News AMD previews Ryzen 3rd generation at CES

[Hitman928]

More info to come. . .

 

[exquisitechar]

Those figures don't make sense for system power...9900k system should be drawing more with that CB score?

 

[ub4ty]

BTW the 9900K in the CB15 test was running at 4.7Ghz, since running within 95W TDP would only yield a CB15 score of around 1760.

Unbelievable that that the 9900k runs at 170Watts+... Thats near the suck down of my friggin 16 core threadripper. Jesus christ.. They literally riced out a civic (ringbus) w/ NoS on that processor ... (Vttaaacchhh turboooo swwwwwuuuufffffftttt)

 

[LTC8K6]

View attachment 2307
Look closely below the top die, what do you see when its titled at this angle?


R.I.P - Shintel

We don't need childish games, though.

They don't make AMD look good to me.

It makes them look childish.

Just put out a good product at a good price and you won't need to play such games.

Let your product do the talking.

 

[firewolfsm]

How is it a childish game? It's not necessarily intentional, it's just there on the package design that there's room for another chiplet. It makes sense from a design perspective to make room for two chiplets from the beginning. This way they won't need to change a thing when a 12/16 core refresh comes out 6/12 months after this 8 core product. I think it gives the Ryzen 3 product more life, and frees up R&D for GPUs and APUs; 4000 series processors might not even change anything else.

 

[Kedas]

Sine it's the 8 core that would be the 3600X 4.0/4.8GHz 95W (according to speculation)
Usng a Ryzen 12 or 16 core compared to 9900k would have made it a lot harder for us to compare.

Since it's with an I/O die the big question will be latency, good enough....

Still a half year before launch for EPYC and Ryzen I assume that means at least one more stepping? I want to build my system end July, so that will be a close call....
I assume threadripper will then also launch almost at the same time.

 

[LTC8K6]

How is it a childish game? It's not necessarily intentional, it's just there on the package design that there's room for another chiplet. It makes sense from a design perspective to make room for two chiplets from the beginning. This way they won't need to change a thing when a 12/16 core refresh comes out 6/12 months after this 8 core product. I think it gives the Ryzen 3 product more life, and frees up R&D for GPUs and APUs; 4000 series processors might not even change anything else.

You mean it doesn't actually have a crack at Intel printed there?

 

[ub4ty]

We don't need childish games, though.

They don't make AMD look good to me.

It makes them look childish.

Just put out a good product at a good price and you won't need to play such games.
Let your product do the talking.

You mean it doesn't actually have a crack at Intel printed there?

wew lad.. I sense much butthurt
Time to step away from the keyboard for a while.
I'm not even going to respond much further.

 

[Kedas]

Are we sure that the Ryzen I/O die is 14nm? i don't think it was mentioned.

 

[Topweasel]

You mean it doesn't actually have a crack at Intel printed there?

No he was talking about the obvious indentations from pin connections in the substrate that line up perfectly for a second die of the exact same size as the top right.

 

[Hitman928]

You mean it doesn't actually have a crack at Intel printed there?

I don't see any cracks about intel in the picture. Maybe you were confused about ub4ty's commentary being on the chip?

 

[DrMrLordX]

Looks like my 1800x and I have an extra 1-2 months together. Was hoping for a firm April launch. Oh well!

Otherwise, very nice. It's good to see AMD take a shot squarely at perf/watt in their CBR15 demo which tells us as much about expected Rome performance as anything else. Would like to have seen some AVX2 as well but what can you do? We'll get more benches later.

Regardless it takes my 1800x over 200W according to HWiNFO64 to pull a CBR15 score of 1815 (best I ever did). I am not worried about the mem latency issue since Zen/Zen+ already had an IF linked between CCXs and the mem controller. That isn't changing with the new chiplet topology. IF speeds should be faster, I hope?

 

[Hitman928]

Are we sure that the Ryzen I/O die is 14nm? i don't think it was mentioned.

Wasn't mentioned but I'd be surprised if it wasn't.

 

[Topweasel]

Are we sure that the Ryzen I/O die is 14nm? i don't think it was mentioned.

It is. AMD stated why they did this tech wise cores work better on a smaller clock, higher clock speed then the uncore stuff. But the reality is even more simple. AMD will be using a GF sourced IO chip till their WSA runs out. 14nm is probably better, more forgiving, and cheaper than 12nm option but maybe next year they will move to that. But it will GF till they don't have to buy anything from them again.

 

[moinmoin]

IF speeds should be faster, I hope?

Bandwidth must be higher due to PCIe 4 alone. The big wildcard is latency though. I hope we get numbers for that before launch.

 

[LTC8K6]

wew lad.. I sense much butthurt
Time to step away from the keyboard for a while.
I'm not even going to respond much further.

Ha ha! You did that well.

But I don't get butthurt.

I still like AMD and will continue to recommend them when they are a good fit for someone's needs.

I remember when some company printed the notes of the Intel theme on their mobo as a crack at Intel.

 

[ub4ty]

Wasn't mentioned but I'd be surprised if it wasn't.

Yep it's 14nm

 

[LTC8K6]

I don't see any cracks about intel in the picture. Maybe you were confused about ub4ty's commentary being on the chip?

I even imagined that I saw it. I've been working too many hours...

 

[DrMrLordX]

Bandwidth must be higher due to PCIe 4 alone. The big wildcard is latency though. I hope we get numbers for that before launch.

If bandwidth goes up and link width stays the same, then latency must go down. Putting the mem controller off-chiplet like that isn't going to impose much extra latency from signal delays.

 

[Topweasel]

Ha ha! You did that well.

But I don't get butthurt.

I still like AMD and will continue to recommend them when they are a good fit for someone's needs.

I remember when some company printed the notes of the Intel theme on their mobo as a crack at Intel.

Honestly I think you are going to have issues with AMD for awhile. TR and EPYC are supposed to be fun names that are a bit childish. Even Ryzen is though less "Bro" like. I mean it goes back father when Athlon or Althon 64 had a gun pointing up from Texas to point to Santa Clara. But I think AMD has gone back to it's early Althon roots, aim high, try hard, have fun, and poke the bear a little bit for a giggle.

 

[ub4ty]

Ha ha! You did that well.

But I don't get butthurt.

I still like AMD and will continue to recommend them when they are a good fit for someone's needs.

I remember when some company printed the notes of the Intel theme on their mobo as a crack at Intel.

K, you've calmed down a little. There was no crack at intel. It's simply an obvious observation from the chip they showed. Intel isn't even a part of the discussion. AMD has had a "platform" since the beginning and they are iteratively progressing it. The new big innovation is the "chiplet" paradigm and an I/O chip. Them introducing that to AM4 which is "new" news signifies the obvious : They intend to put more than one compute chiplet on AM4. This is further realized by the positioning of the first die and then when you see the traces for a 2nd die. Intel was dead to me the moment Ryzen platform was announced... AMD's new design solidifies that again. But this was already known. So, there are no jabs at intel. This is essentially what everyone expected...

 

[ub4ty]

Honestly I think you are going to have issues with AMD for awhile. TR and EPYC are supposed to be fun names that are a bit childish. Even Ryzen is though less "Bro" like. I mean it goes back father when Athlon or Althon 64 had a gun pointing up from Texas to point to Santa Clara. But I think AMD has gone back to it's early Althon roots, aim high, try hard, have fun, and poke the bear a little bit for a giggle.

Good Marketing which makes Intel seem IBM like .. cold, irrelevant, and filled with gray hairs. The young people swoon over AMD partly due to this clever marketing and memes pour out daily. AMD is winning a generation... Intel is stuck in its old ways. Doesn't bode well. If some people knew the 'slang' name for threadripper that young people use, their heads would pop off their shoulders and go into orbit.

 

[LTC8K6]

Honestly I think you are going to have issues with AMD for awhile. TR and EPYC are supposed to be fun names that are a bit childish. Even Ryzen is though less "Bro" like. I mean it goes back father when Athlon or Althon 64 had a gun pointing up from Texas to point to Santa Clara. But I think AMD has gone back to it's early Althon roots, aim high, try hard, have fun, and poke the bear a little bit for a giggle.

I don't have issues with AMD at all.

I frequently suggest AMD CPUs to people.

In fact, I just tried to steer a poster to AMD from Intel.

https://forums.anandtech.com/thread...-11-performance-i7-9700k-vs-i7-7820x.2559440/

 

[Hitman928]

Yep it's 14nm

That's an Epyc slide, the question was about the Ryzen presentation today. However, like I already mentioned, I'd be shocked if it wasn't also on 14 nm.

 

[zinfamous]

killing it (maybe)

It looks like there is room for an extra chiplet, with maybe a special X, X2, XTX, BE, XT-XCX-X version to be released later, after solid binning and once they've gotten IPC locked down. I like sticking with 8 cores and working on clocks well above Zen and Zen+ for the first release of Zen2 Ryzen (3) chips. (that's very confusing, but I think I got it right)

Is there also enough room to cram a 7nm cut Vega die + HBM stack on that thing? Can the IO die be moved around as well? Since I don't know anything about design restrictions for these dies, I was wondering if the chiplet could be rotated 90degrees, the IO moved a bit to the upper left, and then have room to cram a Vega die +1-2 stacks HBM? I'm guessing this isn't really possibly and I am just being stupid, but I like to think this design is fully modular: high-end 8/16 APUs with ~470/480? level performance, or desktop/HEDT CPUS from 12/24 to 16/32.

 

[EXCellR8]

Why not keep your 1700 till Zen 3 atleast?

That could end up happening, depending on what sort of products are being offered at the time and if I feel that I really need the boost in clock speed. The 1700 has been a fantastic chip for my needs so if I were to keep it around longer, I wouldn't mind much at all.