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

[londiste]

The leaked details sound very suspicious to me. 3.6 GHz base clock and 4.7 GHz all-core boost? 8700K has 3.7 GHz base on 6 cores. If 9900K can do 3.6 GHz on 8 cores within the same 95W power envelope as well as boost all cores to 4.7 GHz out-of-box, that last + in 14++ process would really be deserved with the efficiency boost.

 

[bug]

Some "95W parts" consume twice as much as others. TDP as we know it is long... long gone. Today its used to describe power consumption for base frequency, but even that is incorrect.
At his maximum potential, said 9900K would be able to cross the 200W line with ease.

It's true that the CPU is now allowed to draw more power than its specified TDP. However, it can only do so until the temps rise high enough. Which is probably 10 seconds or so (highly dependent on cooling). The average TDP is the one on the box though.
And yes, I am aware that at times, both AMD and Intel took pride in their rated TDPs representing the max power draw as opposed to the competitions average. It seems those days a re behind us.

 

[R0H1T]

It's true that the CPU is now allowed to draw more power than its specified TDP. However, it can only do so until the temps rise high enough. Which is probably 10 seconds or so (highly dependent on cooling). The average TDP is the one on the box though.
And yes, I am aware that at times, both AMD and Intel took pride in their rated TDPs representing the max power draw as opposed to the competitions average. It seems those days a re behind us.

That's not true at all, even at stock the 8700k can pull more than 95W easily in heavy workloads, like AVX or AVX2 usage for certain applications.
There's no such thing as avg TDP, what you're talking about is probably avg power draw - which may change massively even with a change in boards, as stock voltages can vary.
TDP's generally an indication for cooling the said processor.

Every spec sheet Intel puts out?

None of them are official, yet.

 

[jabbadap]

It's true that the CPU is now allowed to draw more power than its specified TDP. However, it can only do so until the temps rise high enough. Which is probably 10 seconds or so (highly dependent on cooling). The average TDP is the one on the box though.
And yes, I am aware that at times, both AMD and Intel took pride in their rated TDPs representing the max power draw as opposed to the competitions average. It seems those days a re behind us.

Intel defines TDP as this:

TDP
Thermal Design Power (TDP) represents the average power, in watts, the processor dissipates when operating at Base Frequency with all cores active under an Intel-defined, high-complexity workload. Refer to Datasheet for thermal solution requirements.

That high complexity workload might have AVX load, at least anandtech said that on one of their review(Although I'm not 100% sure was it known fact or just guess from reviewer according to load test results). And yeah TDP != power draw.

 

[ZoneDymo]

My next build will still be AMD though

 

[londiste]

That high complexity workload might have AVX load, at least anandtech said that on one of their review(Although I'm not 100% sure was it known fact or just guess from reviewer according to load test results). And yeah TDP != power draw.

Provided that thermals are OK, Intel CPUs will run full load including AVX at base clock with power consumption at or below TDP.

Even 8700K will do that easily. Actually, while the base clock is 3.7 GHz, 8700K will run Prime95 Small FFTs (the worst case scenario for Intel CPUs) at 4.0/4.1 GHz within TDP. This matches well with the usual expected all-core boost of 4.3 GHz (default AVX offset being -2).

While Intel does a lot of things, they have very nice specs and technical papers and their products tend to match all that to the letter (although often enough after you manage to decode that "to the letter" part).

 

[jabbadap]

Provided that thermals are OK, Intel CPUs will run full load including AVX at base clock with power consumption at or below TDP.

Even 8700K will do that easily. Actually, while the base clock is 3.7 GHz, 8700K will run Prime95 Small FFTs (the worst case scenario for Intel CPUs) at 4.0/4.1 GHz within TDP. This matches well with the usual expected all-core boost of 4.3 GHz (default AVX offset being -2).

While Intel does a lot of things, they have very nice specs and technical papers and their products tend to match all that to the letter (although often enough after you manage to decode that "to the letter" part).

Well yeah clocks are guaranteed with sufficient cooling(130W cooler by intel specs). Then there's is of course out of intel's hand limiting factor motherboard manufacturer and their vrms and bios settings.

 

[GlacierNine]

The leaked details sound very suspicious to me. 3.6 GHz base clock and 4.7 GHz all-core boost? 8700K has 3.7 GHz base on 6 cores. If 9900K can do 3.6 GHz on 8 cores within the same 95W power envelope as well as boost all cores to 4.7 GHz out-of-box, that last + in 14++ process would really be deserved with the efficiency boost.

I agree. Adding 33% more cores at essentially the same base frequency, without changing manufacturing process?

There's just no way that results in the same TDP and power envelope as the previous generation, even at the base clocks. When you add in Turbo Boost, you're already way over the 95W TDP for these chips. You were also over it for the 8700K compared to the 7700K, but this is a step up again, in terms of real-world power use.

 

[londiste]

I agree. Adding 33% more cores at essentially the same base frequency, without changing manufacturing process?

They did change the manufacturing process. These should be on the new improved 14+++ process.

 

[jabbadap]

They did change the manufacturing process. These should be on the new improved 14++ process.

Well increased die size also helps cooling. Thus core count to GHz is not necessary linear equation.

 

[R0H1T]

They did change the manufacturing process. These should be on the new improved 14+++ process.

Speculation atm, remember the devil's canyon with same TIM but improved thermals wrt 4770k

 

[Durvelle27]

Yawn

 

[RejZoR]

That's hardly an achievement. I mean, R7 2700X was released back in April 2018. That's 3 months ago and it's not even a flagship from the Zen+ lineup (which is why AMD reserved the R7 2800X imo). It would be a bit shameful for Intel to release a slower product 3 months later lol.

But hey, competition is good. We missed this kind of nudging for quite few years between AMD and Intel.

 

[GlacierNine]

Well increased die size also helps cooling. Thus core count to GHz is not necessary linear equation.

You're right, it does, but increased heat output over a wider area is still increased heat output. The amount of thermal energy output by this chip is still higher. It's just no longer such a challenge to move it around. Ultimately your limiting factor is still the capacity of the heatsink to dissipate the heat once it's been conducted away from the die.

Let's say we have a true 100W TDP part with a 100mm square die area, no IHS and we put a cooler onto it capable of dissipating 200W.
If we increased the die size while keeping TDP the same, the temperature might go down a bit, because the heatsink could become more efficient, spreading the heat over a wider area of the fins, meaning less hotspots.

But if we increased that TDP to 200W, and compared the temperatures of the two, with the same heatsink, the temperatures would be the same, because once you hit the thermal capacity of the heatsink, it no longer matters how quickly or evenly the heat can enter the heatsink to be dissipated - the heat can only leave that heatsink so quickly once it's there.

The moment our example die hits 201watts of heat output, we enter a situation where no amount of die area would enable us to cool the chip, because the heatsink is simply unable to dissipate that much. It would be in thermal runaway, slowly getting hotter and hotter until eventually the chip hit TJMax and shut off.

So sure, die size helps cooling - but more heat is still more heat, and you still need a bigger heatsink to be able to dissipate more heat.

 

[efikkan]

I agree. Adding 33% more cores at essentially the same base frequency, without changing manufacturing process?

It's going to consume a bit more, but not 33% more, they did not increase the die by 33%. Dies are more than just cores.

 

[GlacierNine]

It's going to consume a bit more, but not 33% more, they did not increase the die by 33%. Dies are more than just cores.

Which is why my post says "adding 33% more cores" and not "adding 33% more die area". Not that die area is the be all end all of this - as I explained above.

The top SKU also has 4mb more L3 Cache than the 8700K, 16 over 12MB, so, again, more heat generation from that.

The heat increase here is not going to be 33% but it is still going to be significant.

 

[bug]

Which is why my post says "adding 33% more cores" and not "adding 33% more die area". Not that die area is the be all end all of this - as I explained above.

The top SKU also has 4mb more L3 Cache than the 8700K, 16 over 12MB, so, again, more heat generation from that.

The heat increase here is not going to be 33% but it is still going to be significant.

There you go: https://en.wikichip.org/wiki/intel/core_i7/i7-8700k
Cores+L3 cache ~50% of the die area.

 

[londiste]

Going from 4 cores (Kaby Lake, 7700K) to 6 cores (Coffee Lake, 8700K) the only change was +2 cores, basically the same uncore and same GT2 iGPU. From 126 mm² to 149 mm². 2 more cores should put the die size to the neighborhood of 175 mm². 17% increase, roughly.

That actually matches to what AMD was saying about 4 cores of both manufacturers being 40-something mm².

 

[GlacierNine]

There you go: https://en.wikichip.org/wiki/intel/core_i7/i7-8700k
Cores+L3 cache ~50% of the die area.

Ah, good, I actually went looking for a die shot of the 8700k but didn't leave google, so only found shots of the older HEDT 6 cores.

So with that, we add 2 cores. We now have 33% more of the parts that comprise 50% of the die area, meaning that on the same manufacturing process the die is now ~16% larger and all of that 16% produces heat.

As I've been saying this whole time - that is not a trivial increase. From whence would anyone get the idea, knowing this, that the TDP could possibly stay at 95W?

And that's even assuming it was actually 95W after the move from Kaby Lake to Coffee Lake, which I personally doubt very much, as wasn't Intel's rationale for Z370 not being backwards compatible, that Z370 and hex cores, needed more power, that entry level Z170 and Z270 boards weren't necessarily able to provide with their VRMs, that were designed for quad core?

 

[efikkan]

Which is why my post says "adding 33% more cores" and not "adding 33% more die area". Not that die area is the be all end all of this - as I explained above.

The top SKU also has 4mb more L3 Cache than the 8700K, 16 over 12MB, so, again, more heat generation from that.

The heat increase here is not going to be 33% but it is still going to be significant.

We should expect something in the range of ~15-20%, all depending on the final clocks and binning. But remember that the memory controller, PCIe and DMI controllers, etc. will be unchanged. Also, L3 cache don't consume a lot of energy, just die space.

It's worth mentioning that we don't know the final clocks and TDP, I sure hope Intel will set more fair TDP.

 

[londiste]

From whence would anyone get the idea, knowing this, that the TDP could possibly stay at 95W?

And that's even assuming it was actually 95W after the move from Kaby Lake to Coffee Lake, which I personally doubt very much, as wasn't Intel's rationale for Z370 not being backwards compatible, that Z370 and hex cores, needed more power, that entry level Z170 and Z270 boards weren't necessarily able to provide with their VRMs, that were designed for quad core?

8700K has base clock of 3.7 GHz, 7700K has 4.2 GHz. I do not really see a problem with both having the same TDP considering the definition some posts back.

TDP definition aside, Intel has a tendency to work with certain specific TDP numbers. 105W, 95W, 65W, 35W and they'll try to find the matching frequency/voltage combination for the chip on the efficiency curve.

 

[bug]

Ah, good, I actually went looking for a die shot of the 8700k but didn't leave google, so only found shots of the older HEDT 6 cores.

So with that, we add 2 cores. We now have 33% more of the parts that comprise 50% of the die area, meaning that on the same manufacturing process the die is now ~16% larger and all of that 16% produces heat.

As I've been saying this whole time - that is not a trivial increase. From whence would anyone get the idea, knowing this, that the TDP could possibly stay at 95W?

And that's even assuming it was actually 95W after the move from Kaby Lake to Coffee Lake, which I personally doubt very much, as wasn't Intel's rationale for Z370 not being backwards compatible, that Z370 and hex cores, needed more power, that entry level Z170 and Z270 boards weren't necessarily able to provide with their VRMs, that were designed for quad core?

Let's not forget that the leak says 9900k will have a 3.1GHz base freq, down from 8700k's 3.7. That, if true, goes a long way towards cutting the cooling needs.
Still, as always, we shouldn't even try to pinpoint what 9900k will be and how it will perform. We got a ballpark figure, if the leak is genuine. That's all we know and discussing won't help us discover anything else.
Oh I'm also pretty sure this will be priced outside my comfort zone

 

[GlacierNine]

We should expect something in the range of ~15-20%, all depending on the final clocks and binning. But remember that the memory controller, PCIe and DMI controllers, etc. will be unchanged. Also, L3 cache don't consume a lot of energy, just die space.

It's worth mentioning that we don't know the final clocks and TDP, I sure hope Intel will set more fair TDP.

Well, we're not REALLY discussing what Intel quotes the TDP at, so much as what the real TDP is actually going to be in practice, compared to that quote.

I don't for one millisecond believe that 95W has been the real-world TDP of any Intel chip for quite some time. I think that ideal died with the introduction of per-core boost. (Which IIRC was Turbo Boost 2.0?) Ever since that day, we've been seeing base clocks go down and boost clocks and core counts go up, while TDP has been locked to this mystical 95W figure the entire time.

Your guess is as good as mine as to why, but I rather suspect you know what my guess is.

 

[jabbadap]

Ah, good, I actually went looking for a die shot of the 8700k but didn't leave google, so only found shots of the older HEDT 6 cores.

So with that, we add 2 cores. We now have 33% more of the parts that comprise 50% of the die area, meaning that on the same manufacturing process the die is now ~16% larger and all of that 16% produces heat.

As I've been saying this whole time - that is not a trivial increase. From whence would anyone get the idea, knowing this, that the TDP could possibly stay at 95W?

And that's even assuming it was actually 95W after the move from Kaby Lake to Coffee Lake, which I personally doubt very much, as wasn't Intel's rationale for Z370 not being backwards compatible, that Z370 and hex cores, needed more power, that entry level Z170 and Z270 boards weren't necessarily able to provide with their VRMs, that were designed for quad core?

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.

 

[londiste]

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