Well, I will agree that the core micro-architecture of the CPU cores, between Skylake and Kaby Lake was (allegedly) "the same". That doesn't mean that it's the same exact piece of silicon, just binned differently, because of the media-decoding block additions. It was a different die. (Edit: Therefore, not "the same CPU".)
What is so remarkable is that while Intel have always had issues with clock speeds with die shrinks, AMD haven't.
Kaveri(GF28SHP) and Carrizo(GF28A) are the same FEOL node. With GF28A having a denser BEOL stack than GF28SHP. The clock regression is technically process-related. Kaveri is virtually the "HPC" node with the 13-track library and Carrizo is virtually the "Mobile" node with the 9-track library. The node change is with 28HPA which is Bristol/Stoney. Which had the higher mobility implant.Then moving from 28nm -> 28nm (different node) in going from Kaveri to Carrizo brought even further clockspeed regressions (though those regressions may have been more uarch-related than process-related).
Isn't it a different stepping, implied re-spin/new die. Meaning its binning profile has been improved for the mature process?Also, again, the 9900KS doesn't even have that level of differentiation.
Correct, I should have said AMD have not always had issues. Absolutely there are times they have dropped clocks with process shrinks, but the amazing thing is that with Zen -> Zen+ -> Zen2, as you get smaller the heat density issue become bigger, and it's amazing that they've been able to not only rapidly shrink processes but also increase clocks as they shrink. I think given what has happened with Intel over the last several years that's really commendable.Ehhh not true.
90nm->65nm was a disaster for AMD. I think 65nm was the last process they launched before spinning off GF? No wait it was 45nm. 45nm was relatively successful in comparison. AMD 65nm just plain sucked for them. Then later, the move from 32nm SOI to 28nm brought clockspeed regressions. Then moving from 28nm -> 28nm (different node) in going from Kaveri to Carrizo brought even further clockspeed regressions (though those regressions may have been more uarch-related than process-related). Then moving from either 28nm node to 14nm brought even more clockspeed regressions! AMD basically slid downhill from 5GHz+ to 4-4.1 GHz from 32nm to 14nm. Oh and let's try to ignore GF20nm because it officially reached a clockspeed of 20nm (RIP Nolan, Amur).
What they gained in the process was better density and better power consumption within a desirable clockspeed range. GF14LPP (and 12nm) were so much better power-wise than 32nm or either 28nm node between the 3-3.6 GHz range that it's not even funny. Without those changes, Zen would never have made any headway. Intel experienced similar improvements - better power within a desired clockspeed range - when moving to newer nodes that admittedly brought them lower clocks. 10nm promises more of the same assuming they can actually use it seriously.
Are you giving AMD credit for process shrinks by Glofo and TSMC? I'd be more interested in how AMD fared when they were running their own fabs. AMD has the luxury of going with whichever semiconductor company is doing well that is not Intel, is that an achievement? Not going down with Glofo when there are better options would seem like a no-brainer to me.Correct, I should have said AMD have not always had issues. Absolutely there are times they have dropped clocks with process shrinks, but the amazing thing is that with Zen -> Zen+ -> Zen2, as you get smaller the heat density issue become bigger, and it's amazing that they've been able to not only rapidly shrink processes but also increase clocks as they shrink. I think given what has happened with Intel over the last several years that's really commendable.
No, I'm giving them credit for a product that has increased clock speed, and stable power usage on sequential process shrinks. I'm not interested in how AMD fared when they were running their own fab and I fail to see how that matters to the end result.Are you giving AMD credit for process shrinks by Glofo and TSMC? I'd be more interested in how AMD fared when they were running their own fabs. AMD has the luxury of going with whichever semiconductor company is doing well that is not Intel, is that an achievement? Not going down with Glofo when there are better options would seem like a no-brainer to me.
Oh okay, the process has nothing to do with it. HeheNo, I'm giving them credit for a product that has increased clock speed, and stable power usage on sequential process shrinks. I'm not interested in how AMD fared when they were running their own fab and I fail to see how that matters to the end result.
it's amazing that they've been able to not only rapidly shrink processes but also increase clocks as they shrink
So AMD with their Ryzen line haven't had rapidly shrinking processes with increased clock speeds? I don't get the point you're trying to make, other than trying to catch me in a semantic trap (whatever).Oh okay, the process has nothing to do with it. Hehe
In your last post I quoted, you literally said:
3.8 ghz is low ? and 4.1 ghz is much higher ? Its not even 10% My 1800x would do 3.8 (all core, every day) and my 3900x does 4.1 all core every day. You have quite the knack for overstatement or understatementI think his point is that AMD is getting the silicon from another foundry. In any case, AMD has a much easier time getting equal or faster clockspeeds from die shrinks because they were starting with a quite low clockspeed on the original Zen.
OK, but Intel started at 3.6GHz boost on 32nm, 3.9 on 22nm, and 3.8 on 14nm, and now they're at 5GHz boost on 14nm+^18, so... they started at low boost speeds too.I think his point is that AMD is getting the silicon from another foundry. In any case, AMD has a much easier time getting equal or faster clockspeeds from die shrinks because they were starting with a quite low clockspeed on the original Zen.
Claiming that Zen started from slow clockspeeds is quite dubious IMO.OK, but Intel started at 3.6GHz boost on 32nm, 3.9 on 22nm, and 3.8 on 14nm, and now they're at 5GHz boost on 14nm+^18, so... they started at low boost speeds too.
Talking about all core turbo/overclock. Considering that intel can do 5 ghz all core, *yes* low 4ghz all core that early zen could do (on a good sample, if I recall correctly some had to struggle to even reach 4.0 all core) is low. As for overstatement or understatement, I challenge you to find one statement of mine that is clearly false. You may not like what I say, or resent a contrasting viewpoint or context, but that is not my problem.3.8 ghz is low ? and 4.1 ghz is much higher ? Its not even 10% My 1800x would do 3.8 (all core, every day) and my 3900x does 4.1 all core every day. You have quite the knack for overstatement or understatement
Did you read my reply ? both 3.8 overclocked on the 1800x and 4.1 ghz overclocked on the 3900x are all-core overclocked. Both of those are not the highest overclock, but since I run them@100% all day every day, I like temps in the 70's. As for the overstatement/understatement, it because you are comparing Intel to AMD. You seem to think that GHZ is everything, and its not. In some thinks, a Ryzen 8 core can beat an Intel 8 core at lower clocks. You seem to think everything revolves around GHZ, and it does NOT, and thats fact, not opinion.Talking about all core turbo/overclock. Considering that intel can do 5 ghz all core, *yes* low 4ghz all core that early zen could do (on a good sample, if I recall correctly some had to struggle to even reach 4.0 all core) is low. As for overstatement or understatement, I challenge you to find one statement of mine that is clearly false. You may not like what I say, or resent a contrasting viewpoint or context, but that is not my problem.
You seem to think that GHZ is everything, and its not.
OK, but Intel started at 3.6GHz boost on 32nm, 3.9 on 22nm, and 3.8 on 14nm, and now they're at 5GHz boost on 14nm+^18, so... they started at low boost speeds too.
I didn't get this from his posts. Zen 3000 series might have a slight IPC advantage over the Intel 9000 series, but it is slight, where the higher clocks of Intel edge out AMD for ultimate performance.
Maybe... But a 3700x 8 core (85 watts) vs a 9900ks 8 core (142 watts) 57% more power ???? What is incredible about that ? That they are so handicapped on power ?
Since you did not provide a link, I am going to go look for it.
Edit: from their conclusion:
Intel has extended its lead in gaming over AMD's lineup, but you have to keep that in perspective. The Ryzen 9 3900X offers a lower price point and is more agile in heavily-threaded workloads, particularly in productivity applications. It's also comes with other advantages, like support for PCIe 4.0. If you're not chasing the bleeding edge of gaming performance or overclockability, the Ryzen 9 3900X still offers a compelling blend of price and performance in both gaming and productivity applications.
So, as we already knew it wins in games by a little, but gets destoyed at anything else at the same price or more. And usess 57A% more power (and heat)
So why again is this good ?
Thank you for actually reading my post and making a reasonable response instead of putting words into my mouth and apparently thinking you can read my mind.I didn't get this from his posts. Zen 3000 series might have a slight IPC advantage over the Intel 9000 series, but it is slight, where the higher clocks of Intel edge out AMD for ultimate performance.
From a value stand-point, AMD is the clear winner. That is, it performs almost as fast as Intel (core for core) effectively (IPC * clock speed) and is priced competitively, not to mention the power consumption advantage. THAT said, the Intel 9900KS is still the fastest 8 core chip out there due to the all-core 5Ghz turbo, stock.
I know these forums have often been involved in brand wars (even more so with the GPU side), but I think this remains factual.
Intel is still the absolute performance king for mainstream at the cost of double power consumption and more $$$. This usually means that Intel is usually not the best choice for most people with a budget.
But, then there are people like me... I just want a 9900KS. I guess the idea of all turbo @ 5Ghz appeals to me. Would I notice the difference? Not likely. Is it worth the price increase? No, it isn't. But will I probably get one? Yep. To each their own.
I was REALLY hoping that Zen could hit 5Ghz clocks, as I would have went AMD on my next build.
You seem to think that GHZ is everything, and its not. In some thinks, a Ryzen 8 core can beat an Intel 8 core at lower clocks. You seem to think everything revolves around GHZ, and it does NOT, and thats fact, not opinion.
No, I am trying to dispel the rumor that clocks rule, as some posters seem to maintain. This is NOT off-topic or non-sequitur as you say. Some are talking about ghz, and some about process shrinks. While this is a 9900ks preview, the comparison to AMD is not really OT.This is an odd reply. The discussion taking place has been about GHZ (i.e., increasing frequency with process shrinks), not overall performance. Your post is a bit of a non-sequitur given that context.
That is not always true.
No, I am trying to dispel the rumor that clocks rule, as some posters seem to maintain.