Discussion Intel current and future Lakes & Rapids thread

Page 257 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
So, you can't back up your claim that the y-axis is "where this slide utterly fails to sniff test "?

Sunny Cove is crap on frequencies (what we are talking about). Not unusable crap, but crap. 40% better than crap is not impossible.
Its not over yet. But stockholders don't create lawsuits for nothing.
 
  • Like
Reactions: Tlh97 and Drazick

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
So you were lying when you implied they were held legally liable?
I was not lying, as you said I implied, since its not over yet, and I believe they will be help liable when it is over. Stop twisting my words, you are as bad as Intel.
 
  • Like
Reactions: Tlh97 and Drazick

dullard

Elite Member
May 21, 2001
24,998
3,327
126
I was not lying, as you said I implied, since its not over yet, and I believe they will be help liable when it is over. Stop twisting my words, you are as bad as Intel.
I'm not sure why you didn't just come out and say you personally believe that Intel will be held legally liable. That is way more straightforward than these words:
With Intel's track record of lying about everything lately...You don't get held legally liable for missing deadlines, just lying about them.
Marking your opinions as opinions just goes over so much better on forums. Certainly better than stating your opinions as facts, then coming back and saying that people thinking that you said them are facts are twisting your words.
 
  • Like
Reactions: Exist50

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
I'm not sure why you didn't just come out and say you personally believe that Intel will be held legally liable. That is way more straightforward than these words:

Marking your opinions as opinions just goes over so much better on forums. Certainly better than stating your opinions as facts, then coming back and saying that people thinking that you said them are facts are twisting your words.
To you... The same problem you are having with the others on the chart.
 
  • Like
Reactions: Tlh97 and Drazick

Hitman928

Diamond Member
Apr 15, 2012
5,182
7,633
136
Like I said above, a better graph would have labels (such as units). See post #6391. But, as long as we are talking about percentages, units aren't needed.

That's the whole point, the obfuscation in the chart would allow Intel to make even a tiny improvement seem huge, we have no way of knowing what the actual improvement is. For instance, what if the y-axis is normalized to 100 MHz and the x-axis is in 0.01 V steps. Would you be really impressed at that point? I'm not saying that's what Intel did, but based upon the graph, could you prove that they didn't do this? I'll reserve judgement until we see actual numbers, but the chart tells us nothing without a sense of scale or real units and numbers applied.
 

dullard

Elite Member
May 21, 2001
24,998
3,327
126
To you... The same problem you are having with the others on the chart.
I do not understand what you are trying to convey. Could you please rephrase? The chart shows roughly 40% higher frequency at the voltage that gives 2.5 (presumably GHz) on Sunny Cove. The chart may be incorrect marketing fluff. Independent reviews can verify that in a few weeks. But, that is exactly what the chart says right now.

What I think dmem was stuck on was that the graph is not 40% frequency gains at all voltages, just at that one single voltage. At the highest Sunny Cove voltage shown on the graph, the gain was only 17.9% (3.9 to 4.6). 17.9% is quite achievable with newer architecture. I personally think that dmens just didn't read the chart correctly. It was a badly made chart, I agree with that. But, you can still get some information from it.
 

ondma

Platinum Member
Mar 18, 2018
2,718
1,278
136
The lack of X-axis numbers renders left-right arrow means your left-right arrow is absolutely meaningless for comparative purposes.

You can try to draw conclusions from the vertical axis since it is supposedly linearly scaled, and that is where this slide utterly fails to sniff test: if the claim is they managed to achieve a general 40% increase in frequency on the same voltage (just ignore power for now) on a typical, average sampled part from two very similar core architectures, we are talking about gains that used to take multiple process generations to achieve. That is not happening, unless Intel completely abandoned low power operational characteristics, which while possible, won't cheat third-party power benchmarks no matter how much they re-define TDP, PL* windows, whatever. You can add on top the troubled development history of the 10nm process, but really that is not even necessary to render these claims odorously offensive.

My professional opinion, this is borderline fraudulent marketing. My take is that they can avoid legal liability by simply claiming they are not comparing average parts.
WTH are you talking about? That graph clearly shows one thing: at the same voltage, WC attains a 3.5 ghz, while SC attains 2.5 ghz. That is clearly shown in the graph, whether the x axis is labelled or not, or whether it is linear or not. You can choose to believe that or not, but that is what the graph shows.
 

dullard

Elite Member
May 21, 2001
24,998
3,327
126
That's the whole point, the obfuscation in the chart would allow Intel to make even a tiny improvement seem huge, we have no way of knowing what the actual improvement is. For instance, what if the y-axis is normalized to 100 MHz and the x-axis is in 0.01 V steps. Would you be really impressed at that point? I'm not saying that's what Intel did, but based upon the graph, could you prove that they didn't do this? I'll reserve judgement until we see actual numbers, but the chart tells us nothing without a sense of scale or real units and numbers applied.
It was obfuscated. Yes, that is correct.

But it did give just enough information to reach a few conclusions. If the graph is independently verified to be correct, then Willow Cove has way better frequencies at low voltages and 17.9% better frequencies at high voltages. It doesn't say anything about IPC. It doesn't say anything about power required. It simply says the frequency is higher at each voltage which is not that much of a stretch given how low frequencies are on Sunny Cove.

You are correct that we don't know what those voltages are. But, if the graph is correct, then the Willow Cove shown on that graph is a much improved mobile processor, and a decent but not great desktop processor improvement.
 
Last edited:

dullard

Elite Member
May 21, 2001
24,998
3,327
126
WTH are you talking about? That graph clearly shows one thing: at the same voltage, WC attains a 3.5 ghz, while SC attains 2.5 ghz. That is clearly shown in the graph, whether the x axis is labelled or not, or whether it is linear or not. You can choose to believe that or not, but that is what the graph shows.
Some people just can't read graphs, or refuse to read them. I'm not quite certain which of the two options is going on here.
 

btarlinian

Junior Member
Jun 23, 2020
8
15
51
That's the whole point, the obfuscation in the chart would allow Intel to make even a tiny improvement seem huge, we have no way of knowing what the actual improvement is. For instance, what if the y-axis is normalized to 100 MHz and the x-axis is in 0.01 V steps. Would you be really impressed at that point? I'm not saying that's what Intel did, but based upon the graph, could you prove that they didn't do this? I'll reserve judgement until we see actual numbers, but the chart tells us nothing without a sense of scale or real units and numbers applied.

The y-axis units are very obviously GHz. You can go look up the range of clockspeeds available on Sunny Cove in publicly available Icelake processors to verify that if you're concerned by the lack of an explicit label. The only plausible argument here is that Intel is simply making stuff up or doing some sort of really biased comparison (e.g., comparing i3 bins of ICL to i7 bins of TGL, which as I showed earlier is implausible based on the 1065G7 to 1165G7 clock comparison.)

In regards to shareholder lawsuits, they are a dime a dozen, and occur almost any time bad news causes stock prices to drop and are not particularly strong evidence of Intel's press releases being fraudulent. As everyone knows, everything everywhere is securities fraud.
 

Cardyak

Member
Sep 12, 2018
72
159
106
Well, Willow Cove providing a “Mid-Single digit IPC gain“ over Sunny Cove is more than a little disappointing.

Golden Cove is going to have to be an absolute monster to overtake Zen 3, let alone Zen 4.
 

dmens

Platinum Member
Mar 18, 2005
2,271
917
136
WTH are you talking about? That graph clearly shows one thing: at the same voltage, WC attains a 3.5 ghz, while SC attains 2.5 ghz. That is clearly shown in the graph, whether the x axis is labelled or not, or whether it is linear or not. You can choose to believe that or not, but that is what the graph shows.

Did you even read the post?

- A comparative claim on the x-axis would be the voltage differential to reach the same frequency. That claim cannot be made at all with that graph, as in, when I said, the x-axis cannot be used to make a comparison.

- A comparative claim on the y-axis would be the frequency differential at the same voltage. Whether you choose to believe those ridiculous numbers is up to how gullible or ignorant someone is to believe such low-grade marketing FUD.
 
  • Like
Reactions: Tlh97

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
It was obfuscated. Yes, that is correct.

Obfuscated or not, Dullard is right. The product will have the final word anyways.

10nm SF is a good gain plus Icelake sucks in frequencies. What's so hard to believe? We know from products that sucking in frequencies are likely true for all frequency levels because Icelake sometimes regresses in battery life compared to the predecessor.

NBC tests show 1065G7, the top Icelake part clocks at 2.4-2.6GHz using 25W running Prime95 stress test, and 1.8-2GHz at 15W. The Cometlake 10510U can clock from 2.7-2.9GHz at 25W. At 15W, 10510U is almost at 1065G7 25W level clocking at 2.3GHz.

1597449885202.png
1597449905732.png
1597449936271.png

It's clearly frequency, since Sunny Cove goes to 3.9GHz in the 1065G7 product and the graph is slightly under 4GHz.

Even Willow Cove's graph is based on a real product as we know the Turbo will be in the 4.7GHz range. That's exactly what straight line up extrapolation from Sunny Cove gives us.

Call it coincidence or not, but the second graph labeled "Increased Power Efficiency" is at 3.6GHz, which is the Turbo frequency for one of the leaked -Y variants. Along with slightly improved uarch, it essentially means same performance as the 25W Icelake but at 10-12W, or fanless envelopes.
 
Last edited:

Hitman928

Diamond Member
Apr 15, 2012
5,182
7,633
136
It was obfuscated. Yes, that is correct.

But it did give just enough information to reach a few conclusions. If the graph is independently verified to be correct, then Willow Cove has way better frequencies at low voltages and 17.9% better frequencies at high voltages. It doesn't say anything about IPC. It doesn't say anything about power required. It simply says the frequency is higher at each voltage which is not that much of a stretch given how low frequencies are on Sunny Cove.

You are correct that we don't know what those voltages are. But, if the graph is correct, then the Willow Cove shown on that graph is a much improved mobile processor, and a decent but not great desktop processor improvement.
The y-axis units are very obviously GHz. You can go look up the range of clockspeeds available on Sunny Cove in publicly available Icelake processors to verify that if you're concerned by the lack of an explicit label. The only plausible argument here is that Intel is simply making stuff up or doing some sort of really biased comparison (e.g., comparing i3 bins of ICL to i7 bins of TGL, which as I showed earlier is implausible based on the 1065G7 to 1165G7 clock comparison.)

In regards to shareholder lawsuits, they are a dime a dozen, and occur almost any time bad news causes stock prices to drop and are not particularly strong evidence of Intel's press releases being fraudulent. As everyone knows, everything everywhere is securities fraud.

All I'm saying is that until I see actual graphs or hard numbers, I don't trust marketing slides, and that is what these really are. Could reality be 100% what Intel is implying here? Sure. But I'm not going to hang my hat on these graphs. I've had marketing people use slides like this for products I've worked on where I would say they were lying about what the product was really capable of, but technically they did it in a way that you could say was accurate, just extremely misleading at face value. I'm all for renewed competition in the x86 space, I just want more than fuzzy slides before thinking we're there.
 

Hans de Vries

Senior member
May 2, 2008
321
1,018
136
www.chip-architect.com
Yes, it would be nice to have the x-axis numbers. Yes, it is a marketing slide. But, if you were an engineer (or similar occupation), how come you don't realize that x-axis values are not needed to answer this question from the slide? See below were a few red lines were drawn to get the answer, all without any x-axis values:
View attachment 27985



If only these graphs weren't so crappy inconsistent. The same curve is very different on two consecutive pages.

VF curve1.jpg
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
If only these graphs weren't so crappy inconsistent. The same curve is very different on two consecutive pages.

I assume the slides just for being marketing purpose, it does not matter that they are inconsistent - because there are no real numbers behind - they show something rather qualitatively instead of quantitatively.
 
  • Like
Reactions: teejee

ondma

Platinum Member
Mar 18, 2018
2,718
1,278
136
Did you even read the post?

- A comparative claim on the x-axis would be the voltage differential to reach the same frequency. That claim cannot be made at all with that graph, as in, when I said, the x-axis cannot be used to make a comparison.

- A comparative claim on the y-axis would be the frequency differential at the same voltage. Whether you choose to believe those ridiculous numbers is up to how gullible or ignorant someone is to believe such low-grade marketing FUD.
Did you even read my post? I specifically said all the graph shows is that at the particular voltage chosen, the the graph shows the frequency difference. I never said anything about different voltages, as, duh, obviously no numbers are given for the voltage.

I actually think we probably agree on that, although it seems you insist on directing your venom against Intel toward any other poster who does not share your feelings.
 

dmens

Platinum Member
Mar 18, 2005
2,271
917
136
Did you even read my post? I specifically said all the graph shows is that at the particular voltage chosen, the the graph shows the frequency difference. I never said anything about different voltages, as, duh, obviously no numbers are given for the voltage.

I actually think we probably agree on that, although it seems you insist on directing your venom against Intel toward any other poster who does not share your feelings.

Anyone who believes anything Intel says in the year 2020 needs to have their heads checked.
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
Anyone who believes anything Intel says in the year 2020 needs to have their heads checked.
Except if it's bad news for Intel, I suppose?
Icelake is "crap" until Tigerlake shows up with better frequency and power numbers and then Icelake is no longer "crap." Not a great day for doomsayers.
 
  • Like
Reactions: vstar and ondma

eek2121

Platinum Member
Aug 2, 2005
2,904
3,906
136
All I'm saying is that until I see actual graphs or hard numbers, I don't trust marketing slides, and that is what these really are. Could reality be 100% what Intel is implying here? Sure. But I'm not going to hang my hat on these graphs. I've had marketing people use slides like this for products I've worked on where I would say they were lying about what the product was really capable of, but technically they did it in a way that you could say was accurate, just extremely misleading at face value. I'm all for renewed competition in the x86 space, I just want more than fuzzy slides before thinking we're there.
Anyone who believes anything Intel says in the year 2020 needs to have their heads checked.

You don’t need to believe Intel. The evidence is out there. We have proof these chips are hitting 4.7 Ghz. Intel has told the press TGU is a 15w part. I am as sceptical as the next person when it comes to marketing claims, but Ice Lake didn’t hit 4.7 Ghz and nobody in their right mind would believe that U series CPUs would consume as much power as H series. We all knew that once Intel got 10nm straightened out they would have much better products. Only time will tell if it is enough to compete. I suppose you guys are going to call Ian a liar as well for detailing Tiger Lake?
 
  • Like
Reactions: geegee83 and vstar

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
If only these graphs weren't so crappy inconsistent. The same curve is very different on two consecutive pages.
Sunny Cove curve is identical, but Willow Cove pencils in a 10% frequency boost. For a company that stayed very tight-lipped about TGL, the way they opened their mouth does seem to lack clarity. I'll refrain from further comments until Sep 2nd.

Untitled-1.jpg
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
Call it coincidence or not, but the second graph labeled "Increased Power Efficiency" is at 3.6GHz, which is the Turbo frequency for one of the leaked -Y variants. Along with slightly improved uarch, it essentially means same performance as the 25W Icelake but at 10-12W, or fanless envelopes.
While voltage is a big part of the story, frequency increase still takes a (linear) toll on power usage. That should be accounted for in this discussion, although the way Intel presented the information invites the audience to ignore it.

The only clear thing about Intel's disclosures is the people in this forum who claimed Sunny Cove was inherently bad at high frequencies (independent of the process node) will have to eat some high crow dosage. We now have confirmation Willow Cove is nearly identical to Sunny Cove, except for L2/L3 cache structure. So the same architecture that barely hit 4Ghz is now aiming for 4.7+ Ghz.