Discussion Intel current and future Lakes & Rapids thread

[Dayman1225]

I'm starting to think TGL-U is shipping as A0.

I'm sure the engineering team working on this project would love that but the chances of that happening is near zero.

 

[Ajay]

I'm sure the engineering team working on this project would love that but the chances of that happening is near zero.

Do you think the Intel engineers are iterating to improve functional yield or clocks?
With today’s simulation software and FPGA test chips, I’d think rolling good A0 silicon is a high probability.

 

[yeshua]

I'm sure this result has been discussed ad nauseam but I still admire it so much I'm gonna post it once again:

https://browser.geekbench.com/v5/cpu/1058208 / https://browser.geekbench.com/v5/cpu/1058208.gb5 (mind the second link).

What's interesting is that Tiger Lake

1) easily boosts to 4.2GHz so Intel definitely has progress in terms of frequency
2) base frequency is 2.3GHz which is a whole lot higher than the existing Ice Lake CPUs
3) Both ST and MT results are extremely impressive 1400/4920 points which is higher than any low voltage four-core CPU. In fact the MT result is even more impressive.

Looks like 10nm+ is finally starting to work for them though they have yet to crack even higher frequencies like AMD has managed with the TSMC 7nm node (4.6GHz).

I'm a little concerned there's seemingly zero increase in IPC vs ICL. Intel promised at the very least a 10% uplift and it's just not there. Considering the frequencies the sample ran at, it should have scored 1540 points in ST mode.

 

[lobz]

That's not even his worst take on the subject, he tweeted right after that:

https://twitter.com/x/status/1219987829465346048

To show how bad Intel is going to "pulverize" AMD in mobile offerings, he linked to a comparison of an unreleased, most likely top of the line TGL product against a 2500u. Yes, he compared intel's best unreleased product to AMD's mid tier 14 nm (not even 12nm) mobile processor launched over 2 years ago.

The man is golden!!! Good laughs, thanks for the share 😀

 

[Exist50]

A little confused here. CPU design at Hillsboro is gone? Only Israel is doing 'Core' development?

To clarify, only Core (capitalization important) development in Oregon is dead. Unless something has radically changed, they still handle CPU/SoC development, just using IDC's cores. If memory serves, IDC handled Skylake, its derivatives, and Icelake, while the Oregon team did Tiger Lake. Last I heard, IDC was going to do Rocket Lake and Alder Lake, while the Oregon team would do Meteor Lake.

Also, in response to the Keller speculation, I'm unsure of the exact timeline, but am pretty sure this was before he arrived. Maybe around 2017? How things might have changed since then, I can't say.

 

[itsmydamnation]

I'm a little concerned there's seemingly zero increase in IPC vs ICL. Intel promised at the very least a 10% uplift and it's just not there. Considering the frequencies the sample ran at, it should have scored 1540 points in ST mode.

Intel didn't say 10% IPC they said 10% performance, they can get that with boost/base clock improvement alone.

 

[Nothingness]

Do you think the Intel engineers are iterating to improve functional yield or clocks?
With today’s simulation software and FPGA test chips, I’d think rolling good A0 silicon is a high probability.

FPGA and other verification tools only ensure functional correctness. The rest like frequency and power can only be estimated by various tools which accuracy is far from perfect. I guess that's where you get most of the steppings from.

That being said even functional correctness is not always guaranteed before implementation and having silicon helps you run more cycles which might uncover new functional bugs.

 

[Exist50]

FPGA and other verification tools only ensure functional correctness. The rest like frequency and power can only be estimated by various tools which accuracy is far from perfect. I guess that's where you get most of the steppings from.

That being said even functional correctness is not always guaranteed before implementation and having silicon helps you run more cycles which might uncover new functional bugs.

Also, they require your design to be synthesizable.

 

[mikk]

I'm a little concerned there's seemingly zero increase in IPC vs ICL. Intel promised at the very least a 10% uplift and it's just not there. Considering the frequencies the sample ran at, it should have scored 1540 points in ST mode.

Not correct, they promised a double digits CPU uplift, this a combination of IPC and frequency. Current Geekbench results are stepping 0 based, later steppings are faster.

 

[exquisitechar]

I'm sure this result has been discussed ad nauseam but I still admire it so much I'm gonna post it once again:

https://browser.geekbench.com/v5/cpu/1058208 / https://browser.geekbench.com/v5/cpu/1058208.gb5 (mind the second link).

What's interesting is that Tiger Lake

1) easily boosts to 4.2GHz so Intel definitely has progress in terms of frequency
2) base frequency is 2.3GHz which is a whole lot higher than the existing Ice Lake CPUs
3) Both ST and MT results are extremely impressive 1400/4920 points which is higher than any low voltage four-core CPU. In fact the MT result is even more impressive.

Looks like 10nm+ is finally starting to work for them though they have yet to crack even higher frequencies like AMD has managed with the TSMC 7nm node (4.6GHz).

I'm a little concerned there's seemingly zero increase in IPC vs ICL. Intel promised at the very least a 10% uplift and it's just not there. Considering the frequencies the sample ran at, it should have scored 1540 points in ST mode.

The scores are good but indicate 0 improvement over ICL clock for clock. The overall score is somewhat inflated by the insane crypto score, the integer/FP scores are less impressive and they are what actually matters. The frequency isn’t all that great, this is a 28W part and the 28W ICL 1068G7 is already supposed to boost to 4.1GHz, it’s just been vaporware so far. Intel said it’s coming, though.

I'm sure the engineering team working on this project would love that but the chances of that happening is near zero.

Haven’t we already seen a newer stepping? Just not in a GB result.

 

[DrMrLordX]

I'm a little concerned there's seemingly zero increase in IPC vs ICL. Intel promised at the very least a 10% uplift and it's just not there. Considering the frequencies the sample ran at, it should have scored 1540 points in ST mode.

I'm sure I've said it at least once before, but there is the possibility that the cache restructuring in Tiger Lake may cause it to perform better in certain narrow cases vs. Ice Lake that aren't represented well in Geekbench 5.

 

[Adonisds]

That's not even his worst take on the subject, he tweeted right after that:

https://twitter.com/x/status/1219987829465346048

To show how bad Intel is going to "pulverize" AMD in mobile offerings, he linked to a comparison of an unreleased, most likely top of the line TGL product against a 2500u. Yes, he compared intel's best unreleased product to AMD's mid tier 14 nm (not even 12nm) mobile processor launched over 2 years ago.

That's interesting, I just found out I got blocked by him after replying to the tweet where he was comparing the 9900K to the 64 cores Rome in single threaded performance. I'm sure it wasn't anything toxic.

 

[uzzi38]

That's interesting, I just found out I got blocked by him after replying to the tweet where he was comparing the 9900K to the 64 cores Rome in single threaded performance. I'm sure it wasn't anything toxic.

Welcome to the club then

He blocks anyone that says something reasonable back to him that disagrees with what he says.

 

[Ajay]

To clarify, only Core (capitalization important) development in Oregon is dead. Unless something has radically changed, they still handle CPU/SoC development, just using IDC's cores. If memory serves, IDC handled Skylake, its derivatives, and Icelake, while the Oregon team did Tiger Lake. Last I heard, IDC was going to do Rocket Lake and Alder Lake, while the Oregon team would do Meteor Lake.

Also, in response to the Keller speculation, I'm unsure of the exact timeline, but am pretty sure this was before he arrived. Maybe around 2017? How things might have changed since then, I can't say.

That checks with what I've seen for hiring at Hillsboro. Thanks.

 

[Ajay]

I'm sure I've said it at least once before, but there is the possibility that the cache restructuring in Tiger Lake may cause it to perform better in certain narrow cases vs. Ice Lake that aren't represented well in Geekbench 5.

Depending on what was done, it could even be hurting PPC in order to reach higher clocks. We really won't know till we gets some details (which may not be till release).

 

[RetroZombie]

Keller started Intel's high-performance ARM core development instead of Ocean Cove, something like K12 he was not able to finish in AMD.

I seams Jim Keller is wasting his time with Tremont:
What's Significant about Intel Tremont?

Which is an huge disappointment, was he hired by the intel marketing?

 

[lobz]

Intel didn't say 10% IPC they said 10% performance, they can get that with boost/base clock improvement alone.

To clarify this: people here said 8% for weeks, saying that it'll still be ahead of Zen 3, because 18% (Sunny) + 8% (Willow) is 27,4% higher IPC than SKL, therefore faster than Zen 3 and than life etc. If you read this thread back, they really seemed to paint the picture as if it was not just that simple, but also certain to happen.

 

[liahos1]

Intel just posted record financial results
- 20.2bln of Q4'19 revenue, wall street expected 19.2bln
- Guided to 19bln of Q1'20 revenue, wall street expected $17.2
- DCG was up 19% yoy at 7.2bln, beating analyst esimates of 6.4bln
- CCG was at 10bln, beating analyst estimates of 9.74bln
- They guided to 17bln of capex vs street estimates of 16bln. And reiterated increasing wafer capacity by another 25% this year.

Notably, DCG ASPs were UP 5% yoy, despite ROME.

More interestingly - Intel stated
"10nm yield ahead of expectations… accelerating transition with nine product releases in 2020 "

Seems like they are doing well. I wonder what the accelerated products are?

 

[krumme]

Intel just posted record financial results
- 20.2bln of Q4'19 revenue, wall street expected 19.2bln
- Guided to 19bln of Q1'20 revenue, wall street expected $17.2
- DCG was up 19% yoy at 7.2bln, beating analyst esimates of 6.4bln
- CCG was at 10bln, beating analyst estimates of 9.74bln
- They guided to 17bln of capex vs street estimates of 16bln. And reiterated increasing wafer capacity by another 25% this year.

Notably, DCG ASPs were UP 5% yoy, despite ROME.

More interestingly - Intel stated
"10nm yield ahead of expectations… accelerating transition with nine product releases in 2020 "

Seems like they are doing well. I wonder what the accelerated products are?

Intel is for the most part in a monopoly market so when they are supply constrained, and demand is fairly stable, prices, asp and revenue goes up. Ask OPEC. Same mechanism. Intels portfolio is weak but it really matters little now and next year because amd is so insignificant.
Cant continue in 3 years time unless 10nm and 7nm comes om track, because TSMC 5nm is comming full steam no matter what.

 

[liahos1]

Intel is for the most part in a monopoly market so when they are supply constrained, and demand is fairly stable, prices, asp and revenue goes up. Ask OPEC. Same mechanism. Intels portfolio is weak but it really matters little now and next year because amd is so insignificant.
Cant continue in 3 years time unless 10nm and 7nm comes om track, because TSMC 5nm is comming full steam no matter what.

Thank you for the business analysis, but what do you think the accelerated 10nm products are?

 

[TheGiant]

Thank you for the business analysis, but what do you think the accelerated 10nm products are?

I am hoping for 10nm desktop parts and some 16C HEDT with possible more than 4 mem channels (or server)
I wonder what one API brings to the business, maybe more of shared CPU/GPU resources for general desktop computing....

 

[Dayman1225]

Thank you for the business analysis, but what do you think the accelerated 10nm products are?

Bob said:

• Tigerlake
• Snow Ridge SoC
• First dGPU
• Xeon for DC, Storage and Networking
• AI ASICs

and I assume more, since thats 7.

 

[Exist50]

Bob said:

• Tigerlake
• Snow Ridge SoC
• First dGPU
• Xeon for DC, Storage and Networking
• AI ASICs

and I assume more, since thats 7.

Might be multiple within some of those categories.

 

[jpiniero]

Thank you for the business analysis, but what do you think the accelerated 10nm products are?

Lakefield Refresh I would guess.

 

[RetroZombie]

"10nm yield ahead of expectations… accelerating transition with nine product releases in 2020 "

Interesting how someone already 4 years late can use the word accelerating.