AMD 12 core 24 thread Zen2 engineering sample benchmark leaked

[GoldenLag]

1 minute ago, Stefan Payne said:

Naa, they are confirmed to use GDDR6.

But yeah, the speculation was that AMD will use the Zen2 CPU Die as much as they can, for selecting the CPUs the best they can.

you never know. china might want something. 

1 minute ago, Stefan Payne said:

Now that the Consoles *might* use the same CPU Chips, AMD might actually be able to do some heavy selection for some higher end, higher clocked Chips like "the other side" does. That is only possible because they have many CPUs to ship and can select the best of the best of the best for the Desktop Higher End Marked.

super high likelyhood AMD will just massproduce this one chip. they have no reason to do anything else. they will simply be masters at binning

[Trixanity]

1 hour ago, GoldenLag said:

actually the answer is closer to maybe. its unlikely, but they can if they want to

Yes and I could run a marathon if I wanted to. Yet it isn't happening. I'd say the likelihood of either are similar.

 

AMD would want a platform change before switching and that's gonna coincide with either Zen 3 or 4 depending on whether they want to be first adopters and go through that mess or wait for maturity and a healthier market.

 

I really have to question the logic behind the "can if they want to" argument. If we applied that across the board we're missing out on a lot of things that they could just do if they really wanted to. I could make you a list as thick as a phone book if it sounds good to you. I mean I could if I wanted to.

[GoldenLag]

2 minutes ago, Trixanity said:

I really have to question the logic behind the "can if they want to" argument.

what i mean by "can if they want to" is not in relation to regular desktop systems, but more like embedded systems aswell as chinese consoles. 

[Trixanity]

1 minute ago, GoldenLag said:

what i mean by "can if they want to" is not in relation to regular desktop systems, but more like embedded systems aswell as chinese consoles. 

Semi-custom work doesn't really count as by its very nature it's made to be specialized in some way meaning if the customer is willing to pay they'll make you whatever you need. It opens up a lot of avenues because the economics and marketing is taken out of the equation. If Sony wants a processor that runs gddr6 instead of DDR4 they'll get it if they're willing to pay for it. Same way they got a Jaguar processor with gddr5 attached it. The key here is it's what the customer pays for and what AMD can deliver. AMD has a portfolio but ultimately it's the customer that decides what it should be.

[PatXioPC]

Well on this topic, I always speculated that AMD won't launch a full 16 core with SMT chip on the consumer-end market because they shouldn't. Only a single memory channel tied to a CCX that packs 8 cores would highly bottleneck the zen engine from the throughput i/o. That's what also happened with the TR 2990wx vs Eypc. 

Now since there's another separate die (ccx) and on the presentation, it had 8 cores; how come with two sums up to only 12? What about the extra space? I speculate - could there be an igpu cluster?

[GoldenLag]

12 minutes ago, PatXioPC said:

Well on this topic, I always speculated that AMD won't launch a full 16 core with SMT chip on the consumer-end market because they shouldn't. Only a single memory channel tied to a CCX that packs 8 cores would highly bottleneck the zen engine from the throughput i/o. That's what also happened with the TR 2990wx vs Eypc. 

except that isnt what happened to the 2990wx. what happened to that was the windows sqeduallerbug. this was uncovered January/december 

 

TL;DW: if it was a memmory bottleneck we would see it in full swing in Linux. also to note that this is leach die and not direct access like Zen 2. so 16 core is very likely. though unshure if they will do a staggered launch

13 minutes ago, PatXioPC said:

Now since there's another separate die (ccx) and on the presentation, it had 8 cores; how come with two sums up to only 12? What about the extra space? I speculate - could there be an igpu cluster?

APU variant is rumoured top be sqedualled Q4 2019 or Q1 2020. traces on the substrate itself point to a second core complex with up to 16 cores. and since memmory bandwhidt isnt much of an isse, that is likely what they are aiming for. 

 

current speculation point to a different IO die for APUs. 

[Trixanity]

19 minutes ago, GoldenLag said:

except that isnt what happened to the 2990wx. what happened to that was the windows sqeduallerbug. this was uncovered January/december 

 

TL;DW: if it was a memmory bottleneck we would see it in full swing in Linux. also to note that this is leach die and not direct access like Zen 2. so 16 core is very likely. though unshure if they will do a staggered launch

APU variant is rumoured top be sqedualled Q4 2019 or Q1 2020. traces on the substrate itself point to a second core complex with up to 16 cores. and since memmory bandwhidt isnt much of an isse, that is likely what they are aiming for. 

 

current speculation point to a different IO die for APUs. 

Ian Cuttress says APUs will remain monolithic (costs, area and complexity probably). I assume he has talked to someone at AMD.

[GoldenLag]

17 minutes ago, Trixanity said:

Ian Cuttress says APUs will remain monolithic (costs, area and complexity probably). I assume he has talked to someone at AMD.

APUs at this moment arent monolithic (if you arent talking about APU consoles like xbox/ps4)

 

cost of the 7nm die is currently small. APUs are also between 10 and 13 months away (estimate based on release sqedual)

 

there is no complexity as the GPU part is connected with PCIe. we allready have Zen APUs using modular design with Vega and Zen 1. Zen 2 APU is that but with a 7nm GPU (presumably Navi)

 

 

Navi will remain monolithic and not be a modular design like first speculated about 1,5-2 years ago. 

 

if the APUs were monolithic they would be more expencive. not only in terms of design but also cost of manufacture. 

[Crunchy Dragon]

On 1/24/2019 at 6:28 PM, Lurick said:

For the record, I think AMD has done an awesome job with Ryzen and hopefully pushes things further with Ryzen 2, I'm just tried of all this smoke being blown about some magic chip that's going to save us from the darks of Intel. I'll wait for benchmarks but I could see it being that AMD might not even need to hit anywhere near a 5GHz boost to give Intel even more run for their money and I hope they keep pushing because I want them to make Intel sweat hard.

Agreed.

 

12c/24t on the consumer socket is already pushing Intel quite hard, even the 9900K is just an 8c/16t chip. Combine core count with presumably a good IPC improvement, and AMD is set to lay on quite a bit of pressure.

[porina]

3 minutes ago, GoldenLag said:

APUs at this moment arent monolithic (if you arent talking about APU consoles like xbox/ps4)

I just had a quick google for an image of a 2200G delid, and it looks monolithic to me. It is a single piece of silicon. Do you understand differently or am I missing something?

 

Also on the bandwidth thing, it will come down to use cases, but we have three factors that are assumed to be increasing: more cores, more clock, more IPC. Offsetting that, more total cache. And the limiting concern, not more ram bandwidth, if we assume a minor supported speed bump as not terribly significant. The FPU upgrade to catch up with Intel is enough to ring alarm bells, and might just be saved with enough cache in the right place.

[GoldenLag]

3 minutes ago, porina said:

I just had a quick google for an image of a 2200G delid, and it looks monolithic to me. It is a single piece of silicon. Do you understand differently or am I missing something?

huh, i just had a doublecheck. i assumed it was 2 dies mostly due to 8 PCIe lanes were missing. wow i feel like an idiot. also due to cost of production. 

 

i still stand firm on Zen 2 APU being 3 different dies. though possibly using a different IO die

[The Benjamins]

10 minutes ago, porina said:

I just had a quick google for an image of a 2200G delid, and it looks monolithic to me. It is a single piece of silicon. Do you understand differently or am I missing something?

 

Also on the bandwidth thing, it will come down to use cases, but we have three factors that are assumed to be increasing: more cores, more clock, more IPC. Offsetting that, more total cache. And the limiting concern, not more ram bandwidth, if we assume a minor supported speed bump as not terribly significant. The FPU upgrade to catch up with Intel is enough to ring alarm bells, and might just be saved with enough cache in the right place.

AFAIK the Chinese console is 2 chips.

EDIT: NVM

[yian88]

SO there will be only 12 core Ryzens with chiplets that failed 8 core tests for threadripper 16 core's, or they will release 8-12 core variants of Ryzen and keep 16 core as reserve against late 2019 against intel, either way intel is fuckked, 8-10 core intel i9 wont keep up with 12 core Ryzen and the price of intel part wont make sense anyway, while AMD will still have the 16 core ACE up their sleeve, remember those "moar cores" meme jokes about AMD in the past? who is laughing now intel?  AMD is

[GoldenLag]

2 minutes ago, yian88 said:

SO there will be only 12 core Ryzens with chiplets that failed 8 core tests for threadripper 16 core's, or they will release 8-12 core variants of Ryzen and keep 16 core as reserve against late 2019 against intel, either way intel is fuckked, 8-10 core intel i9 wont keep up with 12 core Ryzen and the price of intel part wont make sense anyway, while AMD will still have the 16 core ACE up their sleeve, remember those "moar cores" meme jokes about AMD in the past? who is laughing now intel?  AMD is

wait, what sources on staggered launch?

 

like we know its a likely, but it would be a dikk move by AMD imo. depending on yields. they might have a cuckton of 8 core parts. 

[Trixanity]

41 minutes ago, GoldenLag said:

APUs at this moment arent monolithic (if you arent talking about APU consoles like xbox/ps4)

 

cost of the 7nm die is currently small. APUs are also between 10 and 13 months away (estimate based on release sqedual)

 

there is no complexity as the GPU part is connected with PCIe. we allready have Zen APUs using modular design with Vega and Zen 1. Zen 2 APU is that but with a 7nm GPU (presumably Navi)

 

 

Navi will remain monolithic and not be a modular design like first speculated about 1,5-2 years ago. 

 

if the APUs were monolithic they would be more expencive. not only in terms of design but also cost of manufacture. 

Laptops have a limited scope to work with. I don't think a massive package of 2-3 chips to work with is what they want - both in regards to laptop internal design and custom board design. Since they use the same for desktop and mobile it'll color the development for both segments. OEMs are already ruling out AMD due to lack of subsidies and development resources. They'd tell AMD to fuck off if they tried to sell them a 3 chip package.

 

So what they'd want is essentially a similar package to what Raven ridge is now. That would be a reduction in most or all relevant factors. Hence it'll be a single die aka monolithic.

[pas008]

24 minutes ago, yian88 said:

SO there will be only 12 core Ryzens with chiplets that failed 8 core tests for threadripper 16 core's, or they will release 8-12 core variants of Ryzen and keep 16 core as reserve against late 2019 against intel, either way intel is fuckked, 8-10 core intel i9 wont keep up with 12 core Ryzen and the price of intel part wont make sense anyway, while AMD will still have the 16 core ACE up their sleeve, remember those "moar cores" meme jokes about AMD in the past? who is laughing now intel?  AMD is

true but we still dont know what prices these 7nm will be yet, hopefully about the same

[Stefan Payne]

38 minutes ago, GoldenLag said:

i still stand firm on Zen 2 APU being 3 different dies. though possibly using a different IO die

I'm not sure. Depens if AMD wants to make a High End or Lower end APU.

Looking at the APUs right now, I'd rather suspect them to be lower end, similar to the APUs right now -> up to 4 COres and 8 Threads. Though the GPU is pretty strong (704 Cores)

2 hours ago, Trixanity said:

Yes and I could run a marathon if I wanted to. Yet it isn't happening. I'd say the likelihood of either are similar.

As I said, depends. Its not as cut and dry as it was in the past.

It really depends on the situation, the Cost of the Dies and so on. If the Zen3 are available by then and are not more expensive, than its improbable.

But AMD could still use Zen2 dies for whatever reasons (ie they have a couple of Million lying around. In the past they had to throw them away, now they can use them basically forever, until they are gone and are not bound to any socket, Memory or whatever.

Its quasi like CPUs with Chipset on Chip.

 

Something Intel had done with Pentium (2?) Mobile.

 

Right now its comparable with the ASUS P65UP5 thing...

The Motherboard where you could put in Pentium MMX, Pentium PRO or even Pentium 2, depending on the CPU Board you put in. It wasn't bound to one Plattform because the "Chipset" was on the CPU Card...

 

With Zen its the same. The CPU Core Die is independant of the Plattform. The I/O Die is not. That is the thing that has to change and will...

 

Quote

AMD would want a platform change before switching and that's gonna coincide with either Zen 3 or 4 depending on whether they want to be first adopters and go through that mess or wait for maturity and a healthier market.

Yes, the Plattform Change will happen but that doesn't have to correlate with Zen3 or 4, it is independant. That is the Advantage they have.

They can do an I/O Die with whatever Memory they feel like and PCIe Lanes and so on.

 

Also its entirely possible that they save the Zen3 and 4 Dies for the higher end and continue to sell cheaper Zen2 variants to get rid of those Dies. They don't have to throw those away, only the I/O Die...

 

 

1 hour ago, PatXioPC said:

Well on this topic, I always speculated that AMD won't launch a full 16 core with SMT chip on the consumer-end market because they shouldn't.

Why shouldn't they?!

Because you don't want to?
With the 2990WX we know that Memory (Bandwith) isn't the Issue and its enough for 16 Cores on AM4.

 

Quote

Only a single memory channel tied to a CCX that packs 8 cores would highly bottleneck the zen engine from the throughput i/o. That's what also happened with the TR 2990wx vs Eypc. 

No, it did not, see Wendel's Video.

Its the Windows Scheduler that totally craps it up, not the Bandwith. 

Also we're talking about high bandwith like DDR4-3200.

Last gen we had no Problemo with 4 Cores on DDR3-1333...

 

Just look at that:

While DDR4 gives a bit higher perofrmance, its not that much...

So its entirely possible and not as bad as you claim...

 

Quote

Now since there's another separate die (ccx) and on the presentation, it had 8 cores; how come with two sums up to only 12? What about the extra space? I speculate - could there be an igpu cluster?

No, its 2 castrated 8 Core Dies with 6 Cores enabled on each Die.

58 minutes ago, Trixanity said:

Ian Cuttress says APUs will remain monolithic (costs, area and complexity probably). I assume he has talked to someone at AMD.

Yeah, makes sense because it will be price senseitve.

The question is:
What about the CPU? Keep them at 4 Cores or move them to 8??

What about the GPU? Keep them at 11 CU or increase them further?

 

My guess would be a slight increase to 16 CU and probably 8 Cores for the CPU...

 

Or will they stay at 4 Cores and 11CU??
That's also possible as that would mean that they will go down in TDP and AMD looks at high efficiency/low TDP things.

Like for example Tablets.

[GoldenLag]

1 minute ago, Stefan Payne said:

I'm not sure. Depens if AMD wants to make a High End or Lower end APU.

Looking at the APUs right now, I'd rather suspect them to be lower end, similar to the APUs right now -> up to 4 COres and 8 Threads. Though the GPU is pretty strong (704 Cores)

the APUs are actually most interesting when it comes to laptops. while interesting for desktop, that is not where the revenue lies today.

3 minutes ago, Trixanity said:

They'd tell AMD to fuck off if they tried to sell them a 3 chip package.

you know said 3 chip package acts like a single chip. its essentially no different than the 2500u (or at least that would be the pricniple behind it). its not like its magically more complex to omplement. 

 

5 minutes ago, Trixanity said:

I don't think a massive package of 2-3 chips to work with is what they want

its still the size of the AM4 substrate (roughly, perhaps slightly larger). laptop manufacturers have no issues integrating a graphics chip like the mx150. something the mobile APU more or less eliminates (unless they really want something more powerfull then the APU). so size with the exception of the smallest of ultraportables isnt an issue. 

 

7 minutes ago, Trixanity said:

Since they use the same for desktop and mobile it'll color the development for both segments.

they are likely to use the same 7nm core chips for everything. the IO die is what is likely to change between system types. 

 

7 minutes ago, Stefan Payne said:

My guess would be a slight increase to 16 CU and probably 8 Cores for the CPU...

sounds about right. going too large will result in bad idle powerdraw. rx 560 (16 CU) is 123mm^2. the core dies being roughly 78mm^2. idk the areashrink of 7nm from 14nm, but that looks pretty close to me at being parity in size. 

 

currently the 2700u is 210mm^2. which leaves a little room for the IO die, didnt find numbers on the IO die, but they might swap it if they dont need to connect 2 coredies together. 

[Trixanity]

36 minutes ago, Stefan Payne said:

With Zen its the same. The CPU Core Die is independant of the Plattform. The I/O Die is not. That is the thing that has to change and will...

 

Yes, the Plattform Change will happen but that doesn't have to correlate with Zen3 or 4, it is independant. That is the Advantage they have.

They can do an I/O Die with whatever Memory they feel like and PCIe Lanes and so on.

 

Also its entirely possible that they save the Zen3 and 4 Dies for the higher end and continue to sell cheaper Zen2 variants to get rid of those Dies. They don't have to throw those away, only the I/O Die...

You're conflating technical possibilities with reality. Again.

 

Reality is for economic reasons AMD would want to launch a new platform together with new products and to do that they'd want to introduce a compelling processor which would mean performance improvements AKA new uarch.

 

It would also cause an uproar if AMD sold various uarchs on the same platform unless they'd make a very clear distinction which seems unlikely. They'd likely go under the same model name.

 

So your claim seems a fantasy at this point. 

 

Also, if AMD has an abundance of Zen 2 chiplets lying around that haven't been packaged then they've fucked up somewhere.

26 minutes ago, GoldenLag said:

you know said 3 chip package acts like a single chip. its essentially no different than the 2500u (or at least that would be the pricniple behind it). its not like its magically more complex to omplement. 

 

its still the size of the AM4 substrate (roughly, perhaps slightly larger). laptop manufacturers have no issues integrating a graphics chip like the mx150. something the mobile APU more or less eliminates (unless they really want something more powerfull then the APU). so size with the exception of the smallest of ultraportables isnt an issue. 

 

they are likely to use the same 7nm core chips for everything. the IO die is what is likely to change between system types. 

Integrating an MX150 is a trade-off. It adds cost and complexity but you get Nvidia resources and performance benefits. If you haven't noticed Ryzen is dead in the water despite it seemingly being the obvious choice.

 

If AMD says they're not doing APU chiplets then your theory is pretty much dead. It hinges on a GPU chiplet or a combined IO/GPU chiplet and if AMD says it isn't happening this time then I can't tell you anything other than it's monolithic. It's apparently too expensive to package multiple chips for the BoM cost conscious mobile market. I suspect there are many more factors involved but it's hard to say which ones matter the most to each OEM. All I know is they've gone on record saying it's easier and cheaper to work with Intel and AMD throwing any curve balls their way doesn't encourage them any further.

[Taf the Ghost]

2 hours ago, PatXioPC said:

Well on this topic, I always speculated that AMD won't launch a full 16 core with SMT chip on the consumer-end market because they shouldn't. Only a single memory channel tied to a CCX that packs 8 cores would highly bottleneck the zen engine from the throughput i/o. That's what also happened with the TR 2990wx vs Eypc. 

Now since there's another separate die (ccx) and on the presentation, it had 8 cores; how come with two sums up to only 12? What about the extra space? I speculate - could there be an igpu cluster?

There is only a minor bottleneck after 6 cores per Memory Channel. But it's quite minor, and that's at 2666. Pushing up to 3200 or higher will be fine in pretty much everything. And if you need that ~8% better performance per core, you're buying more cores anyway on HEDT, at minimum.

[Taf the Ghost]

1 hour ago, GoldenLag said:

wait, what sources on staggered launch?

 

like we know its a likely, but it would be a dikk move by AMD imo. depending on yields. they might have a cuckton of 8 core parts. 

All of the companies stagger releases, Intel just paper launches half the generation early. The only question is how they're staggering them.

[GoldenLag]

3 minutes ago, Trixanity said:

Integrating an MX150 is a trade-off. It adds cost and complexity but you get Nvidia resources and performance benefits. If you haven't noticed Ryzen is dead in the water despite it seemingly being the obvious choice

i mean, you get Cuda and the 2GB of dedicated Vram. you dont get extra performance (unless the update their mobile GPU linup significantly). or AMD cucks up memmory significantly with APUs (which we know nothing off). 

 

5 minutes ago, Trixanity said:

If AMD says they're not doing APU chiplets then your theory is pretty much dead. It hinges on a GPU chiplet or a combined IO/GPU chiplet and if AMD says it isn't happening this time then I can't tell you anything other than it's monolithic.

i mean, i could be wrong, but it wouldnt make financial sence to make another coredie. it would be roughly 140mm^2 plus IO die which puts it at around 210mm^2. on 14nm that is cheap, but 7nm that is expencive. 

 

7 minutes ago, Trixanity said:

It's apparently too expensive to package multiple chips for the BoM cost conscious mobile market.

AMD will be packaging the entire setup ready for slaughtering. AMD is allready doing multichip setups on each substrate. both consumer and server. and they have been doing it with HBM for a long time (more of a afterthought at that point). 

http://www.cpu-world.com/CPUs/Core_i7/Intel-Core i7 i7-8550U.html

going by these numbers. AMD wont exactly be suffering on the bottomline with the chips speculated earlier in this thread. judging by what these chips are, it wont exactly be budget parts. or at least not the best silicon they can produce. 

11 minutes ago, Trixanity said:

Also, if AMD has an abundance of Zen 2 chiplets lying around that haven't been packaged then they've fucked up somewhere.

considering the multitude of uses the Zen 2 CPU die has, dont think its gonna be much of an issue. also partly because of the glofodeal. 

 

12 minutes ago, Trixanity said:

All I know is they've gone on record saying it's easier and cheaper to work with Intel and AMD throwing any curve balls their way doesn't encourage them any further.

i am no laptop manufacturer. but the 7nm buzzword is gonna be hard to avoid in the laptopspace. and we know intel likes to encourage the use of their chips with subsidies, we saw this especially the last time AMD had competetive chips. 

 

 

we are probably not gonna see to many laptops using AMD chips in 2019, but ive heard speculation Microsoft is slightly interested to implement it into Surface products. the biggest issue being if they are able to buy thunderbolt controllers.

[GoldenLag]

1 minute ago, Taf the Ghost said:

All of the companies stagger releases, Intel just paper launches half the generation early. The only question is how they're staggering them.

as much as AMD wants to create hype through the year. hopefully they release 16 and 12 core first. i know a lot of people just waiting for those exact chips. especially the 16 core ones. 

[Taf the Ghost]

Just now, GoldenLag said:

as much as AMD wants to create hype through the year. hopefully they release 16 and 12 core first. i know a lot of people just waiting for those exact chips. especially the 16 core ones. 

AMD's 50th Anniversary is in May, we have the Gigabyte slides with X570 launching at Computex, so, while things are still up in the air, I'm rather expecting AMD to run some event on May 1st to announce them all.

[Taf the Ghost]

It's also worth remembering that the scuttlebutt from CES is that it's the motherboards that have been the delay. It *is* a PCIe generation change, so some hiccups were always on the table.