Intel Core i9-9900k 8c/16t, i7-9700K 8c/8t, i7-9600k 6c/6t 2nd Gen Coffee Lake CPU's + Z390

https://www.guru3d.com/news-story/intel-to-release-low-power-35w-core-i9-9990t-cpu.html

35W version of the 8 core spotted

"i9 9900T"
1.7ghz base
3.3 all core turbo
3.8 single core turbo

In the ballpark with the leak a while ago with 2.1ghz base clock 8 core "H" series - which being H (mobile) is assumed as cap 45W

I did some testing on my 9900K
stock 4.7 results in about 130W on 16 thread TSBench
95W power limit results in about 4.2ghz
65W power limit results in about 3.8ghz
from there I had to use speedshift/EPP as a low power limit would be ignored and it would sit on base clock of 3.6
EPP of 177 = ~3.3ghz and 50W consumption
EPP 229 = ~2.12ghz and 27W
EPP 238 is just under 1.7ghz and just under 21W

 

high temps and power sucking at high clocks aside, its quite crazy how efficient that chip becomes at lower clocks! were still talking about desktop cpu at 8 cores, 16 threads here, imagine that!

 

Intel's IGP-less i9-9900KF, i7-9700KF CPUs First Listings Appear Above MSRP Pricing
Raevenlord Monday, Monday, January 28th 2019 11:29 Discuss (11 Comments)
https://www.techpowerup.com/251950/...cpus-first-listings-appear-above-msrp-pricing

"... The i9-9900KF has been found online for $582.50, more than $50 above the i9-9900K's $530 street pricing, while the 9600KF is listed at $308.75, more than $60 over the street pricing of Intel's i7-9600K.

Remember that these prices are gouged on account of limited availability on the market; when these are more widespread and the market sees stocks in line with demand expectations, these should bottom down.

But then again, we've seen Intel's products being price-increased for a while as the company struggled to keep its production up to the demand, amidst a constrained and supposedly already second-line 14 nm process."

 

The 9900K has been available on Newegg for a while now i believe. Not sure why anyone would pay the premium for the KF unless it is just out of pure curiosity, a mistake or for a review. I am wondering if Intel will even release enough of them for the price to come down.

 

Yup, if the 'F' CPU's really are "floor sweepings" - failed iGPU / features turned off - re-cycling recovery sku's then there will be an even more limited supply than the full feature 'K' models.

If only a small percentage of dies fail in just the right way that they can be recovered as 'F' sku's, that's not going to be enough to fill the channel.

There might be some that think it's worth exploring the OC / power / thermal differences between the iGPU / feature disabled 'F' CPU's and the stock 'K' sku.

Intel says there is no difference, and for that I am assuming they mean there are no parameter / tuning differences in the firmware.

It's possible that with no iGPU "tax" there will be additional headroom for power / thermals, even cache, so it's probably something to look for soon, some reviews that explore the potential differences.

 

im very skeptical, but open to proven otherwise

 

https://www.reddit.com/r/thinkpad/comments/astmup/official_2019_refresh_specs_release_dates_and/

Intel 10nm Ice Lake laptops coming by June 2019 according to this technical details disclosure.

Expect 10nm desktop by year end 2019 or earlier. Competition is heating up fast.

 

Crappy 10nm low power mobile CPU's without iGPU's, maybe. No desktop 10nm CPU's until next year, and real high TDP 10nm laptop CPU's after that.

Intel is prioritizing Datacenter deliveries for 14nm and will continue the same for 10nm when it finally passes QA. Intel needs to deliver 10nm for the Datacenter first.

Intel is "out of the competition" until next year.

 

https://www.tomshardware.com/news/lenovo-laptop-intel-ice-lake-10nm,38674.html

June 2019 for Intel 10nm laptops. Gives them a full 6 months to get 10nm desktop parts ready for Holiday 19.

 

Those are low power CPU's, you know for the 2in1 laptops, and slim carry laptops, not the kind we like, not gaming laptops.

Intel is a total bust on powerful high performance 10nm Datacenter, Desktop, and Gaming laptop CPU's for 2019, and probably beyond.

7nm is the future Intel seems to be chasing now. 10nm is an all but dead, end.

 

I'll believe the 10nm desktop when I see it. Mobile parts are smaller dies, meaning that they are easier to get usable yields. Intel has pushed 10nm off for 3 years now. They even created a token dual core laptop CPU last year.

Since then, they had to gut the density of the chips and scrap specific implementations of scaling elements to create something that could have the necessary yields to produce 10nm. In fact, the 10 core Comet Lake is 14nm and planned for this summer to fall. The designs for HEDT and Server are also planned on 14nm until next year, it seems.

So I will wait to see Intel get 10nm on desktop as they have missed TOO many deadlines regarding 10nm to date to believe fully.

For 7nm, I find myself arguing the opposite. Earliest production chips for that node should be 2021, with a slide to 2022 being possible.

That is perfectly acceptable as their 7nm is closer to TSMC 3nm or Samsung 3nm/2nm nodes. Samsung will have that around 2021 and I believe TSMC is planning 3nm around 2022 or 2023, while being able to switch their lines over from 5nm to 3nm. Samsung will be there first, with Intel potentially second and TSMC third, which is a bit awkward for the largest fab in the world (unless Samsung purchases GF which went up for sale in the past couple weeks, which would be more for the IP portfolio).

In all honesty, I see Intel trying to get back onto the December mobile chips, January/February desktop chips, HEDT/Server Computex/early third quarter cadences. This is something that stopped being used as much after Kaby Lake, IIRC.

With that said, I am excited to see how they do with 10nm considering the first chip released last year kinda sucked, they fused off the iGP, and it ran slower on frequency than the prior 14nm gen it was to replace (although it had a higher transistor count, meaning, theoretically, a higher IPC which can make up for frequency to a degree).

So we are expecting mobile chips and cooper lake from Intel about the same time AMD drops desktop and server chips around mid-year. I'm happy they have finally gotten the yields issue under control, but with the changes they made to the process, there is a chance these chips may just be on par with 14nm+++++++++.

 

It was my understanding foveros won't be used on mainstream chips for the start of 10nm, instead being used for low power solutions related to atom type chips and possibly low power mobile. Do you have more information on it that I may have not seen as I haven't looked into it since CES and this looks like speculation of inclusion videos rather than Intel's current statements on when and what production lines will include it?

I even ate my words on it being impressive once they disclosed they are putting cache (likely L3 and potentially L4 or the L4 on the DRAM layer on the chip) on the layer below the CPU cores. Just want to know if there is something I missed, because prior statements said they wouldn't include it on all product lines at first.

Sent from my SM-G900P using Tapatalk

 

I think they are planning low power solutions first, but hopefully we see this in desktop eventually. It makes sense, they can mix and match process technologies, and use chiplets essentially to make future high CPU counts. Seems to be the most innovative thing out of Intel in a long time. The future seems pretty exciting.

 

It's my understanding the reason they aren't doing desktop and server with 3D stacking yet is the heat required for high frequency computing. The design shown has four low power cores and one high power core in a big little configuration. If they had all high powered cores, dealing with heat becomes more difficult.

Because of that, this isn't going to go against Zen 2, rather Zen 3 or beyond.

A good way to combat the heat is large IPC gains and lower frequencies, something that may need to wait for 7nm Intel or around there, as I'm unsure 10nm, with density being similar to AMD, that it could provide enough density and extra transistors to make it viable, even with the reduction in power from the smaller node.

With that said, Intel is working on chiplets, with the first major addition being two 24-core chips connected by a data fabric with the cascade AP Xeon chips, which require water cooling and are rumored to be bga. Over 5000 connection points on that chip!

After doing that, and with proof of concept on disintegrating the I/O, memory controllers, and cache, intel may look into a 2.5D bridge similar to AMD for 10nm.

But this is just my speculation.

Sent from my SM-G900P using Tapatalk

 

It wasn't just the yield problem holding up 10nm shipments, it was that Intel couldn't get a performance bump over current 14nm process on the current 10nm process. In fact that problem, the performance problem, was the first reason I heard for 10nm not being ready - the yield problem mention came afterwards.

That's why Intel's 7nm fab got such a funding push and even though we hear little about it publicly, it's been rumored to be the favored next jumping off point from 14nm. The 10nm process isn't thought to be the one to win between 7nm and 10nm in the Intel fab competition.

Also, the delay from 2015 is now 4 years here in 2019, and it will be 5 years if Intel delivers Datacenter / Desktop CPU's in 2020.

Here's a chart from Intel, no longer available on the Intel site, not sure of the year published by Intel, but I think it was 2010:


Here's an Intel presentation from 2014 that re-uses the chart with a few edits:

Transistor Scaling: The Age of Innovation
By Kaizad Mistry, Intel Corporation, Portland, OR
https://nanohub.org/resources/20880/watch?resid=20937
https://nanohub.org/resources/20935/download/2014.04.08-Mistry-NEEDS.pdf

Under the Supporting Documentation tab, Presentation Slides, the chart of interest is slide #49:
https://nanohub.org/resources/20935/download/2014.04.08-Mistry-NEEDS.pdf

 

You are correct, which is why 14nm++ was limited until 10nm wasn't ready. They then did coffee refresh. Looking at the press day in 2017 (March), they showed that 10nm was a side grade from 14nm++, or slightly worse transistor performance. But, you can increase number of transistors to increase IPC to compensate. But increasing transistors means it doesn't maximize how small the die is, bringing up yield concerns. Also, with the ambitions on cobalt over active gate, cobalt contract implementation, use of ruthenium, and going to a single dummy gate for scaling, the chips became more defect sensitive related to yields.

This was compounded by EUV not being available and relying on quad or more patterning.

Sent from my SM-G900P using Tapatalk

 

James with Rockit Cool sent me a email this morning. They are working on and testing a bracket for processors in order to connect directly to die without worrying about cracking your CPU die. They tested a 9700K yesterday. He said he was able to reach around the same temperatures at 5.3 ghz with direct die to waterblock as he was seeing at around 5.0 ghz previously. Not sure the bump in voltage he had to make but seems pretty impressive. They will be testing a 9900K next.

 

keep us updated, good info

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

I will be doing some testing of my own when I get ahold of one. Hopefully next week. My 9900k is currently delided with clu but I will do some before and after tests and take some pictures. Hopefully by next weekend.

 

Did you sand down the die as well?

 

No, Rockit Cool delided the processor for me. They just cleaned it up and used one of there copper IHS. Don’t think I would trust myself with sanding down the die.

 

OK. Was curious since reports are delidding the i9-9900K yields only slightly better temps (like 1 or 2C) unless sanding down the die's top 1/10 mm or so. But even 1C may help if you're on the edge.

 

Ya, I am not really sure as it was a new processor and did not do any testing before delid. It will be interesting to to see how much of a difference direct contact makes though.

 

according to SL delidding nets more like 4-8C on average, with sanding down the die only good for additional 1-2C.

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Hmmm. Der8auer reported just the opposite on a you tube post. Wonder if his sample size was smaller than Sillicon Lottery? (Either that Or I'm mis-remembering the video)

 

there were some differences though, i.e. Der8auer sanded the die much further down than SL did. iirc it wasnt economically feasible to go the extra distance on a larger scale, since ud have to worry more about scratching the PCB and traces when removing more of the die. so SL reported only 1-2C difference when sanding down to "safe" levels, whereas Der8auer went further and thus also had higher temp differences vs. stock height.

delidding and putting LM between die and IHS netted the same range for both SL and Roman.

btw, finally tuned in my everyday clocks ended up with 4.6 ghz all core with cache at 4.3 ghz. thats about all my four vrm phases would let me have at full load although benching is possible at higher clocks as well currently checking what kinda offset ill need for avx loads, then its off to tuning system agent, vccio and ram.

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Remove much more of silicon than needed can trow up other unwanted/unexpected problems like screw up the protection barriers... http://forum.notebookreview.com/threads/next-level-lapping.820439/#post-10754250

 

exactly, one of the main reasons why such cooling mods would only make sense for short term benching and not using the silicon long term. not worth those few degrees...

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

But remember Roman had little delta in temps by just delidding... In other words if you weren't going to sand, might as well stay with the solder as the delid only had a slight improvement in heat removal.

 

compared to previous gens the delta was indeed smaller, but far from insignificant. 4-8C could make up for 1-2 multis, depending on chip and voltage.

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

If I remember correctly SL said you will gain 0.5 multi. But I think they will be on the safe side regarding fully stability.

 

In my case I imagine i will gain 3-4 C as i am currently running Gelid GC Extreme between the heat spreader and the EK Waterblock with CLU between the die and the heat spreader and will apply CLU between the die and the EK Water block. I stopped by Rockit Cool yesterday and they are still working on the bracket but expect it to be done soon.

Edit: If anyone is interested in doing a formal review you could probably get one from James for free i would imagine. I will not be doing anything other than Aida64 before and after temps at 5.0ghz with the same bios settings and then try to overlock higher at the same voltages or a minor bump.

 

Intel Expected to Regain Top Chipseller Ranking

https://www.eetimes.com/document.asp?doc_id=1334414

 

Yeah, ram wasn't the issue, RAM with high GPU cost and high CPU cost due to an Intel shortage caused preassembled to be proceed uncomfortably or out of stock.

Intel has already been projected to alleviate their CPU shortage by middle of this year by industry insiders. But I do not expect regain of market share, rather holding strong where they are until second quarter. We have to remember, Intel had 99% of the server market and a crushing lead in the other segments. Even with AMD being mid-single digits, roughly, in the server market, yet gaining more ground in mainstream, they always has market dominance.

As to recovering selling more chips (referring to mind factory's numbers), cheaper ram does not necessarily make expensive CPUs more palatable, while CPUs for the same price but without the iGP doesn't seem to be as compelling at the same clocks.

I would argue that if the prices are correct on Zen 2 products, Intel will not just have to fulfill volume (talking ending the shortage), rather Intel will have to engage in dropping prices. But if they do this reactively rather than proactively, it won't be as positive as people expect.

This isn't to say I am not wrong here. Just an alternative analysis. Considering most computers are laptops now, and Intel is releasing 10nm mobile this summer, it may very well be true overall. But that is my two cents.

Sent from my SM-G900P using Tapatalk

 

WTH does that have to do with CPU sales? Nothing.

Samsung's largest Chip sales by % is Nand sales, and with the decline of Nand price the dollar sales amount will decrease, allowing Intel's unchanged sales to rise above again to #1.

This means nothing good or bad for Intel, it's a Samsung story at best, mostly it's a good news story for Nand prices.

"Thanks to a steep plunge in the memory market that is expected to drag Samsung's sales down by 20% compared to last year, Intel will once again assume the mantle as the leading chipseller in 2019, IC Insights projects."
...
"The memory market is expected to decline by about 24% this year compared to last year. Roughly 83% of Samsung's chip revenue in 2018 was derived from memory chip sales. With the market downturn, IC Insights projects that Samsung's chip sales will slip to about $63.1 billion, down from $78.5 billion last year.

Intel, meanwhile, is projected to have chip sales of about $70.6 billion this year, up just slightly from $69.9 billion last year.

In 2017, amid a prolonged boom in memory chip prices and sales, Samsung passed Intel to take the No. 1 spot in semiconductor revenue. Prior to that, Intel had been the leading chip seller from 1993 to 2016."

Intel still hasn't delivered Datacenter / Desktop 10nm parts to production, and is still years behind in 7nm deployment, it's unclear whether Intel will be able to deliver any meaningful production on 10nm this year at all, and it might be the end of next year before 10nm Datacenter and desktop CPU's arrive - if at all.

Intel is going to eventually hit that cross over point where 14nm stops selling, and there is no 10nm / 7nm production to take over.

That's the real news to wake up to, not some sales ranking based on Nand price fluctuations.

 

And you expect that Intel will sit down look at AMD eat the whole cake for himself?

 

I don't know Intel will have much choice for a year or so. 2021 or so with 7nm is a different story. They are furthest along with implementing cobalt contacts of any company. As to EUV, others are further along, but that is also because Intel waited too long on other companies getting it working (the hardware used to create the chips, which Intel does not design itself).

Right now, rumors are 14nm comet lake will come end of this year or into next year and have 10 cores. As to IPC, many believe ice lake will keep them in the lead there, but 10nm itself has clocking issues, as it is a couple hundred MHz lower than 14nm. 10nm+ should resolve that.

I'm wondering if comet lake is adapting ice lake to 14nm++ for high performance and doing others on 10nm+ ice lake/sunny cover next year.

Either way, competition is back. Industry insiders have speculated AMD will take the multi core crown for the year, while Intel may keep the single core performance crown. But that is just rumors and speculation.

The primary question is if a price war will be started this summer (which rumored AMD prices suggest it will). If it does, all consumers win, regardless of whose chips you buy! That is an exciting prospect!

Sent from my SM-G900P using Tapatalk

 

Yes, absolutely that is what I am saying. Good interpretation.

Intel can continue selling low yield high cost 14nm CPU's like it has been, but even that narrow unprofitable path is petering out due to yield, power, and thermal limitations reducing shipments and missing deliveries.

Intel has screwed up so badly by painting themselves into the smallest of 10nm corners without a viable exit strategy.

Intel believes that 7nm is Intel's 10nm exit strategy, but that's most likely false hope continuation of the false hopes from Intel's 10nm failures to deliver.

Hopefully AMD doesn't screw up too badly along the way, and as long as AMD simply continues to deliver higher core counts, higher clocks, and finer automatic tuning for the best performance, AMD is going to be able to make, sell, deliver - and "eat" all the cake.

Mmmmm.

 

HaHa

 

https://wccftech.com/intel-comet-la...st-will-also-have-8-core-and-6-core-variants/

10 core Comet Lake Incoming 2H 2019. 14nm++++++++++++ ain't dead yet.

 

The real clowns are Intel in their failure to matriculate to 10nm and showing their "Big Red Noses" (Beep! Beep!) while pumping higher and high power and heat through 14nm dies that are getting larger and larger - with smaller and smaller yields and market relevance.

Seriously, Intel are sprouting some pretty Ludicrous Clown Shoes

 

I will very likely skip this one. I will wait for 10nm next year. 8 core 16 threads is more than enough for my needs. We need to see improvements in other areas to speed up calculations, throwing more cores at it won't fix this. The core war while good to an extent will just get silly if we don't improve in other areas.

 

Then you are ignoring other changes AMD has some to the microarchitecture. There are substantial changes, like using UMA instead of NUMA, increased cache size and changed pre-fetch, reduced latency on IF2, etc. So it isn't just adding cores, IPC is rumored at 11-15%, which means it should be better than Intel's.

Now, next year, Intel will also change their pipeline and may increase IPC by about 11%, but that means just playing leap frog. Also, that is going against Zen3, which may be more of a change than zen to zen+. Also, there is going to be a large floating point increase for AMD.

So, it's always best to wait to see actual performance for one's needs.

Did you see Intel's pricing rumored for the new cascade lake server chips? Nearly $20K.

Sent from my SM-G900P using Tapatalk

 

If AMD releases a CPU that blows Intel out of the water in terms of gaming then no doubt I will give them a chance. Until then, I will wait for 10nm Intel. If that doesn't bring substantial changes, I will again wait. This 8/16 CPU at 5.0Ghz will hold me over for some time I think with a 4K 144hz panel. GPU will be my bottleneck for some time I think. I have my eyes on AMD for next gen, but they have a decent gap to close in terms of gaming performance with Zen 2. They will do it I think, but will they exceed Intel by any meaningful margin to make me swap over? I don't think they will.

 

Now that I can agree with. Even if the sample shown at CES was the 65W variant of the CPU (which leaks suggest it was), that was going against a stock 9900K, not an overclocked one. But that begs the question what the full 95W chip can do, whether the average 11% performance deficit was closed (it was closer to a 22% deficit average in productivity workloads, IIRC) with the 65W versus the stock 9900K, and if the CPUs will meaningfully overclock.

Either way, if you already own a 9900K, I don't know if a 30%+ increase would be possible over it, even if the full 95W does outperform the 9900K when overclocked for both CPUs.

If that is the case, then I'd say wait, not just for 10nm Intel, though, but also to see changes implemented for Zen 3. If the cost of active interposers is low enough by then (looking at 22nm-32nm active interposer costs because some companies doing integration don't do 2.5D or 3D on sizes larger than that, although GF is for sale and who knows what the purchaser will do with their line, if AMD will move the I/O die to TSMC and what integration they offer, etc.), then we could see a significant jump in performance (for TSMC 7nm EUV, you get like 35% area reduction, 10 or 15% power reduction, and no idea on performance impact, whereas 5nm EUV over 7nm is estimated at around 15% performance improvement). So the active interposer may come with Zen 3 or the Zen after that.

But, if wanting significantly more, I think Intel's average IPC increase per gen is around 11%. So there is a lot to watch for....

Sent from my SM-G900P using Tapatalk

 

First Photo is with Delid (CLU between die and Copper IHS). Gelid GC Extreme between Copper IHS and EKWB Waterblock.



Second Photo is Thermal Grizzly Conductonaut as i did not have any CLU with a direct connection between the die and the EKWB water block. The maximum CP package was only 1 C different but you can see the average was a bit lower (around 4C for the package).




In order to use the bracket made by Rockit Cool i had to take off the plastic washers on the EKWB (Photo of washers below).



Bracket installed with some normal thermal paste to make sure it is making contact.



Bracket is making contact!! I know, too much and did not cover the hole die, but i did not care much as i just wanted to make sure it was making contact.



Cleaned up and installed Grizzly Liquid Metal.



You all can be the judge of the thermals. The voltages may appear to be a little different, but all settings in the bios were the same.

 

Not a lot of difference between the two then. I know you mentioned there was a bigger difference between the averages, but you'd have to be careful in controlling your experiment if you're gonna refer to the average as a reliable comparison. The benchmark run would have to be same, and the point at which you press the "Reset" button of the HWInfo Sensors would have to be the same, as well as the point at which you do the final screenshot. First off I notice that the lowest package power in Watts is quite different between the two runs - with 145W of the 'colder one' vs 161W of the 'hotter one', so that leads me to believe you might have pressed the "Reset" button of the HWInfo Sensors at a different point during the testing, all of which can skew the averages. Your experiment might need to be more tightly controlled to ascertain if there is a true difference in cooling performance here, because the temperature margins are quite small.

 

Understand, not used to running comparisons in a controlled environment. It seems the difference is negligible; however, once the bracket is installed it will be faster to clean up and re-paste then dealing with the Copper IHS and cleanup underneath it as well. I like the idea of the bracket. I know he sent the bracket off to some reviewers so maybe they will do a more controlled review.

 

That's good then. I also forgot to mention that you'd have to control your room temperature, or at least have a temperature measurement of it during the two runs, and you'd either have to do the comparison benchmark run on a completely cold PC from startup, or you would have to preheat your loop until it had reached steady state at which point you would then do your benchmark run - probably the latter is better, preheated loop. I'd be surprised if your mod has made the cooling worse though, it's probably a little better if anything - although thinking about it it's probably more reliant on your heatsink being completely flat if you don't use the IHS, so that's something to consider. The IHS probably would help get better temperatures if your water block is not completely flat - because more surface area with IHS to spread the heat out to a wider area of the water block if there are some imperfections in the flatness of your water block. (but then that would also depend on how flat the IHS is on the top surface though too - lots of variables - I suppose that's why some people lap their IHS to make it flatter.) (I've seen reviewers measure flatness by placing the edge of a metal ruler on top of the surface to see if there is any light creeping through any gaps where it's not flat).