Discussion Intel current and future Lakes & Rapids thread

[Timorous]

Yes, a real schedule does account for those things. Which is why a real schedule doesn't specify production down to the week over a year in advance.

I work in Automotive. The 1st production date is set years in advance.

 

[Spicedaddy]

H35 4-cores isn't a real H CPU. Should be named U35.

H45 is the real H, not out yet.

 

[jpiniero]

Obviously Intel wanted an H branded 11th Gen processor out there but doesn't trust they can actually yield that much of H45.

 

[uzzi38]

Obviously Intel wanted an H branded 11th Gen processor out there but doesn't trust they can actually yield that much of H45.

There's still no reason to believe TGL-H would have drastically worse yields than TGL-U - not to the point that they wouldn't even try launching right now.

The plan was just it to be ready for launch later, not now. That's all.

 

[Gideon]

It looks TGL_H is comnig in the end of Q2.

XMG expects Intel Tiger Lake-H (8-core) to launch in late Q2 - VideoCardz.com

XMG: Intel Tiger Lake-H to launch in June? We may finally have a possible launch date for Tiger Lake H(45). In a lengthy post on Reddit, an XMG representative provided an update on the current stock situation of NVIDIA GeForce RTX 30 GPUs and AMD Ryzen 5000 CPUs codenamed Cezanne. The company...

videocardz.com

This makes me think Alder Lake will not be here before Q4, otherwise it's gertting ridiculous

 

[jpiniero]

It looks TGL_H is comnig in the end of Q2.

XMG expects Intel Tiger Lake-H (8-core) to launch in late Q2 - VideoCardz.com

XMG: Intel Tiger Lake-H to launch in June? We may finally have a possible launch date for Tiger Lake H(45). In a lengthy post on Reddit, an XMG representative provided an update on the current stock situation of NVIDIA GeForce RTX 30 GPUs and AMD Ryzen 5000 CPUs codenamed Cezanne. The company...

videocardz.com

This makes me think Alder Lake will not be here before Q4, otherwise it's gertting ridiculous

There's always the paper launch option.

 

[DrMrLordX]

It looks TGL_H is comnig in the end of Q2.

Wow. That's pretty embarrassing.

 

[uzzi38]

Wow. That's pretty embarrassing.

Not really - kind of expected just by the fact that TGL-H35 launched at all tbh.

 

[misuspita]

If it does come that late, then I don't think AlderLake arrives faster than start of Q2'22. Posssibly last week of Q1'22 to sound like they delivered in Q1....

 

[RTX2080]

Retail version 11900KF:

旧帖归档 - Chiphell - 分享与交流用户体验

旧帖归档Chiphell - 分享与交流用户体验 ,Chiphell - 分享与交流用户体验

translate.google.com

seems too hot and too power hungry for 4.8Ghz@1.325v, with AVX3 comsume 250watts

 

[JoeRambo]

Looks around right for AVX512 workload, if anything that is ~700-900Gflops of performance @4.8Ghz.

 

[SAAA]

Feels like power consumption talk is easy these days yet everyone is only looking at max power without mentioning workloads.

220W FX was bad for it used that much power without even getting close to competition 1-4 threaded performance of that time, at most they managed some win with all 8 thread on very parallel workload tests, while at least today's power hogs aren't as far from the much more power efficient choices on 90% use cases.

I bet you in any game both comet and rocket lake won't pull anywhere near 250W, temperatures are another issue as even 1W with bad heat transfer could turn a spot hundred of degrees hot, here the larger 11th gen chips should mitigate somewhat.

 

[Asterox]

Retail version 11900KF:

旧帖归档 - Chiphell - 分享与交流用户体验

旧帖归档Chiphell - 分享与交流用户体验 ,Chiphell - 分享与交流用户体验

translate.google.com

seems too hot and too power hungry for 4.8Ghz@1.325v, with AVX3 comsume 250watts

As expected, backporting CPU arhitecture is never good thing.

Can anyone remember, when was last time Intel did this CPU backporting? Too close to Alder Lake+new Socket, blah Rocket Lake is so an unnecessary and pointless move or mess.

 

[ondma]

As expected, backporting CPU arhitecture is never good thing.

Can anyone remember, when was last time Intel did this CPU backporting? Too close to Alder Lake+new Socket, blah Rocket Lake is so an unnecessary and pointless move or mess.

Why is it pointless? Granted it has the same power consumption issues as CL, but with better IPC. So it is a net gain. I think it depends on when Alder Lake comes out. If Alder Lake comes out Q3 or even Q4 2021 with good availability and performance, then one could argue RL is hardly worth it. But if AL is delayed until 2022, RL will fill a gap. Plus if they extend RL to lower end chips, it will give a performance increase, take some pressure of 10 nm production, and power will be much less of an issue in lower core count chips that are not pushed to max frequency.

Obviously, it is not ideal, but sometimes you just have to "go with what you got".

 

[dullard]

Feels like power consumption talk is easy these days yet everyone is only looking at max power without mentioning workloads.

220W FX was bad for it used that much power without even getting close to competition 1-4 threaded performance of that time, at most they managed some win with all 8 thread on very parallel workload tests, while at least today's power hogs aren't as far from the much more power efficient choices on 90% use cases.

I bet you in any game both comet and rocket lake won't pull anywhere near 250W, temperatures are another issue as even 1W with bad heat transfer could turn a spot hundred of degrees hot, here the larger 11th gen chips should mitigate somewhat.

When pushing the limits, Rocket Lake will be a power consuming monster....For 28 seconds...Then go back to rated power.

Power timing often gets lost in reviews. Either (A) the benchmark is so fast that the processor is really using a lot of power. Or (B) the benchmark is long and the power consumption is at the cooling capability of the system. This is a subtle but important difference that gets lost. Report that it uses 200+W, but then neglect to mention that for 99.99% of the benchmarking, it used 125W. A max power well over 200W is a lot of power. But it is also quite temporary.

That is unless you yank off the supplied cooler, replace it with something that can cool more, then the processor will be faster and use more power. But you basically put yourself in that situation. Do you want it to be faster and use more than the rated power? You can do that. But you should review it with that choice clearly highlighted and the effects clearly described.

 

[uzzi38]

Why is it pointless? Granted it has the same power consumption issues as CL, but with better IPC. So it is a net gain. I think it depends on when Alder Lake comes out. If Alder Lake comes out Q3 or even Q4 2021 with good availability and performance, then one could argue RL is hardly worth it. But if AL is delayed until 2022, RL will fill a gap. Plus if they extend RL to lower end chips, it will give a performance increase, take some pressure of 10 nm production, and power will be much less of an issue in lower core count chips that are not pushed to max frequency.

Obviously, it is not ideal, but sometimes you just have to "go with what you got".

For what it's worth Intel have publically said ADL launches in 2021 and they've been telling OEMs the same as well.

Expecting it to launch in Q4 is a good bet.

And with all of Intel's launched they do at the least ensure the lower-end SKUs are in stock. The -K SKUs tend to be low stock for at least a month (has been this way since CFL), so again I'd go with that expectation.

IMO if you can avoid buying RKL you should. Just waiting to see what ADL turns out as will probably be worth it. If your system is absolutely chugging in the first half of next year or you're building one from scratch, then buying RKL is reasonable provided you stay away from the i9 IMO.

The 11900K and 11900 are both likely going to end up wastes. Vastly higher prices for somewhat better bins just isn't worth it. But the 11700K and below? There's real potential for them to be worth it if Intel keeps the pricing the same as last gen for each product tier.

 

[Hulk]

As expected, backporting CPU arhitecture is never good thing.

Can anyone remember, when was last time Intel did this CPU backporting? Too close to Alder Lake+new Socket, blah Rocket Lake is so an unnecessary and pointless move or mess.

2:19

"Ice Lake hasn't worked. The cores inside Ice Lake did not work. The idea of bringing them to the desktop next year baffles me."

Am I missing something? Hasn't Intel been producing Ice Lake parts for like 2 years?

 

[tamz_msc]

1.4 V for an all-core clock of 4.8GHz is too high. That's why you have it reaching 98°C. Must be some auto-OC or out of whack BIOS.

 

[eek2121]

1.4 V for an all-core clock of 4.8GHz is too high. That's why you have it reaching 98°C. Must be some auto-OC or out of whack BIOS.

Wait until he finds out how amps/wattage changes the equation. 🤣

 

[uzzi38]

1.4 V for an all-core clock of 4.8GHz is too high. That's why you have it reaching 98°C. Must be some auto-OC or out of whack BIOS.

It was running 1.325V, not 1.4V.

Also, how are you determining what is too high and what is not on a new architecture that hasn't been released publically yet?

 

[tamz_msc]

It was running 1.325V

CPU-Z reads 1.4 V. Even if it is 1.325 V it's still too high for 4.8 GHz.

Also, how are you determining what is too high and what is not on a new architecture that hasn't been released publically yet?

Because at this point, the V-f curve after multiple iterations of 14nm would be bunched up really close to one another, being primarily a function of the process node rather than architecture, unlike Zen 2 -> Zen 3 where there was still scope for refinements as evidenced by the V-f curve of those two architectures.

 

[uzzi38]

CPU-Z reads 1.4 V. Even if it is 1.325 V it's still too high for 4.8 GHz.

Because at this point, the V-f curve after multiple iterations of 14nm would be bunched up really close to one another, being primarily a function of the process node rather than architecture, unlike Zen 2 -> Zen 3 where there was still scope for refinements as evidenced by the V-f curve of those two architectures.

If you read the article the leaker specifically stated the voltage was 1.325V.

Also, V/f curve is dependant on both node and architecture. Zen 3 is on the same process node as the Zen 2 XT SKUs, yet it clocks higher at the same voltage.

 

[tamz_msc]

If you read the article the leaker specifically stated the voltage was 1.325V.

Also, V/f curve is dependant on both node and architecture. Zen 3 is on the same process node as the Zen 2 XT SKUs, yet it clocks higher at the same voltage.

The leaker doesn't have a reliable voltage measurement because they choose to trust one software reading over another. It could be anything, but I'm inclined to believe that it's pretty high.

Zen 3 definitely saw improvements in the V-f curve but that was because TSMC and AMD also refined the 7nm node as well. While we don't have direct evidence for this yet, the existence of Lucienne alongside Cezanne points to this being the case.

 

[uzzi38]

Zen 3 definitely saw improvements in the V-f curve but that was because TSMC and AMD also refined the 7nm node as well. While we don't have direct evidence for this yet, the existence of Lucienne alongside Cezanne points to this being the case.

Referring to Vermeer here:
"It's the same 7nm recipe we are/were using for 3000XT"
"n7 + better AMD secret recipe"

- Robert Hallock on a Discord group chat.

 

[tamz_msc]

Referring to Vermeer here:
"It's the same 7nm recipe we are/were using for 3000XT"
"n7 + better AMD secret recipe"

- Robert Hallock on a Discord group chat.

That secret sauce is BEOL improvements.