Intel Core i9-13900KS Could be World's First 6 GHz Processor

[fevgatos]

Good thing 6 Ghz won't be an all core workload then.

Sure but then you dont need a tec. Im doing 5.8 on a small air cooler

 

[dgianstefani]

I would argue edge case frequency workloads such as 6 Ghz single core boost is precisely the ideal situation to use a TEC. As you may know, voltage leakage is increased at higher temperatures, so there is a kind of thermal/voltage runaway to maintain stability if pushing high current without adequate cooling.

The cooler a chip is, the less voltage is required to run it stably, as a general rule.

Sure but then you dont need a tec. Im doing 5.8 on a small air cooler

 

[fevgatos]

I would argue edge case frequency workloads such as 6 Ghz single core boost is precisely the ideal situation to use a TEC. As you may know, voltage leakage is increased at higher temperatures, so there is a kind of thermal/voltage runaway to maintain stability if pushing high current without adequate cooling.

The cooler a chip is, the less voltage is required to run it stably, as a general rule.

Yeah, at around 80c stability starts going out the window and you need more and more voltage. That's why im running 5.6 on 3 cores instead of 5.8, to be stable at 5.8 I need to push 1.58 volts

 

[TheoneandonlyMrK]

Yeah, at around 80c stability starts going out the window and you need more and more voltage. That's why im running 5.6 on 3 cores instead of 5.8, to be stable at 5.8 I need to push 1.58 volts

Oh, because you declare 5.8 every other day without that caveat of 3x5.6.

Clearly you're running a manually honed setup that's had way more time and thought put in then the average Joe yet. Use your setup as a cludgeon to bait AMD user's with bullshit comments like mine runs fast cool and efficient, perhaps but not without an amount of effort most Won't put in.
Thanks for the clarity, post noted.

 

[Outback Bronze]

The cooler a chip is, the less voltage is required to run it stably, as a general rule.

Can this also be a case of better silicone too? Better silicone = lower voltage = lower temps?

I'm finding the better silicone you've got the better off you are, as a general rule for the average overclocker.

 

[fevgatos]

Oh, because you declare 5.8 every other day without that caveat of 3x5.6.

Clearly you're running a manually honed setup that's had way more time and thought put in then the average Joe yet. Use your setup as a cludgeon to bait AMD user's with bullshit comments like mine runs fast cool and efficient, perhaps but not without an amount of effort most Won't put in.
Thanks for the clarity, post noted.

You think someone got baited and thought im running ALL Core at 5.8? lolk, thought its obvious we are talking about ST speeds, although in my case I use it for 3 cores for more consistent boosting

 

[TheoneandonlyMrK]

Sure but then you dont need a tec. Im doing 5.8 on a small air cooler

Hmmn , bullshit comments meant to mislead evidence found.

 

[dgianstefani]

Can this also be a case of better silicone too? Better silicone = lower voltage = lower temps?

I'm finding the better silicone you've got the better off you are, as a general rule for the average overclocker.

Not necessarily, there is "better silicone" for different things, i.e. a top binned mobile chip may have fantastic frequency range at a low voltage, but will not scale to higher frequencies at high voltages as well as a differently binned chip. Leaky, hot silicon can sometimes be the absolute best at overclocking, assuming you can cool it.

There's also chips that have excellent core silicon but terrible IMC, fast cores that aren't fed with fast memory at low latency are bandwidth starved, and won't actually end up faster than a different chip with worse cores but a more normal IMC e.g.

 

[fevgatos]

Hmmn , bullshit comments meant to mislead evidence found.

Whats bullshit about it? I've done 5.8 on 3 cores on an air cooler, have uploaded benchmarks on CPU Z and cinebench. The heck are you talking about?

 

[DeathtoGnomes]

Or maybe boosted Intels frequency and doubled the power consumed? Since Intel is advertising the 6ghz, the idea of an Intel top chip sucking 400watts seems plausible now.

Nvidia set the bar for higher wattages and Intel just followed suit. 400w just to keep up with the 5.7Ghz.

 

[napata]

Ipc increase over alderlake? Doubt

You think they increased L2 to waste die space? More L3 too.

 

[R-T-B]

Can this also be a case of better silicone too?

Drop the e and maybe.

 

[Chrispy_]

Yes, yes.... of course 6GHz was inevitable, we've been north of 5GHz for a few years now.

The real question is "at what cost?"

6GHz at 250W simply isn't worth it when 5.5GHz is only 100W.

 

[Tropick]

Leaky, hot silicon can sometimes be the absolute best at overclocking, assuming you can cool it.

Can absolutely confirm. Was benching a delidded GTX 580 Classified for fun, had 55ish% ASIC quality reported by GPU-Z but that thing could take a ludicrous amount of voltage.

Had it on an Kraken X73 AIO direct die cooled with Conductonaut LM paste. At 1.3v the thing was an absolute fireball sucking down 384W TBP. Needed that AIO running max pump/fan speed to keep temps in check but it was game stable at 1035MHz and bench stable at 1067MHz. Sometimes the worse the silicon the harder you can push it.

I scored 6 037 in Fire Strike

AMD Ryzen 5 5600X, NVIDIA GeForce GTX 580 x 1, 32768 MB, 64-bit Windows 10}

www.3dmark.com

 

[Sabotaged_Enigma]

A strange competition is going on here, or to say the competition is going the wrong way.
Abandoning power efficiency and making the last juice of gigahertz is just far from smart. I'm saying this on both AMD and Intel.
What's the actual/realistic meaning of huge amount of power bringing that high clock speed? Who's gonna buy a CPU and see that number in monitoring software flying so high and making himself happy?
After all, if it's good, it's good; if it's rubbish, it's rubbish. Frequency game changes nothing.

 

[Outback Bronze]

Leaky, hot silicon can sometimes be the absolute best at overclocking, assuming you can cool it.

Yes, I agree but I was referring to the "average overclocker" so on air or simple water cooler.

 

[aQi]

Time to bring that old chiller out.

Lol, just the comment i was looking for

 

[dgianstefani]

Yes, I agree but I was referring to the "average overclocker" so on air or simple water cooler.

There's not much room for the average overclocker these days, chips are pushed fairly well out of the box. If you're going to overclock it needs to be done properly, or in many cases performance will actually be lost. I.e. only really worth OC if you eliminate cooling and power issues, then manually tune for frequency and latency.

Adjusting for more efficiency is a different ball park.

 

[The Von Matrices]

A strange competition is going on here, or to say the competition is going the wrong way.
Abandoning power efficiency and making the last juice of gigahertz is just far from smart. I'm saying this on both AMD and Intel.
What's the actual/realistic meaning of huge amount of power bringing that high clock speed? Who's gonna buy a CPU and see that number in monitoring software flying so high and making himself happy?
After all, if it's good, it's good; if it's rubbish, it's rubbish. Frequency game changes nothing.

Dennard scaling is over. It used to be that CPUs were limited by silicon switching speed long before power became a limiting factor; now the opposite is true. And since power efficiency gains are declining with each node shrink, the main way to increase performance now is to increase power. If TDP limits were fixed we'd be seeing only ~10% performance improvements with each node shrink because even though more transistors were on the chip, clock speeds would have to decline to keep constant power.

 

[watzupken]

Fixed with good IHS substrate and solder, further fixed if you delid and apply liquid metal. For the people who care about 200Mhz 6Ghz numbers, this isn't a significant obstacle.

There's also TEC coolers which Intel has recently pushed. Just an engineering/design challenge really. CPU's aren't going to change anytime soon in their die sizes compared to GPUs, and the bleeding edge of performance will always cost that exponential bit more.

When you have this much heat output and an IHS over it, it’s going to run hot no matter how good the IHS substrate and solder is. Don’t forget GPU dies are bare, and the GPU chip itself don’t actually draw as much power as this CPU itself. The 350W power draw is collectively used by the components on the board. Intel’s been marketing CPU with very high clock speed where most conventional coolers can no longer manage the power it requires and heat output that results from it. I would be interested to see if Raptor Lake will require a significant power bump when actual review comes up.

 

[Sabotaged_Enigma]

If TDP limits were fixed we'd be seeing only ~10% performance improvements with each node shrink because even though more transistors were on the chip, clock speeds would have to decline to keep constant power.

Sorry but I don't think so. I still believe with a well-designed microarchitecture and a well-selected process code, we would get pretty decent performance gain. Moving from Zen 2 to Zen 3 took only a microarchitecture improvement other than process node or power limit or huge transistor count increase. A 19% IPC lift is still a huge leap even frequency is not increased a lot.
So I say the competition is going the wrong way. The major way I believe is still designing excellent microarchitecture and making it efficient. Otherwise, why not make a 10 nm version of Pentium 4 and make it 10 GHz, and say look we've got the first 10 GHz CPU in the world with small-sized footprint saying "oops if you ignore rubbish performance and astonishing power consumption". What about 7 nm version of 10 GHz Bulldozer? Does anybody like it or does it make any sense?

 

[Minus Infinity]

So Intel has edited the scripts and massaged what they said was a 8-9% IPC uplift into magically now 15% just a b it better than AMD's 13%. Simply amazing effort from Intel's marketers.

We all knew 13900KS would go for broke for bragging rights, power be damned. Brain dead and egregious waste of resources. Huang will do the same with Lovelace. Both will be under immense pressure going forward and Intel has to raise prices. 13900KS will be $850+ IMO. WIll be a perfect match for the ludicrous 4090Ti, better get that 1.5kW PSU order in quick.

 

[trparky]

Given that high end AIOs struggle with 12900KS what kind of cooling solution you'd need to cool this alleged CPU?

A nuclear power station style cooling tower.

 

[ratirt]

So Intel has edited the scripts and massaged what they said was a 8-9% IPC uplift into magically now 15% just a b it better than AMD's 13%. Simply amazing effort from Intel's marketers.

We all knew 13900KS would go for broke for bragging rights, power be damned. Brain dead and egregious waste of resources. Huang will do the same with Lovelace. Both will be under immense pressure going forward and Intel has to raise prices. 13900KS will be $850+ IMO. WIll be a perfect match for the ludicrous 4090Ti, better get that 1.5kW PSU order in quick.

and we are supposedly be in a power crisis I agree with the Intel went for the 6Ghz to be better than AMD. As the first to get there. It would seem Intel has an obsession over being the best at everything and that would also mean enormous power consumption as well They have a very nice strides in that area to be honest.

 

[Jimmy_]

hmmm.. IPC gains over ADL - doubtful to me a bit, as I haven't seem anywhere regrading IPC gains or may they ll cover during their launch events -