way though for PC enthusiasts, just need Intel to sort out their 10nm mess now so we can have continued competition beyond 2020

There is some exaggeration there. First the IPC difference was closer to 11% for Zen 1. Improved memory and CCX latency and a few minor changes has brought it closer to 7-8% IPC difference. Also It seems wrong to use 2700x turbo settings vs. Overclocked imaginary 9900k clocks. You are looking closer to 4.3GHz for both MT and ST there. That's also on un-patched launch Zen + against Zen 1/Coffelake info. There was drop in performance on Intel CPU's. not great but still a percent or two. The difference is significant though. Hard to see in any given application (the 20-25% difference will probably equal to 5% in low res gaming and 10% in Productivity tools) but regardless significant. Though there is a near zero chance that that consumer non-APU Zen 2 has an 8c limit.
 
Last edited:

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
There is some exaggeration there. First the IPC difference was closer to 11% for Zen 1. Improved memory and CCX latency and a few minor changes has brought it closer to 7-8% IPC difference. Also It seems wrong to use 2700x turbo settings vs. Overclocked imaginary 9900k clocks. You are looking closer to 4.3GHz for both MT and ST there. That's also on un-patched launch Zen + against Zen 1/Coffelake info. There was drop in performance on Intel CPU's. not great but still a percent or two. The difference is significant though. Hard to see in any given application (the 20-25% difference will probably equal to 5% in low res gaming and 10% in Productivity tools) but regardless significant. Though there is a near zero chance that that consumer non-APU Zen 2 has an 8c limit.

The IPC difference between Pinnacle Ridge and Coffee Lake (GPZ patched) is 14.4% on average, unless you omit all workloads which take advantage from wider than 128-bit instruction.

YuvqEF9.png


2700X is also unable to reach 4.3GHz on all cores. 4.2GHz is the typical best case scenario, but that means you will lose 150MHz worth of ST performance.
At stock the typical clocks are 4.35GHz for ST workloads and 4.05GHz for all-core workloads.
 
  • Like
Reactions: Drazick

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
Though there is a near zero chance that that consumer non-APU Zen 2 has an 8c limit.

I think the odds are significantly above zero, that Desktop Ryzen 3000 series is still 8 core. I am close to 50:50 on 3000 series desktop Ryzen having more than 8 cores. I certainly wouldn't be the least bit surprised in either case.

Some enthusiasts are quick to leap to the conclusion that: everything possible will arrive in the next iteration.

The thinking is: More cores are possible at 7nm, therefore it must have more cores.

But business doesn't necessarily work that way. They may want to stretch out the arrival of benefits over multiple generations, to give greater reason to upgrade on future generations, and in this case a smaller, more profitable die on fledgling 7nm process.

So Zen 2, may rely on real promised architecture IPC improvements, along with clock speed improvements, on an otherwise tried and tested topology of two 4C-CXXs .

Perhaps Zen3 with more cores also arrives with the next advance in memory bandwidth to feed all those cores.

Definitely a decent chance of another 8 core series IMO.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
The IPC difference between Pinnacle Ridge and Coffee Lake (GPZ patched) is 14.4% on average, unless you omit all workloads which take advantage from wider than 128-bit instruction.

2700X is also unable to reach 4.3GHz on all cores. 4.2GHz is the typical best case scenario, but that means you will lose 150MHz worth of ST performance.
At stock the typical clocks are 4.35GHz for ST workloads and 4.05GHz for all-core workloads.

I am at work and don't have time to search for any competing pints so if you say so, I'll take the word of a single benchmark.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
I think the odds are significantly above zero, that Desktop Ryzen 3000 series is still 8 core. I am close to 50:50 on 3000 series desktop Ryzen having more than 8 cores. I certainly wouldn't be the least bit surprised in either case.

Some enthusiasts are quick to leap to the conclusion that: everything possible will arrive in the next iteration.

The thinking is: More cores are possible at 7nm, therefore it must have more cores.

But business doesn't necessarily work that way. They may want to stretch out the arrival of benefits over multiple generations, to give greater reason to upgrade on future generations, and in this case a smaller, more profitable die on fledgling 7nm process.

So Zen 2, may rely on real promised architecture IPC improvements, along with clock speed improvements, on an otherwise tried and tested topology of two 4C-CXXs .

Perhaps Zen3 with more cores also arrives with the next advance in memory bandwidth to feed all those cores.

Definitely a decent chance of another 8 core series IMO.

No there isn't, we have a decent idea of what Rome/Starship is. We also know that EPYC can't stay at 32c for very long. The fact they aren't doing a Zen+ EPYC and that they are doing a Zen 2 EPYC tells even more. There is a near zero chance that AMD will be using an 8c die for EPYC 8k, TR3 or Ryzen 3. Now obviously there is a chance of a change of numbers, with Intels move to i9 in mainstream the 10+ core Ryzen's maybe Ryzen 9's or something but AMD would be silly not use the dies with increased cores with those cores unlocked on the desktop to one up Intel on the consumer core count.