Speculation: Ryzen 3000 series

[CHADBOGA]

10% better perf/clock in ST in CB, and likely more in MT, is not exactly what can be called the same IPC level...

Perhaps he should have said single core performance, rather than IPC, but people have been using those terms interchangeably now for quite a while.

 

[CHADBOGA]

That is all true but don't forget that AMD of today is a moving target. They have Zen 3 ready to intersect Sunny Cove core and Zen 4 one year after that to extend the lead. They are aware of the opportunity that was presented and intel knows that as well. What needs to happen in next two years for intel to regain the competitive edge is :

1) AMD to slip tremendously with Zen3/Zen4 and/or TSMC to fudge the 7nm+/6nm/5nm
2) intel to execute flawlessly with Sunny Cove(+) cores and their own process nodes.

Too many factors need to align so that intel can regain the lead in next 2-3 years. I'm of opinion that this will not happen and intel will bleed heavily their market share the next 2 years.
They are a massive company that can rebound from the slump their are in right now, but we need to be realistic. Behemoths move slowly and take a lot of time to take a swing. When they do they inflict a lot of damage. AMD is now inflicting a death of thousand small cuts to intel trying to bleed them before they can recuperate.

I'd be pleasantly surprised if the gap between Zen 2, Zen 3 & Zen 4 is just 12 months in each instance.

Having said that, Intel seem to be so lost, that it is literally impossible to predict when they will be back on track.

 

[JDG1980]

The funny thing is, if Intel's 10nm process hadn't been such an utter, catastrophic failure, Zen 2 would most have likely fit into the historical pattern since Phenom I (disregarding the Bulldozer era) of AMD's latest and greatest being a generation behind its Intel counterpart in single-thread performance, assuming the early reports about Ice Lake's performance have any substance to them. Possibly it would have just ended up being a repeat of the original Zen's launch, with Intel ahead in gaming and scenarios that required a low-to-mid thread count, and AMD ahead on content creation.

I'm not so sure. Yes, Icelake is probably going to eke out a slight advantage in IPC over Zen 2, but that's only one-half of the picture when it comes to single-thread performance. Intel has been on the 14nm process for so long that they've got it incredibly optimized for high clock speeds - up to 5 GHz from the factory. If they went to an early iteration of 10nm, IPC would go up, but clock speeds would almost certainly have been much lower - maybe not much more than 4 GHz. Even Intel's own charts indicate a clock speed regression compared to 14nm+++ until at least the first refined phase (10nm+). Since the Ryzen 7 3700X boosts to 4.4 GHz, it could conceivably still beat first-generation Intel 10nm Icelake in single-thread performance even if IPC was a bit lower.

 

[StinkyPinky]

I'd be pleasantly surprised if the gap between Zen 2, Zen 3 & Zen 4 is just 12 months in each instance.

Having said that, Intel seem to be so lost, that it is literally impossible to predict when they will be back on track.

Once the smoke has cleared I'd love some detailed breakdown over what the hell has happened at Intel in the last 5 years. They seem to have truly derailed. Of course the desktop space is not their target market these days, but even so it's not a good look for a company of Intel's size.

 

[Hans Gruber]

Once the smoke has cleared I'd love some detailed breakdown over what the hell has happened at Intel in the last 5 years. They seem to have truly derailed. Of course the desktop space is not their target market these days, but even so it's not a good look for a company of Intel's size.

It had been almost 10 years since AMD was competitive with Intel in CPU's. During that time AMD always wanted to compete on price vs. performance. Intel delayed the skylake release years ago because AMD had nothing. There was no need to innovate.

It's the server market for Intel that is really their biggest problem because of the zen technology. Their gravy train may come to an end.

On the bright side of things. This may give Intel needed time to focus on the discrete graphics card market.

 

[maddie]

It had been almost 10 years since AMD was competitive with Intel in CPU's. During that time AMD always wanted to compete on price vs. performance. Intel delayed the skylake release years ago because AMD had nothing. There was no need to innovate.

It's the server market for Intel that is really their biggest problem because of the zen technology. Their gravy train may come to an end.

On the bright side of things. This may give Intel needed time to focus on the discrete graphics card market.

Bright side of things?

In all the confusion happening now, I've been wondering if Raja jumped from the frying pan into the fire.

 

[CHADBOGA]

Once the smoke has cleared I'd love some detailed breakdown over what the hell has happened at Intel in the last 5 years. They seem to have truly derailed. Of course the desktop space is not their target market these days, but even so it's not a good look for a company of Intel's size.

The biggest obvious problem appears to be with their Manufacturing Group and it astonishes me how they let this get so out of hand, but I would be very surprised if there aren't also significant problems in many of their other business units.

I can't even begin to tell you how appalled I feel at such gross mismanagement, so I want to see them really feel some pain for such incompetence and whilst that looks like it will be happening, it is frustrating how long it is taking for that to happen.

Intel have become 1990's IBM.

 

[IntelUser2000]

Once the smoke has cleared I'd love some detailed breakdown over what the hell has happened at Intel in the last 5 years.

Brian Kraznich happened.

The article from CanardPC had great details of what was going on back then. I wouldn't be surprised if development had stalled completely in some departments. They were showing drone presentations at IDF!

 

[exquisitechar]

I'm not so sure. Yes, Icelake is probably going to eke out a slight advantage in IPC over Zen 2, but that's only one-half of the picture when it comes to single-thread performance. Intel has been on the 14nm process for so long that they've got it incredibly optimized for high clock speeds - up to 5 GHz from the factory. If they went to an early iteration of 10nm, IPC would go up, but clock speeds would almost certainly have been much lower - maybe not much more than 4 GHz. Even Intel's own charts indicate a clock speed regression compared to 14nm+++ until at least the first refined phase (10nm+). Since the Ryzen 7 3700X boosts to 4.4 GHz, it could conceivably still beat first-generation Intel 10nm Icelake in single-thread performance even if IPC was a bit lower.

Yeah, some people point to ICL's 18% IPC increase as a sign that Intel will definitely regain the lead once they stop being hamstrung by 10nm, but they forget about clock speeds. Ice Lake is a huge regression in that regard, as the top Comet Lake U SKU will have a 4.9GHz turbo, while Ice Lake will be 3.9GHz (with a 4.1GHz SKU that's apparently MIA), in the same power envelope. A big reason why this is the case is obviously the sad state of 10nm, but a part of it is also probably that Ice Lake itself traded clock speeds for IPC. With 3rd gen Ryzen, AMD did very well to get such a decent clock speed increase with first gen 7nm while increasing IPC, when you consider what others are getting and all the doom and gloom about the lack of frequency increases with new nodes, and even regressions. Moreover, Charlie Demerjian says that Cannon Lake brought an IPC increase too, but that the parts we got were broken and that's why the IPC increase is very small across the board. So ICL is two generations worth of improvements over Skylake, not one.

The important thing for AMD is to keep executing flawlessly with yearly generational improvements, and they will be fine. It seems like they will do exactly that, since Forrest Norrod hinted at Zen 3 matching or exceeding Sunny Cove (?) IPC in his recent interview with Ian Cutress (he said that AMD expected to reach IPC parity or better with Milan...but Intel messed up so they're already there with Rome). 4th gen Ryzen is going to be great, too.

 

[DrMrLordX]

Intel's problem is not architectural design. It's funny that you think 10% (is that an average figure?) upon Skylake architecture is "paltry" when AMD is now getting to Skylake ipc level after how many tries?

About that . . .

10% better perf/clock in ST in CB, and likely more in MT, is not exactly what can be called the same IPC level...

Looks like you beat me to it. Cinebench isn't the only benchmark that appears to be showing Matisse with an IPC advantage over CoffeeLake.

On the bright side of things. This may give Intel needed time to focus on the discrete graphics card market.

And FPGAs, neural processors (Loihi), NICs, network panels, and lots of other stuff. Also, about that IceLake IPC increase . . .

Yeah, some people point to ICL's 18% IPC increase as a sign that Intel will definitely regain the lead once they stop being hamstrung by 10nm, but they forget about clock speeds. Ice Lake is a huge regression in that regard, as the top Comet Lake U SKU will have a 4.9GHz turbo, while Ice Lake will be 3.9GHz (with a 4.1GHz SKU that's apparently MIA), in the same power envelope.

And you beat me to it as well. Granted, 10nm is still a hot mess. We should probably be glad that IceLake-U/Y exists at all (as opposed to the paper-launched Cannonlake).

 

[TheGiant]

About that . . .



Looks like you beat me to it. Cinebench isn't the only benchmark that appears to be showing Matisse with an IPC advantage over CoffeeLake.
.

Cinebench was already abnormal IPC wise for ryzen 2x series, like 3% below coffee lake. So ryzen 3x being like 10% above coffee lake isnt a surprise

But we dont eat cinebench and geekbench.

I am interested in real power/temperature and cooling needed for 8C parts with 10% more perf than coffee lake at 4.4GHz
Boards real compatibility and ocing with 8C parts

This is all getting overhyped. and that never ended well

 

[DrMrLordX]

But we dont eat cinebench and geekbench.

Intel used to crush AMD in every Cinebench. Then Ryzen happened. Now Intel internal memos mock AMD for using Cinebench R15 or R20 as demo benchmarks. Hmm.

 

[TheGiant]

Intel used to crush AMD in every Cinebench. Then Ryzen happened. Now Intel internal memos mock AMD for using Cinebench R15 or R20 as demo benchmarks. Hmm.

It doesnt matter. I wont judge the IPC balance based on we all know is an outlier. I am really looking forward to get a skylake 4,4GHz+10% performance with reasonable real wattage.
Handbrake can use AVX2, lets see what that r3x got...

 

[.vodka]

It doesnt matter. I wont judge the IPC balance based on we all know is an outlier. I am really looking forward to get a skylake 4,4GHz+10% performance with reasonable real wattage.
Handbrake can use AVX2, lets see what that r3x got...

Stock R5 3600, ~49.25 FPS

https://www.reddit.com/r/Amd/comments/c5n8r9

8700k @ 4.7GHz, ~46.76 FPS

My R7 1700 @ 3.9GHz, maxed out, ~ 44.62 FPS

• dram calc tuned timings (14-15-14-14-28 1T)
• 1.7GHz/3400MHz IF/memory
• perfbias AIDA64/GB3 (+3% IPC, Zen+'s tighter cache latency)
• relaxed EDC throttling disabled

Yeah, that's 6 Zen2 cores @ 4.2GHz (unknown memory/IF speed) casually beating 8 maxed out Zen1 cores @ 3.9GHz in SIMD heavy workloads... and beating 6 Skylake cores at a 500MHz deficit. That doubled and improved SIMD hardware on top of the IPC improvements are really paying themselves off.

The higher end models at higher clockspeeds will rip through x264/x265 like they're nothing.

 

[rbk123]

The funny thing is, if Intel's 10nm process hadn't been such an utter, catastrophic failure, Zen 2 would most have likely fit into the historical pattern

I don't understand why people keep pointing to 10nm as something that would keep Intel in the game. All that gets them is lower power consumption as it sure wouldn't be able to clock past 5Ghz. Zen 2 would still be superior, clock for clock, to anything Intel can put on 10nm. They need a next-gen far more than they need 10nm.

 

[exquisitechar]

I don't understand why people keep pointing to 10nm as something that would keep Intel in the game. All that gets them is lower power consumption as it sure wouldn't be able to clock past 5Ghz. Zen 2 would still be superior, clock for clock, to anything Intel can put on 10nm. They need a next-gen far more than they need 10nm.

Clock for clock, Sunny Cove is superior to Zen 2. It's not that a working 10nm alone would do the trick, it's that 10nm being a failure has been holding that "next-gen" back.

 

[realibrad]

I don't understand why people keep pointing to 10nm as something that would keep Intel in the game. All that gets them is lower power consumption as it sure wouldn't be able to clock past 5Ghz. Zen 2 would still be superior, clock for clock, to anything Intel can put on 10nm. They need a next-gen far more than they need 10nm.

What? Why do you think the 3000 chips have more cores? Higher density allows for more transistors in the same area. So Intel could put in more cores. You don't think that would matter? Do you think they would not add more cores?

 

[inf64]

Clock for clock, Sunny Cove is superior to Zen 2. It's not that a working 10nm alone would do the trick, it's that 10nm being a failure has been holding that "next-gen" back.

While I also think that Sunny Cove core should have higher IPC than Zen2 core I also think that difference will not be that big. Zen3 is scheduled to arrive just in time to intersect that core on desktop and server space so AMD is well positioned to counter it with smaller, more power efficient and higher performing Zen iteration.

 

[Abwx]

Clock for clock, Sunny Cove is superior to Zen 2.

Some numbers to support this claim..?.

The 18% stated by Intel is just a marketing gimmick carbon copied from AMD s own 18% from Summit Ridge to Matisse, wich are real, a way to tell everybody that they ll be back to the SR vs SKL ratios.

Now if you look at their slide it s obvious that numbers are stretched thanks to some improved hardware blocks like AES or new instructions like AVX512, granted they can catch up in more legacy apps like Cinema 4D or Blender where they already do well.

 

[exquisitechar]

Some numbers to support this claim..?.

The 18% stated by Intel is just a marketing gimmick carbon copied from AMD s own 18% from Summit Ridge to Matisse, wich are real, a way to tell everybody that they ll be back to the SR vs SKL ratios.

Now if you look at their slide it s obvious that numbers are stretched thanks to some improved hardware blocks like AES or new instructions like AVX512, granted they can catch up in more legacy apps like Cinema 4D or Blender where they already do well.

Intel's claims point to ICL having a bigger increase over SKL than Zen 2 does over Zen/Zen+. Leaked Geekbench results support those claims. The gap between the two is unclear, but I think it's pretty certain that Sunny Cove will have the edge compared to Zen 2, small or not.

As far as I know, AMD never gave an 18% figure. They said 15% in Specint 2006 and 13% in Cinebench, not sure whether that's compared to PR or SR. Intel's 18% is an average that comes from multiple benchmarks, like Cinebench, Geekbench, Spec 2017, and several others.

Your accusations seem unfounded. If anything, I'd sooner believe that AMD changed their slide to point out the 15% Specint 2006 increase instead of the 13% Cinebench increase to look better compared to Sunny Cove. We have older slides that used the smaller Cinebench increase that were changed "last-minute" to use 15%...I also see nothing wrong with Intel including workloads where AVX512 plays a part.

 

[TheGiant]

Some numbers to support this claim..?.

The 18% stated by Intel is just a marketing gimmick carbon copied from AMD s own 18% from Summit Ridge to Matisse, wich are real, a way to tell everybody that they ll be back to the SR vs SKL ratios.

Now if you look at their slide it s obvious that numbers are stretched thanks to some improved hardware blocks like AES or new instructions like AVX512, granted they can catch up in more legacy apps like Cinema 4D or Blender where they already do well.

well we dont have pretty much anything yet

all are leaks

 

[lobz]

Clock for clock, Sunny Cove is superior to Zen 2.

Oh WOW, you must be a hell of an insider to state something like that so confidently...

 

[lobz]

Intel's claims point to ICL having a bigger increase over SKL than Zen 2 does over Zen/Zen+. Leaked Geekbench results support those claims. The gap between the two is unclear, but I think it's pretty certain that Sunny Cove will have the edge compared to Zen 2, small or not.

As far as I know, AMD never gave an 18% figure. They said 15% in Specint 2006 and 13% in Cinebench, not sure whether that's compared to PR or SR. Intel's 18% is an average that comes from multiple benchmarks, like Cinebench, Geekbench, Spec 2017, and several others.

Your accusations seem unfounded. If anything, I'd sooner believe that AMD changed their slide to point out the 15% Specint 2006 increase instead of the 13% Cinebench increase to look better compared to Sunny Cove. We have older slides that used the smaller Cinebench increase that were changed "last-minute" to use 15%...I also see nothing wrong with Intel including workloads where AVX512 plays a part.

Upon reconsideration, you're 100% surely not an insider.

 

[PotatoWithEarsOnSide]

Intel used to crush AMD in every Cinebench. Then Ryzen happened. Now Intel internal memos mock AMD for using Cinebench R15 or R20 as demo benchmarks. Hmm.

To be clear, Intel are complaining to the media about ANY synthetic benchmarks being used.
AMD, to their credit, at E3 ran demos versus an un-gimped 9900k and showed parity in gaming...where traditionally they've been some way behind.
I'm not entirely sure which "real world" benchmark Intel would like to see used, but it is bound to be that undefined one that offers a 40% IPC improvement for Sunny Cove.

Zen/Zen+ weaknesses were commonly known, and AMD has focussed it's attention on these in Zen 2, hence huge AVX improvements, massive IMC improvements, doubling of L3 cache, and an improved front end.

 

[exquisitechar]

Oh WOW, you must be a hell of an insider to state something like that so confidently...

You'll be able to say it confidently too once reviews are out.