Discussion Intel current and future Lakes & Rapids thread

[lobz]

Yes, I did. Can you suggest a better word?

I do not answer to direct provocation.

 

[mikk]

It's way slower than it should be. 64EU Gen 11 is 2x 24EU Gen 9, but a 32EU Gen 11 is 20-30% faster than 24EU Gen 9.

It's scores are comparable to 32EU Gen 11, meaning zero improvement.

Even the top leaked TGL iGPU scores are 20-25% behind the level it should be. 5800+ FS score and 1700+ TS score should be doable as a 28W part. The best TGL gets 4500 FS and 1290 TS.

Unless they want to surprise us or something its not looking very good for samples leaking this late.

Just a thought, maybe Rocketlake will mirror Prescott. Hot with little to no gains.

Can you answer my question? What GPU clock did this sample run with? What memory speeds did it use? If you don't have these two substantial infos, how can you moan about the GPU score? You said the score is comparable, zero improvement. All the G1 Icelake powered DDR4 devices on Notebookcheck scored between 1500-1600, the RKL sample is 20% faster than this, this is zero for you. In general you are way too serious about these leaks.

 

[jpiniero]

Can you answer my question? What GPU clock did this sample run with? What memory speeds did it use? If you don't have these two substantial infos, how can you moan about the GPU score? You said the score is comparable, zero improvement. All the G1 Icelake powered DDR4 devices on Notebookcheck scored between 1500-1600, the RKL sample is 20% faster than this, this is zero for you. In general you are way too serious about these leaks.

https://www.notebookcheck.net/Intel-UHD-Graphics-G1-Ice-Lake-32-EU-Laptop-GPU.422873.0.html

I see 4 scores for G1 Icelake Firestrike on NBC, 1443/1468/1614/1673. 1746 would only be about 5% faster than the Dell. Graphics is pretty similar.

Clocks and Memory Speed on the Rocket were likely low, but they would be low on the Icelake models too. I think the whole purpose of backporting Gen12 was to get the updated features than actual performance gains in IGP gaming.

 

[IntelUser2000]

Intel's claims that 14nm++ uses relaxed pitch. So logically it should be larger right?

No, not according to actual data.

Skylake (client) - Microarchitectures - Intel - WikiChip

Skylake (SKL) Client Configuration is Intel's successor to Broadwell, a 14 nm process microarchitecture for mainstream workstations, desktops, and mobile devices. Skylake succeeded the short-lived Broadwell which experienced severe delays. Skylake is the 'Architecture' phase as part of Intel's...

en.wikichip.org

According to Wikichip, Skylake 4 core is 122mm2, and the 4 core + LLC section is 25.3mm2.

The official die size for the 8700K is 149mm2, or under 2mm2 off the 122 + 25.3 number. If you add 25mm2 to 8700K's figure, we get 174mm2 or under 3mm2 off the official number.

I bet what's happening is Skylake is actually slightly larger at 124-125mm2 since Intel never officially revealed the information.

@jpiniero One post in reddit says that Lenovo testers claim that drivers are in a crappy state and only 3DMark is stable, while games are quite a bit behind.

 

[jpiniero]

Intel's claims that 14nm++ uses relaxed pitch. So logically it should be larger right?

No, not according to actual data.

They did relax it but were able to compensate the density loss by pulling in Single Dummy Gate.

 

[mikk]

https://www.notebookcheck.net/Intel-UHD-Graphics-G1-Ice-Lake-32-EU-Laptop-GPU.422873.0.html

I see 4 scores for G1 Icelake Firestrike on NBC, 1443/1468/1614/1673. 1746 would only be about 5% faster than the Dell. Graphics is pretty similar.

Clocks and Memory Speed on the Rocket were likely low, but they would be low on the Icelake models too. I think the whole purpose of backporting Gen12 was to get the updated features than actual performance gains in IGP gaming.

There are 5 GPU scores: 1521, 1550, 1574, 1739, 1755. The score of 1755 is a LPDDR4-3733 device--> RKL-S don't support LPDDR4. As I said all the DDR4 powered scores are in the range of 1500-1600 and RKL-S 0000 is 20% faster than this. Some might describe this as zero, the thing is nobody has the full system specs of this leak.

 

[IntelUser2000]

DDR4-2666 is only for extreme value systems in desktops while its still quite common in laptops.

With Rocketlake it'll be common to see DDR4-3200, nevermind higher speeds set by users or certain brands. Besides, Xe should be even more bandwidth efficient than Gen 11 considering they are aiming for 2x performance at the same bandwidth. Rocketlake won't be constrained by TDP either.

Regardless all the Xe GPUs are seriously underperforming. If the reddit post is true, then they are still seriously struggling with the drivers. The compiler scheduling is both a curse and a blessing. Do it right and you save power and area, and it also encourages the developer of the hardware to make proper drivers since its such a performance critical feature. But if you screw it up, you can royally screw it up.

 

[mikk]

Regardless all the Xe GPUs are seriously underperforming. If the reddit post is true, then they are still seriously struggling with the drivers. The compiler scheduling is both a curse and a blessing. Do it right and you save power and area, and it also encourages the developer of the hardware to make proper drivers since its such a performance critical feature. But if you screw it up, you can royally screw it up.

A month ago I told you I haven't seen a driver which is close to final. Your answer: I do not believe drivers will improve it significantly.

 

[lobz]

A month ago I told you I haven't seen a driver which is close to final. Your answer: I do not believe drivers will improve it significantly.

I bet he meant not for 3Dmark. I would.

 

[Rigg]

Apparently, a lot of AMD fans replace a CPU yearly (or even more often: https://forums.anandtech.com/threads/my-new-ryzen-5-3600-hits-4-5-ghz-1-28v.2581268/).
Just look around on forums like this one.

Using my thread as an example doesn't help your argument. First of all, I'm not an AMD fan. I'm not an obtuse moron who has some misguided allegiance to a multi billion dollar company who's only real obligation is to it's stock holders. Any preference I have toward recommending AMD to others is purely based on it's price to performance ratio. If Intel were superior in price to performance I'd happily recommend them instead.

Secondly, I switch CPU's all of the time because I build and sell PC's as a hobby/side hustle. I'm far from a typical user. I've probably built more PC's in a month than most PC enthusiasts will build in a lifetime. I'm a terrible example of an "AMD fan" to use for your argument.

I switched from a 2700x to a 9900k before building the system in the thread you linked. I've also owned 8700k and 8700 CPU's. When I speak about these platforms I come from a position of having actually used the parts being discussed.

 

[IntelUser2000]

A month ago I told you I haven't seen a driver which is close to final. Your answer: I do not believe drivers will improve it significantly.

I tend to look for patterns, because its often not deviated. The low Tigerlake scores were not only consistent, but even back then were not far from official unveiling. I had to conclude its pretty close to a final driver. I just had a difficult time thinking 2-3 months before a release they could possibly not have a relatively finished driver. 2-3 months before release is for polishing, not for fixing critical bugs!

Even leakers like rogame thinks the same way. He believes 4700 Firestrike is pretty close to a final score. If its truly 2x, we should be seeing close to 6000.

And the 6300 3DMark 11 score, sheesh, that's terrible! It should be(and I don't mean to bring memes but) 9000!

Yes I was wrong. But release is 1-2 months at the most now. They want to be a leader in graphics and all they can show is the same crappy drivers? Do they want us to think they are hiding the best drivers or something and giving the worst for leakers?

According to the poster not even the perf/watt is better for TGL. Oh boy I hope for their sake it does everything they promised and more because even with that it'll be going against the tour-de-force called Renoir.

And arguably, a better CPU is better than having a better iGPU.

 

[mikk]

I tend to look for patterns, because its often not deviated. The low Tigerlake scores were not only consistent, but even back then were not far from official unveiling. I had to conclude its pretty close to a final driver. I just had a difficult time thinking 2-3 months before a release they could possibly not have a relatively finished driver. 2-3 months before release is for polishing, not for fixing critical bugs!

Regardless of the compiler scheduling, Xe is their biggest gen to gen architecture change since forever, the last big gen to gen change imho was Gen6->Gen7. Just because of this it might look different in this generation.

Even leakers like rogame thinks the same way. He believes 4700 Firestrike is pretty close to a final score. If its truly 2x, we should be seeing close to 6000.

He also believed 2.7/4.3 is the final clock speed and we won't see a faster one which obviously wasn't correct. It isn't meaningful, he don't have more informations.

And the 6300 3DMark 11 score, sheesh, that's terrible! It should be(and I don't mean to bring memes but) 9000!

As I said this device might have run at 15W because of the subpar physics CPU score (25W Icelake is faster), 15W I7-1065G7 scores sub 4000. If not, something is fishy there anyways unless TGL-U is slower than Icelake.

They want to be a leader in graphics and all they can show is the same crappy drivers? Do they want us to think they are hiding the best drivers or something and giving the worst for leakers?

To be honest they could do this to some extend. Are you aware that most of the leaks didn't even have a CPU Turbo in these 3dmark leaks? Only a very small amount of them had a turbo. It's like they tried to hide the high Turbo as long as possible.

 

[IntelUser2000]

I just looked at Gen 11 vs Vega 7 comparisons.

Man, they need a LOT of work to do in terms of optimization. Some games, especially newer demanding ones are 2/3rds of what it should be.

One game that is particularly weak for Gen 9 is Witcher 3. Gen 11 is 3x as fast. The slowest implementation of Gen 11(a Y part) beats the fastest Gen 9 one.

Maybe its not just drivers. Some parts of the hardware could be deficient.

 

[mikk]

Looks like the I7-1165G7 device which is floating around is a DDR4 device!

https://twitter.com/i/web/status/1270322986839691264

I don't know what DDR4 speeds it is using but even if it's using DDR4-3200 Renoir with LPDDR4-4266 has a major benefit in these GPU tests. There are not enough details available from these leaked scores, TDP is also an unknown factor.

 

[lobz]

Looks like the I7-1165G7 device which is floating around is a DDR4 device!

https://twitter.com/i/web/status/1270322986839691264

I don't know what DDR4 speeds it is using but even if it's using DDR4-3200 Renoir with LPDDR4-4266 has a major benefit in these GPU tests. There are not enough details available from these leaked scores, TDP is also an unknown factor.

I'd say, a potential of almost 100% variance in the TDP is the biggest unknown factor with these comparisons right now. If it's 15W, I'll happily be the first one to say: it's a freaking beast!

 

[mikk]

There is another entry with a 4.8 Ghz boost for the i7-1165G7. It is expected that the 4.7 or 4.8 Ghz boost is Intels Thermal Velocity Boost, this would explain the changing turbo speeds in all these entries. I wonder if the i7-1185G7 supports an even higher TVB.

 

[Zucker2k]

There is another entry with a 4.8 Ghz boost for the i7-1165G7. It is expected that the 4.7 or 4.8 Ghz boost is Intels Thermal Velocity Boost, this would explain the changing turbo speeds in all these entries. I wonder if the i7-1185G7 supports an even higher TVB.

4.8GHz on a broken 10nm process? Impossible!!
/s

 

[ondma]

4.8GHz on a broken 10nm process? Impossible!!
/s

Yea, but where are the 6 and 8 core chips. after what 3 years of modifications to the process?

 

[RetroZombie]

4.8GHz on a broken 10nm process?

What?!
Still broken 6 years later?

 

[coercitiv]

We finally have some good news on TGL and it's already being buried under gratuitous taunts. Noise is a must.

Considering Intel has already begun their PR efforts with the press I hope we'll get more juicy info soon.

 

[Zucker2k]

Yea, but where are the 6 and 8 core chips. after what 3 years of modifications to the process?

Good question. However, something tells me this high of a frequency bodes well for the future of the process. I've always wanted to see Intel's process engineers tame this beast so we can see its full potential unleashed.

 

[jpiniero]

Frequency hasn't really been the problem, it's always been about yield. Of course if you are actually intending to sell the product, you need to make sure you get enough chips that can pass the validation.

 

[RetroZombie]

Frequency hasn't really been the problem, it's always been about yield.

Wasn't both?
Who wants a new slower cpu because it has one Ghz less?

 

[jpiniero]

Wasn't both?
Who wants a new slower cpu because it has one Ghz less?

One way to improve yield is to reduce frequency and/or increase voltage (which also increases power draw).

 

[coercitiv]

Frequency hasn't really been the problem, it's always been about yield. Of course if you are actually intending to sell the product, you need to make sure you get enough chips that can pass the validation.

Are you saying their parametric yields were fine but their functional yields were in the gutter?