Intel Core i9-9900K 3DMark Numbers Emerge: Beats Ryzen 7 2700X

[GlacierNine]

Soldered IHS -> better heat conduction thus lowered temps and lowered power(yes running cooler generates less heat). Possible more fine tuned manufacturing process and thus possibility of lower core voltages -> lowered power. What I'm trying to say there are more variables than just tdp and core count.

The solder, a larger die area, a better IHS, etc, will all have an effect, but not the one you're describing.

If you have a 130W CPU under a heatsink that can only dissipate 95W maximum, it doesn't matter whether it's soldered, it doesn't matter how big the die is, it doesn't matter if the IHS is made of a super-conducting alloy with a thermal conductivity of 5000W/m Kelvin. The raw amount of energy is higher than the heatsink can dissipate into the surrounding air, and that heatsink will not be able to keep that CPU cool.

The particular rule I am describing is cast iron. More heat *must* equal more capacity to dissipate that heat, which means larger radiators, larger fin stacks, and faster fans. sure, you could introduce a bottleneck before that point, by using toothpaste TIM, or a wooden heatspreader, but ultimately, getting rid of any bottleneck doesn't allow more dissipation. It allows the full utilisation of the amount of dissipation you had all along. If the heatsink is too small, or the amount of heat is too large, you can't fix that with solder or a better IHS or a larger die area.

The 6700K had a 4GHz base clock. The 7700K was notable for the fact even in stock trim, it's temps would routinely spike to TJMax, and it's base clock was 4.2GHz. With the next gen, we lost 500MHz from base clocks, as the 8700K has a 3.7GHz base clock despite being made on a process 2 generations more optimised than the 6700K. The 9900K is cited to have either a 3.6GHz or 3.1GHz base clock.

They're quite clearly reducing the base clocks and basing the TDP off of those figures, in order to avoid admitting that at full boost, these chips are getting hotter and hotter as they add more cores and push the boost higher.

Running cooler does not generate less heat, not on a CPU.

Actually it kinda does. Keeping your chip cooler means there's less voltage leakage, which means you can reduce your voltage while maintaining stability, which means you use less power, which means the chip produces less heat to begin with.

This all relies on manual tweaking and undervolting though. If your chip is running at stock settings then it will simply run on a volt/frequency curve in the BIOS, and temperature won't be accounted for at all.

 

[R0H1T]

The solder, a larger die area, a better IHS, etc, will all have an effect, but not the one you're describing.

If you have a 130W CPU under a heatsink that can only dissipate 95W maximum, it doesn't matter whether it's soldered, it doesn't matter how big the die is, it doesn't matter if the IHS is made of a super-conducting alloy with a thermal conductivity of 5000W/m Kelvin. The raw amount of energy is higher than the heatsink can dissipate into the surrounding air, and that heatsink will not be able to keep that CPU cool.

The particular rule I am describing is cast iron. More heat *must* equal more capacity to dissipate that heat, which means larger radiators, larger fin stacks, and faster fans. sure, you could introduce a bottleneck before that point, by using toothpaste TIM, or a wooden heatspreader, but ultimately, getting rid of any bottleneck doesn't allow more dissipation. It allows the full utilisation of the amount of dissipation you had all along. If the heatsink is too small, or the amount of heat is too large, you can't fix that with solder or a better IHS or a larger die area.

The 6700K had a 4GHz base clock. The 7700K was notable for the fact even in stock trim, it's temps would routinely spike to TJMax, and it's base clock was 4.2GHz. With the next gen, we lost 500MHz from base clocks, as the 8700K has a 3.7GHz base clock despite being made on a process 2 generations more optimised than the 6700K. The 9900K is cited to have either a 3.6GHz or 3.1GHz base clock.

They're quite clearly reducing the base clocks and basing the TDP off of those figures, in order to avoid admitting that at full boost, these chips are getting hotter and hotter as they add more cores and push the boost higher.

I don't think that was ever in doubt, what most people are wondering is whether this thing will do 5GHz (all cores) easily when OCed. Also what will be the power draw, especially when OCed.

 

[londiste]

Actually it kinda does. Keeping your chip cooler means there's less voltage leakage, which means you can reduce your voltage while maintaining stability, which means you use less power, which means the chip produces less heat to begin with.

Whoa, for some reason I still had the impression that leakage was not that significant. Turns out it is rather huge problem by now. I wish more papers would be public but at least some are.
https://www.cin.ufpe.br/~etgs/getPDF3.pdf
http://www.mpedram.com/Papers/IEICE-leakage-review-journal.pdf

Looks like some people have done practical attempts at measuring the temperature dependence:
https://forums.anandtech.com/thread...-power-consumption-with-the-i7-2600k.2200205/

 

[bug]

I don't think that was ever in doubt, what most people are wondering is whether this thing will do 5GHz (all cores) easily when OCed. Also what will be the power draw, especially when OCed.

Yeah, because we have everything else about this figured out
But yes, people will do that.

 

[GlacierNine]

I don't think that was ever in doubt, what most people are wondering is whether this thing will do 5GHz (all cores) easily when OCed. Also what will be the power draw, especially when OCed.

I expect the majority, if not almost all, of the chips to be capable of 5GHz all core, as long as the temperatures can be kept in check. Intel have been on this process for so long now that yield will be very, very good, there simply won't be a lot of failed transistors on these dies to threaten the stability of the cores at that frequency.

The question is as you say, the power draw, and I'll be frank - I expect these chips to be the kind of toasters that you will *need* high end air cooling to tame, if you want those 5GHz all core boosts, or any overclocking headroom at all. A Hyper 212 will not suffice to overclock these chips - at all.

In fact, I will go out on a limb and suggest that if the rumours are correct, and this is the return of solder to the Mainstream desktop, it's not because intel want the chips to run cooler and satisfy enthusiasts demands. It's because the toothpaste introduces too much variance because of the Z-height between die and IHS, and soldering is the only way they can ensure all of the chips they sell will stay within thermal limits when turbo boost is in effect. If they tried to toothpaste these, they'd end up with chips hitting TJMax even under excellent coolers.

That's my story and I'm sticking to it.

 

[R0H1T]

Yeah, because we have everything else about this figured out
But yes, people will do that.

What there's to figure out? This is the same SKL core, 8 cores, with extra L3 cache. If it's soldered then you can expect 5~10 degrees better temps than a regular 8700k, probably more when OCed but that also depends on the max frequency this chip can realistically achieve.

It will beat 8700k everywhere except the most ST limited scenarios, where an OCed 8700k can win. In games it should be pretty close if not outright beating the 8700k in certain titles.

If we're talking exact numbers then sorry, you'll have to wait.

I expect the majority, if not almost all, of the chips to be capable of 5GHz all core, as long as the temperatures can be kept in check. Intel have been on this process for so long now that yield will be very, very good, there simply won't be a lot of failed transistors on these dies to threaten the stability of the cores at that frequency.

The question is as you say, the power draw, and I'll be frank - I expect these chips to be the kind of toasters that you will *need* high end air cooling to tame, if you want those 5GHz all core boosts, or any overclocking headroom at all. A Hyper 212 will not suffice to overclock these chips - at all.

In fact, I will go out on a limb and suggest that if the rumours are correct, and this is the return of solder to the Mainstream desktop, it's not because intel want the chips to run cooler and satisfy enthusiasts demands. It's because the toothpaste introduces too much variance because of the Z-height between die and IHS, and soldering is the only way they can ensure all of the chips they sell will stay within thermal limits when turbo boost is in effect. If they tried to toothpaste these, they'd end up with chips hitting TJMax even under excellent coolers.

That's my story and I'm sticking to it.

Interesting, we'll see if this theory about Z height is correct. Just need someone to destroy delid their sample

 

[GlacierNine]

What there's to figure out? This is the same SKL core, 8 cores, with extra L3 cache. If it's soldered then you can expect 5~10 degrees better temps than a regular 8700k, probably more when OCed but that also depends on the max frequency this chip can realistically achieve.

It will beat 8700k everywhere except the most ST limited scenarios, where an OCed 8700k can win. In games it should be pretty close if not outright beating the 8700k in certain titles.

If we're talking exact numbers then sorry, you'll have to wait.

Interesting, we'll see if this theory about Z height is correct. Just need someone to destroy delid their sample

I don't think we need to test that one. We already know that removing the adhesive from the IHS and PCB during a delid, is one of the most important factors in obtaining a temperature reduction, and we also know that some chips experience much greater reductions than others during delids with the same thermal compounds being used, depending on whether the user removed the adhesive from the IHS. This has been confirmed by everyone from der8auer to Steve at Gamers Nexus, and Kyle at HardOCP to boot.

Hell, even Linus got that right, even though he delidded 3 CPUs on his channel before he thought to use liquid metal on one.

We know beyond reasonable doubt that the z-height variance is there. But, use solder instead of thermal toothpaste, and that variance becomes much less important from chip to chip, which in this case, could allow Intel to sell chips that are binned extremely close to the edge of the envelope.

 

[R0H1T]

I don't think we need to test that one. We already know that removing the adhesive from the IHS and PCB during a delid, is one of the most important factors in obtaining a temperature reduction, and we also know that some chips experience much greater reductions than others during delids with the same thermal compounds being used, depending on whether the user removed the adhesive from the IHS. This has been confirmed by everyone from der8auer to Steve at Gamers Nexus, and Kyle at HardOCP to boot.

Hell, even Linus got that right, even though he delidded 3 CPUs on his channel before he thought to use liquid metal on one.

I know that, but the theory also hinges on the supposition that smaller Intel chips cannot be soldered, for ~ reasons. What could be more interesting is if Intel improved the 14nm(?) process along with packaging & assembly for the rumored i9, if true then the current owners should feel even more aggrieved.

 

[John Naylor]

1. "Beats", like core counts, doesn't mean much ... in gaming the 2700x was slower than two generations if Intel chips before it. Benchmarks ? How many folks spend their PC time running benchmarks ? Until I see application scores, I'm just not interested.

2. TDP has nothing to do with how much power the CPU draws..... it related to the necessary heat the Heat Sink is required to pull away from the CPU to keep it running properly. As the power draw is not static, as long as the heat sink can absorb an average heat load = to the TDP, the CPU will perform as designed. If not discussing TDP, then don't call it TDP.

3. The drop to a base frequency of 3.1 ,makes sense of a heat control mechanism. It's a logical response to the media's focus on core counts as well as a significant % of consumers who mistakenly assume that extra cores are going to do something for them.

4. A Hyper 212 is suitable for moderately overclocking any CPU; expect 10C improvement going form a $25 cooler to a $40 cooler (Mugen Max / Fuma).

5. Delidding was a thing on OIvy Bridge because it actually resulted in OC improvements ... since than, delidding has been done simply for deliddings sake based upon the assumption that "cooler is better". I don't see the point ... have not had an OC thermal limited since IB, om every OC since, have hit a voltage wall before the thermal wall.

 

[HD64G]

Let's wait for the final clocks and price and then we will judge if its good value against the 2700X.

 

[TheOne]

I'm more interested in pricing and seeing how the 9700K compares to the 8700K.

 

[bug]

@John Naylor TDP has everything to do with the power drawn. Because power drawn is the only thing that generates heat.
But as you have noted, TDP is more of an average indicator, mostly used to size the heatsink, whereas in forum discussions power draw most of time refers to peak, even instantaneous usage.

 

[efikkan]

People should always factor in a margin when selecting a cooling solution. Not only can the TDP be a little off, the computer will get full of dust and the condition of the air also have some impact. But still, the best CPU cooler in the world wouldn't matter if the case doesn't have good airflow.

Even pre-built "workstations" from OEMs like Dell and HP don't get cooling right. Just push an Optiplex a little, and it will sound like a "jet engine", with that poor stock Intel fan doing all the work, and that's on CPUs like i7-6700 (65 W).

The TDPs seem to gradually become more dishonest, but Ryzen 2 took this to a new level, where XFR 2.0 boosts extra if you have very powerful cooling on an open rig. And as I said in comments of the Ryzen 7 2700X review, Intel will probably do the same.

 

[jabbadap]

Whoa, for some reason I still had the impression that leakage was not that significant. Turns out it is rather huge problem by now. I wish more papers would be public but at least some are.
https://www.cin.ufpe.br/~etgs/getPDF3.pdf
http://www.mpedram.com/Papers/IEICE-leakage-review-journal.pdf

Looks like some people have done practical attempts at measuring the temperature dependence:
https://forums.anandtech.com/thread...-power-consumption-with-the-i7-2600k.2200205/

Intel uses thermal profiles, this is what is on their datasheet for 95W tdp processor:

So 95W tdp processor will run at 0.22°C/W*95W+43.7°C = 64.7 °C with that 130W cooler. Now if you change processor with soldered IHS that Thermal profile can't be correct anymore, every one knows that using liquid metal lowers the temps not to mention solder. But if you add more cores, solder IHS on and up the freqs to meet that thermal profile again. Intel already uses that thermal profile for four core 91W(@4.2GHz) and six core 95W TDP(@3.7GHz) processors.

I'm more interested in pricing and seeing how the 9700K compares to the 8700K.

Yeah that will be interesting to see eight core processor vs six core sixteen threads one.

 

[GlacierNine]

The processor TDP is the maximum sustained power that should be used for design of
the processor thermal solution. TDP is a power dissipation and component temperature
operating condition limit, specified in this document, that is validated during
manufacturing for the base configuration when executing a near worst case
commercially available workload without AVX as specified by Intel for the SKU
segment. TDP may be exceeded for short periods of time or if running a very high
power workload.

The TDP and Configurable TDP values are the average power dissipation in junction temperature
operating condition limit, for the SKU Segment and
Configuration, for which the processor is validated
during manufacturing when executing an associated Intel-specified high-complexity workload at the
processor IA core frequency corresponding to the configuration and SKU

Segment and Package - S-Processor Line LGA
Processor IA Cores, Graphics Configuration and TDP - 6-Core GT2 95W
Configuration - Base
Processor IA Core Frequency - 3.6 to 3.7GHz
Graphics Core Frequency - 1.15 GHz to 1.2GHz
Thermal Design Power (TDP) (w) - 95

6. Thermal Design Power (TDP) should be used for processor thermal solution design targets. TDP is not the maximum power that the processor can dissipate. TDP is measured at DTS = -1.TDP is achieved with the Memory configured for DDR4 2400/2666 2 DIMMs per channel.

Using these quotes, from that same datasheet, we now know that Intel is quoting TDP at base clocks only. The same base clocks that have gone DOWN every release since Kaby Lake, to the point that if you believe the "3.1GHz" 9000-series leak that just happened, we're now getting 95W TDP on 8 cores at 3.1GHz, rather than 95W TDP on 4 cores at 4.2GHz.

Difference between 7700K Base and Max Turbo Frequency - 300MHz. (And this was a processor known to spike to nearly TJMax without delids)
Difference between 9900K Base and Max Turbo frequency - either 1.4GHz or 1.9GHz depending on which leak you believe.

This, while only hitting the same thermal envelope at BASE clocks. Clearly, then, Intel have not discovered a magic potion for power efficiency or thermal design - they have simply reduced the base clocks and hidden the true TDP behind per-core Turbo specs that they don't share TDP for.

 

[DeathtoGnomes]

So, watercooled again?

TBH, I dont truest any numbers Intel PR puts out, there just reliable proof the numbers are actual. Intel has pulled a fast one on us all before, why would this be any different?