Ryzen 3000 Leaks

[Taf the Ghost]

7 minutes ago, GoldenLag said:

considering the Buzz around AMD joining up with Microsoft for future Surface products it wouldnt be all that surprising if Microsoft decided to put some cash down for AMD aswell. how does the Glofo deal work when AMD isnt the one buying the Wafer, but instead is receiving them from someone like Apple or Microsoft for contract manufacturing. 

Well, AMD has to pay GloFo for wafers at other places, but the I/O dies are still at GloFo. So....

[leadeater]

1 minute ago, Taf the Ghost said:

Well, AMD has to pay GloFo for wafers at other places, but the I/O dies are still at GloFo. So....

That whole thing only lasts for another year though right? That requirement as part of the original contract was till 2020 from what I remember, after that point AMD is a free agent.

[GoldenLag]

1 minute ago, Taf the Ghost said:

Well, AMD has to pay GloFo for wafers at other places, but the I/O dies are still at GloFo. So....

in other words: "we dont know the contents of their agreement, so perhaps maybe?"

 

Glofo is still banking on their nice yielding 14nm node. that node is going to be running for a really long time. wouldnt be surprising if 12nm is used when 5nm chiplets are being considered.

 

besides, that node has a lot of life in it as its very nice for smaller companies wanting a better node to work with

[Taf the Ghost]

1 minute ago, leadeater said:

That whole thing only lasts for another year though right? That requirement as part of the original contract was till 2020 from what I remember, after that point AMD is a free agent.

And it's not like AMD doesn't have a massive amount of leverage now. Though they'll keep producing the I/O die on 14nm for a while, I assume. Though that might be what goes over to Samsung, at some point. Samsung's nodes are a little different than TSMC's, so there's some interesting choices for AMD in the future.

[ZestyJalapeno]

It all sounds a bit fairyland tbh, but the 5Ghz boost on TSMC's hp process and from the node shrink looks possible. Intel was already doing that with the 8086k albeit with the TDP hitting the roof on the 14nm node. What is catching my eye is the fact that 8c/16t is Ryzen 5 now.

 

[Taf the Ghost]

Just now, GoldenLag said:

in other words: "we dont know the contents of their agreement, so perhaps maybe?"

 

Glofo is still banking on their nice yielding 14nm node. that node is going to be running for a really long time. wouldnt be surprising if 12nm is used when 5nm chiplets are being considered.

 

besides, that node has a lot of life in it as its very nice for smaller companies wanting a better node to work with

TSMC still has something like 5% of its total revenue in nodes above 90nm. Some nodes will never die.

[Sir Asvald]

9 hours ago, tp95112 said:

Welp I'm getting buyers remorse right now

Same here. Even though I built my system 3 years too late.  

[GoldenLag]

Just now, Dogeystyle said:

What is catching my eye is the fact that 8c/16t is Ryzen 5 now.

well, its in theory 2 half-functioning chiplets. great to increase binning yields. so its not all that surprising. 

[GoldenLag]

2 minutes ago, Taf the Ghost said:

TSMC still has something like 5% of its total revenue in nodes above 90nm. Some nodes will never die.

if it still makes money it is still contributing to the economy while offer workplaces for lower ranking staff to move up. also its just extra cash for what they built the facility for.

[leadeater]

7 minutes ago, Taf the Ghost said:

TSMC still has something like 5% of its total revenue in nodes above 90nm. Some nodes will never die.

Analogue and signaling/comms stuff doesn't scale down in node size so doesn't gain those inherent performance and power gains from node shrinks, and you can slam more power through them without breaking it down.

 

I can't find a good source of information but I don't think MOSFETs in high power audio amps are below 90nm. There's a whole ton of companies in that 180nm-90nm range in that industry area, everyone below 90nm is basically all in the computer industry.

[Taf the Ghost]

3 minutes ago, leadeater said:

Analogue and signaling/comms stuff doesn't scale down in node size so doesn't gain those inherent performance and power gains from node shrinks, and you can slam more power through them without breaking it down.

 

I can't find a good source of information but I don't think MOSFETs in high power audio amps are below 90nm. There's a whole ton of companies in that 180nm-90nm range in that industry area, everyone below 90nm is basically all in the computer industry.

Texas Instruments is still one of the world's largest Fabs, and I'm not sure they've even reached 90nm yet.

[leadeater]

3 minutes ago, Taf the Ghost said:

Texas Instruments is still one of the world's largest Fabs, and I'm not sure they've even reached 90nm yet.

TI does have 45/40 but I don't know what they use it for.

 

https://en.wikichip.org/wiki/technology_node

[Taf the Ghost]

6 minutes ago, leadeater said:

TI does have 45/40 but I don't know what they use it for.

 

https://en.wikichip.org/wiki/technology_node

Nice find. I went looking one time and they really don't list their node types much of anywhere.

[CarlBar]

6 hours ago, Suika said:

Man, if those core counts and frequencies are even remotely true, I'm curious how some X370 and X470 boards will hold up. I'm not even sure I trust the 6-phase vcore VRMs of the X370 Strix to run a 16 core, let alone overclock it.

 

There's apparently some internal debate going on at AMD over that exact thing.

 

5 hours ago, D13H4RD said:

Welp, if those are true, those are killer as hell

 

But best to be a bit realistic. 

 

Hot damn though 

 

Here's the thing people need to get and get quickly. From a pure thermals PoV a 16 core chip with frequencies and power consumption equivalent to the current 2700x is completely 100% viable. The drop in process node size offers that big a jump in power efficiency. We can also use the 7nm Vega to get an idea of the minimum frequency uplift possibble, and that says around 4.1Ghz, but thats not accounting for the binning advantages of chiplets or especially the way the higher price tag would tend to indicate higher quality silicon selection. 

 

Thermals ad power wise the 7nm process allows for upto a 25% bump in frequency if you want to match thermals/power. So again 8 cores at 4.3Ghz base is well within reason.

 

As a final note before i finish up i'd also point out that whilst 5Ghz Boost clock is a rise above and beyond the node itself, we saw a similar thing with Zen+, the boost frequency went up a lot more than the base so i'd not be surprised to see a north of 5Ghz boost be possibble.

 

Now some of you have probably spotted the fly in the ointment. Between frequency uplift and core count uplift the R9's really should be pulling double the power and thus TDP of the 1700X. Which is quite a bit more than what's being claimed, (though certainly manageable, people pushing their 9900K's to 5Ghz all cores have higher TDP's). They could be just mistating their TDP's.

 

The alternative is that AMD is following Intels lead and dropping Hyper Threading. That would eliminate a lot of their per core performance gains, if not all of them, but it costs intel just as much and i'm pretty much convinced at this point that unless Intel goes to a new socket with Comet lake they've got to be dropping HT on the 10 core part. 

[TOMPPIX]

https://wccftech.com/amd-rx-3080-3070-3060-navi-gpu-specs-prices-leaked-rtx-2070-gtx-1070-gtx-1060-challengers-at-249-199-129/

 

this one smells like bullshit, if it's true then rip Nvidia.

[GoldenLag]

8 minutes ago, TOMPPIX said:

this one smells like bullshit, if it's true then rip Nvidia.

You are forgetting something called mindshare ....

[JuztBe]

13 minutes ago, TOMPPIX said:

https://wccftech.com/amd-rx-3080-3070-3060-navi-gpu-specs-prices-leaked-rtx-2070-gtx-1070-gtx-1060-challengers-at-249-199-129/

 

 

this one smells like bullshit, if it's true then rip Nvidia.

If it's true, to hell AMD with that naming convention. I mean it's smart, really smart marketing wise. But it gets confusing real fast.

 

[GoldenLag]

9 minutes ago, VegetableStu said:

wtf now they're shadowing nvidia's naming scheme as well?!

I mean, if you cant market yourself. Make the competitor market for you aswell.

 

If you dont have mindshare, become the mindshare of others.

[CarlBar]

1 minute ago, GoldenLag said:

I mean, if you cant market yourself. Make the competitor market for you aswell.

 

If you dont have mindshare, become the mindshare of others.

 

Become one with the competition :p?

[RejZoR]

4 hours ago, GoldenLag said:

another interesting thing is that AMD should have a 5% IPC advantage if the IPC gains are remotely true.

 

im still a bit skeptical about the clockspeed of these chips, among other things, though it seems to check out

AMD is doing so good now there is no need for them to overhype things and then reveal a poop. They have the confidence of good products so I think all this is actually true. Give or take 100-200MHz depending on how reliable is the info that leakers got from who knows whose butt.

[leadeater]

2 minutes ago, RejZoR said:

AMD is doing so good now there is no need for them to overhype things and then reveal a poop.

Mythbusters did prove you can polish a turd though.

[Mihle]

36 minutes ago, TOMPPIX said:

https://wccftech.com/amd-rx-3080-3070-3060-navi-gpu-specs-prices-leaked-rtx-2070-gtx-1070-gtx-1060-challengers-at-249-199-129/

 

this one smells like bullshit, if it's true then rip Nvidia.

In the past, AMD CPU leaks with Zen have generally be more accurate than the leaks about Vega was. I trust the Ryzen 3000 series leaks more than I trust the RX 3000 series leaks.

[cj09beira]

5 hours ago, Taf the Ghost said:

AM4 doesn't allow for it, but we have an example of 8 cores per Channel on Desktop right now with the TR 2990WX. If you load up all of the cores, you don't get great scaling, but the leech dies also don't have direct memory access. You don't get great scaling beyond 6 cores per Channel, but you do get some.

 

Still, if we called it the Ryzen 3850 Ti, it'd accomplish the same thing. It's a SKU that exists to exist at the top of the stack. It'll be the basis of every test system that gets used in reviews and on & on. AMD is well aware of what models they sell the most of, and it seems pretty clear it's the R5s. Those are going to be the real value monsters in the stack. And it's also why those Comet Lake 10c rumors keep swirling. 

thing is we do have numbers on how it should perform from epyc testing on 4 channel, and its not that bad.

btw did you notice that the difference between navi 12 and navi 10 seems too small, 25% performance increase seems too small. I can see 2 reasons for it, 1 would be navi is still stuck at 64 Cus, but that seems to go against what other leaks have said, the other is that that sku is a cut down version of the core, what are your thoughts on it? 

 

[WelshDdraig]

If these are true - I'm going to ditch my 6800K and move over to the new Ryzen line-up. Those core counts, at those clock speed at those prices? That is too good to be true.
But if it is - Intel is basically f**ked...

[cj09beira]

2 minutes ago, Mihle said:

In the past, AMD CPU leaks with Zen have generally be more accurate than the leaks about Vega was. I trust the Ryzen 3000 series leaks more than I trust the RX 3000 series leaks.

i think the reason for that was that the leaks way before launch were assuming that all features would be working and be useful (which also explains "Poor Volta"), but that time it didn't workout