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

[TheGiant]

Is there a possibility to use i9-9900K with ECC memory?

 

[coercitiv]

Is there a possibility to use i9-9900K with ECC memory?

Only on the 2 socket version of the Z390 boards. /s

 

[LTC8K6]

Is there a possibility to use i9-9900K with ECC memory?

Not much of a chance of that.

You would need to get the Xeon equivalent of the 9900K.

 

[scannall]

Is there a possibility to use i9-9900K with ECC memory?

Zero chance of that.

 

[Dayman1225]

Is there a possibility to use i9-9900K with ECC memory?

No, but I imagine there will be a Xeon E version of the 9900K

 

[TheELF]

It's a relatively minor(from a user's perspective) implementation difference where AMD's version favors multi-thread and Intel's version favors single thread. It's about how resource sharing/splitting affects performance. Generally, shared resource is better for single thread, and distributed one is better for multi-thread. Your team has limited budget/time, so it has to choose between the two.

AMD get's ~10% less performance out of the "main thread" (running a single thread on 1c/2t) this 10% can be used by the SMT ,AMD get's 10% more SMT out of 10% less general performance...
And if you actually run things on a few cores and it can't boost to top turbo anymore...forgetaboutit.

 

[Markfw]

AMD get's ~10% less performance out of the "main thread" (running a single thread on 1c/2t) this 10% can be used by the SMT ,AMD get's 10% more SMT out of 10% less general performance...
And if you actually run things on a few cores and it can't boost to top turbo anymore...forgetaboutit.

This is totally wrong. AMD is a few % (less than 10) on IPC, and yes, they don't clock as high as Intels' current generation, and AMD's SMT is about 4% better in optimization and performance boost vs HT. They are not related as you are saying.

And your last sentence is totally wrong as well, the less cores running, the higher the turbo boost in clock speed.

 

[TheELF]

This is totally wrong. AMD is a few % (less than 10) on IPC, and yes, they don't clock as high as Intels' current generation, and AMD's SMT is about 4% better in optimization and performance boost vs HT. They are not related as you are saying.

And your last sentence is totally wrong as well, the less cores running, the higher the turbo boost in clock speed.

Well if it's 4% it's within the statistical margin,it's not enough to care about.
Single thread results are from one single thread running...which is when you get the highest turbo clock,run two or more threads and especially on 1c/2t pairs and the IPC takes a dive.

And actually as we have seen on the other thread, on a wide variety of software,when looking at more then one type of software,IPC is about 33% apart.

 

[Markfw]

Well if it's 4% it's within the statistical margin,it's not enough to care about.
Single thread results are from one single thread running...which is when you get the highest turbo clock,run two or more threads and especially on 1c/2t pairs and the IPC takes a dive.

And actually as we have seen on the other thread, on a wide variety of software,when looking at more then one type of software,IPC is about 33% apart.

Not 33%, that is based on overclocking the Intels higher, not IPC. And the 4% is an average of a lot of benchmarks, so, no, its not margin of error.

And even overclocked, I don't believe the 33%, unless its bench outlier. Link something that I can believe.

 

[RichUK]

ZEN 2 will close the gap. Should be quite interesting when it gets released - improved IPC, 7nm and 16 cores.

Just wonder whether it'll clock higher to challenge Intel at 5ghz and finally hit high memory overclocks (i.e. 4k mems).

 

[shortylickens]

I've never delidded in my life.

 

[AtenRa]

Well if it's 4% it's within the statistical margin,it's not enough to care about.
Single thread results are from one single thread running...which is when you get the highest turbo clock,run two or more threads and especially on 1c/2t pairs and the IPC takes a dive.

IPC remains the same. Single Thread Performance is getting lower due to lower clocks.

And actually as we have seen on the other thread, on a wide variety of software,when looking at more then one type of software,IPC is about 33% apart.

Again this is not IPC but Single Core Performance. And if you take other reviews you get different results. For example by taking Anandtechs Ryzen 2 review, the R5 2600X is within 10% of Core i7 8700K at default clocks.
And if you count the higher clocks of the Core i7 8700K, it seems from AT review that Ryzen 2 IPC is higher than CoffeeLake.

 

[TheGiant]

No, but I imagine there will be a Xeon E version of the 9900K

Is it possible to overclock XeonE ?

 

[Dayman1225]

Is it possible to overclock XeonE ?

No

 

[PeterScott]

Probably not as much as full delid +LM, but should be good for a 10C reduction, maybe more on a high power chip like the 9900K.

FWIW delidding + liquid metal typically shaves about 15C off load temps.

Solder is definitely not as good as LM + full delid.

Der8auer delidded a soldered two CPUs and switched to liquid metal and LM ran 8-10 degrees cooler ( i7-5960X) and 4-6 degrees cooler (i7-6950X) than Solder:

https://www.youtube.com/watch?v=Hf8V_UulpBk

LM + Delid is so effective because LM is highly heat conductive and because delid usually removes the adhesive which results smaller gap to CPU as well.

 

[Markfw]

Solder is definitely not as good as LM + full delid.

Der8auer delidded a soldered two CPUs and switched to liquid metal and LM ran 8-10 degrees cooler ( i7-5960X) and 4-6 degrees cooler (i7-6950X) than Solder:

https://www.youtube.com/watch?v=Hf8V_UulpBk

LM + Delid is so effective because LM is highly heat conductive and because delid usually removes the adhesive which results smaller gap to CPU as well.

If LM is so good, whats the downside ? Why would Intel/AMD not use that as their primary instead of solder/TIM ?

 

[Hail The Brain Slug]

Solder is definitely not as good as LM + full delid.

Der8auer delidded a soldered two CPUs and switched to liquid metal and LM ran 8-10 degrees cooler ( i7-5960X) and 4-6 degrees cooler (i7-6950X) than Solder:

https://www.youtube.com/watch?v=Hf8V_UulpBk

LM + Delid is so effective because LM is highly heat conductive and because delid usually removes the adhesive which results smaller gap to CPU as well.

The photo of a supposed 9900K delidded thats been going around the internet shows two things that appear to be in the benefit of enthusiasts:

1) the solder layer looks thin. Much thinner than the ~1mm that gave HW-e and BW-e the large gains from delidding.

2) the application of adhesive looks much less liberal than previous intel designs.

It's possible Intel has actually paid attention this time around and improved the two areas that, even with solder, were sore points

I suppose we'll find out for sure when we see some delids on 9900k's, I assume Der8auer will be doing one.

If they did it right, I would think the gains from delidding would be in line with Ryzen or better, only 3-4C or less.

 

[PeterScott]

If LM is so good, whats the downside ? Why would Intel/AMD not use that as their primary instead of solder/TIM ?

It's conductive, and a free running liquid, so potentially excess might shift from the die to other components and short them out.

It reacts with most other metals. Aluminum in the short term, copper in the medium term and even nickel plating in the long term.

It's just probably too hard to handle/apply correctly and hold up to shipping handling and long term usage.

 

[DrMrLordX]

I've never seen gallium-based TIM corrupt nickel before. It'll sink into copper pretty quickly, though. It doesn't actually react with aluminum, so much as act as a catalyst that permits oxygen free access to aluminum beneath the very thin layer of aluminum oxide that normally protects metallic aluminum from reacting rapidly with oxygen. As a result, your lurvely aluminum turns to aluminum oxide (alumina) very quickly, which is not what you want.

As far as Intel using it as a default TIM . . . I dunno, that would be iffy. A flux pad like Indigo Xtreme might work better for shipping. It's pretty solid at room temps.

 

[StinkyPinky]

ZEN 2 will close the gap. Should be quite interesting when it gets released - improved IPC, 7nm and 16 cores.

Just wonder whether it'll clock higher to challenge Intel at 5ghz and finally hit high memory overclocks (i.e. 4k mems).

16 cores? Has that been confirmed? 10-12 cores seems more likely? If they can can get 4.5 with improved IPC (say 5%?) with a 10 core CPU then it will be more than competitive. But that's a giant "if", whereas the 9900k is probably out in a few weeks (on paper at least)

 

[DrMrLordX]

16 cores? Has that been confirmed? 10-12 cores seems more likely? If they can can get 4.5 with improved IPC (say 5%?) with a 10 core CPU then it will be more than competitive. But that's a giant "if", whereas the 9900k is probably out in a few weeks (on paper at least)

The 9900k may not paper launch until October, with broad availability in November. One year after 8700k, in other words.

After that . . . good luck! But hey, at least it uses solder, right?

 

[Borealis7]

It's conductive, and a free running liquid, so potentially excess might shift from the die to other components and short them out.
It reacts with most other metals. Aluminum in the short term, copper in the medium term and even nickel plating in the long term.
It's just probably too hard to handle/apply correctly and hold up to shipping handling and long term usage.

also, it most probably costs more, and you have to replace it eventually, which is not possible if the hood is soldered, and you will lose heat conductivity over time.

 

[ozzy702]

ZEN 2 will close the gap. Should be quite interesting when it gets released - improved IPC, 7nm and 16 cores.

Just wonder whether it'll clock higher to challenge Intel at 5ghz and finally hit high memory overclocks (i.e. 4k mems).

Timeline on Zen2? I built my buddy a 1700x box a year ago and it's working great but I know he'll want an upgrade sooner or later. Something tells me not to expect as much as many on here are suggesting although I'd love to see the predictions come true. I'll be thrilled if max OC on Zen2 is 4.6ghz for 10 cores with a small IPC and memory latency bump (if possible). At that point assuming reasonable pricing, they'd have to sell a ton to new customers and to 1st gen Zen users.

 

[PeterScott]

Timeline on Zen2? I built my buddy a 1700x box a year ago and it's working great but I know he'll want an upgrade sooner or later. Something tells me not to expect as much as many on here are suggesting although I'd love to see the predictions come true. I'll be thrilled if max OC on Zen2 is 4.6ghz for 10 cores with a small IPC and memory latency bump (if possible). At that point assuming reasonable pricing, they'd have to sell a ton to new customers and to 1st gen Zen users.

Timeline is Epyc Rome early next year, which should reveal what they actual core will be, and I expect 7nm Ryzen desktop much later after volume production is going full blast. It might be mid next year.

Though I think people expecting a trifecta of IPC boost, clock speed boost and core count boost on the desktop might be setting themselves up for disappointment. Time will tell, and Epyc will likely provide clues.

 

[DrMrLordX]

Timeline is Epyc Rome early next year, which should reveal what they actual core will be, and I expect 7nm Ryzen desktop much later after volume production is going full blast. It might be mid next year.

Though I think people expecting a trifecta of IPC boost, clock speed boost and core count boost on the desktop might be setting themselves up for disappointment. Time will tell, and Epyc will likely provide clues.

Pretty sure Matisse hits in early Q2 2019? Should be April.