Speculation: Ryzen 3000 series

[TheGiant]

16C 4.2GHz...no thanks
8C 4,5GHz all core turbo with CFL IPC- getting one

 

[moinmoin]

Those leaked numbers (if factual) are from an ES. With AMD those are pretty much always sandbagged.

 

[gorobei]

the zen3 triple and quad thread stuff is pretty staggering if true.

i dont have an issue believing the part about skipping TR on zen2, with ddr5 pcie4 around the corner there is no reason to make a such a niche, limited time product especially if takes away dice from epyc. the original TR was an afterthought from a small group in the marketing team, if it wasnt part of the masterplan then there is no major issue pruning it away as zen3 solidifies.

letting ms foot the bill for early production of zen3 on xbox seems entirely in the ballpark.

still no indication of how active interposer might fit in.

 

[exquisitechar]

I wouldn't worry about the frequency of that 16c ES just yet.

Hope Adored's right about the stuff he's saying in the replies to that tweet, heh.

Some ZEN 3 rumors.

Stacked memory on the IO die in 2020...these YTers need to stop making clickbait, looks like that's all we'll have with all the actual journalists/analysts getting poached by Intel and the like.

 

[coercitiv]

Hope Adored's right about the stuff he's saying in the replies to that tweet, heh.

You know what's funny though? The boost clocks of this ES are almost the same as the base clocks of the AdoredTV leak.

 

[IntelUser2000]

we don't know what the base/boost of the 12c chiplet would be, we don't know if that rumor is at all tied to reality, etc :shrug: But a 4.2 limit? That would be not swell

Look, 4.2GHz stock is fine, if it can overclock higher. Intel basically found a way to ship near-5GHz chips reliably, when it previously was the overclocking headroom. The chips have minimal headroom nowadays.

5GHz is nevertheless not an easy target. I think Intel is only getting there because they got to play with 14nm for a long long time, and they characterized everything they can. If they were on time with 10nm and 14nm never got to see + variants, we might still be at low 4GHz and possibly 5GHz would have been always for the overclockers.

Every CPU ever produced requires water cooling to get 5.5GHz or above, and beyond 6GHz it needs more and more exotic cooling.

Is AdoredTV a reliable source at all? How did they fare previously.

looks like that's all we'll have with all the actual journalists/analysts getting poached by Intel and the like.

Yes this is unfortunate.

 

[exquisitechar]

You know what's funny though? The boost clocks of this ES are almost the same as the base clocks of the AdoredTV leak.

He either must be very confident in what he's hearing about the clock speeds, or he is determined to stick to what he said before even if it all comes crashing down when these things finally launch.

Either way, it'll be entertaining to see what ends up being true at launch!

Look, 4.2GHz stock is fine, if it can overclock higher. Intel basically found a way to ship near-5GHz chips reliably, when it previously was the overclocking headroom. The chips have minimal headroom nowadays.

5GHz is nevertheless not an easy target. I think Intel is only getting there because they got to play with 14nm for a long long time, and they characterized everything they can. If they were on time with 10nm and 14nm never got to see + variants, we might still be at low 4GHz and possibly 5GHz would have been always for the overclockers.

Every CPU ever produced requires water cooling to get 5.5GHz or above, and beyond 6GHz it needs more and more exotic cooling.

Is AdoredTV a reliable source at all? How did they fare previously.



Yes this is unfortunate.

I know for sure that Adored has gotten some things right that would have been almost impossible to guess before they were officially unveiled, but he also spreads a lot of nonsense. His Navi leak makes no sense, for example. The different die configurations, the performance for the suggested CU counts, the SKUs themselves...none of it checks out, really.

 

[IntelUser2000]

I know for sure that Adored has gotten some things right that would have been almost impossible to guess before they were officially unveiled, but he also spreads a lot of nonsense.

Seem like a viable business strategy for the one with little conscience. You need to be correct in some things, but as long as you have that, you use your skills in clickbait to attract viewers.

 

[maddie]

Seem like a viable business strategy for the one with little conscience. You need to be correct in some things, but as long as you have that, you use your skills in clickbait to attract viewers.

Don't you think, for the GPU side at least, he could have many more views if he pushed Nvidia and their dominant position. Someone cheering on Nvidia, to me, would be hugely more popular.

If he is as you say, then he's following the wrong strategy.

 

[Vattila]

[2019 =] 7*7*7*7 - 7*7*7 - 7*7 + 7/.7.

Wow — I got 7 likes for that.

By the way, did you know that the string "Lisa Su" is seven characters long?

 

[gorobei]

I wouldn't worry about the frequency of that 16c ES just yet.

Hope Adored's right about the stuff he's saying in the replies to that tweet, heh.

Stacked memory on the IO die in 2020...these YTers need to stop making clickbait, looks like that's all we'll have with all the actual journalists/analysts getting poached by Intel and the like.

you might want to read up on active interposer butterdoughnut before you scoff.

edit: given the intel has effectively announced the same concept with foveros, your skepticism on this particular point is bizarre.

 

[maddie]

Wow — I got 7 likes for that.

By the way, did you know that the string "Lisa Su" is seven characters long?

Don't be greedy.

 

[exquisitechar]

you might want to read up on active interposer butterdoughnut before you scoff.

edit: given the intel has effectively announced the same concept with foveros, your skepticism on this particular point is bizarre.

I'm skeptical that it's coming with Zen 3, that's all. Intel is ahead of everyone else in this regard. TSMC claims that SoIC will begin production in 2021.

 

[IntelUser2000]

Don't you think, for the GPU side at least, he could have many more views if he pushed Nvidia and their dominant position.

Mmm, maybe, maybe not.

There's a lot of enthusiasm among the AMD side nowadays because Intel was stagnant for so long. The same could be true of the Nvidia side. Perhaps its better to cover AMD news.

Of course, maybe he just likes covering AMD.

 

[Fanatical Meat]

When do we think there will be reliable leaks, I’m thinking end of this month.

 

[inf64]

When do we think there will be reliable leaks, I’m thinking end of this month.

When you see ES benchmarks it's 1 month to release(roughly). Zen2 will be big but Zen3 will be huge.

 

[Ajay]

Zen2 will be big but Zen3 will be huge.

Why? N7 EUVL doesn’t add much to the table.

 

[gorobei]

I'm skeptical that it's coming with Zen 3, that's all. Intel is ahead of everyone else in this regard. TSMC claims that SoIC will begin production in 2021.

ah.
the thing is that all intel has done is announce foveros, they havent put out a product (i dont count kaby-g emib) that indicates they have a lead. amd started the patent for this in 2012 and had the active interposer university study in 2015. Keller joined intel in 2018, this stuff doesnt happen overnight.

 

[StinkyPinky]

Would love it if AMD back their zen 2 cpus so much they hand samples to the press to review well before release. Do it AMD, you know you want to.

 

[Panino Manino]

letting ms foot the bill for early production of zen3 on xbox seems entirely in the ballpark.

Zen 3 in the next Xbox?
Why use a CPU so powerful for a GPU so limited like Navi must be? This makes no sense for me with the state of RTG.

 

[gorobei]

Zen 3 in the next Xbox?
Why use a CPU so powerful for a GPU so limited like Navi must be? This makes no sense for me with the state of RTG.

if you watch the video kapulek posted, they cover the reason: raytracing on cpu (or rather more likely path tracing). with 3x or 4x threads per core smt the cpu has tons of free cycles, path tracing is better done on cpu than gpu and requires memory to cast all the extra rays (with the large caches on the i/o or the stacked ram it may be more efficient than dedicating ray tracing units on gpu silicon)
if you let the navi lite gpu handle the raster stuff and the cpu side handle the pathtraced lighting there may not be a need for such a powerful gpu in the next xbox.

 

[Mockingbird]

So it looks like Jim/AdoredTV is going put his neck on the line

https://twitter.com/x/status/1102955340725977089

"If the fastest SKUs aren't 5GHz+ then it's been bs, simple as that really. I won't even accept 4.9GHz.

Obviously if stuff like R3 being 6-cores and R7 being 12-cores is false, even with 5GHz then you'd have to say that part was just luck. We'll need the whole picture though."

 

[TheGiant]

Adored can shut it down if this doesn't come true

 

[NostaSeronx]

The simple question to ask; has anyone else achieved 5 GHz at a smaller budget than AMD?

TSMC had Cortex A72 at ~4 GHz, but the changes in A76 pushed it down to ~3 GHz. Tachyum has 4 GHz for its Prodigy, it's however a more simple design than x86.
T216 => 2.8 GHz
T432 => 3.2 GHz
T864 => 4 GHz // 180W?
T16128 => 4 GHz // 360W?

They don't have a need for high GHz desktop users. A64FX is pretty much locked at ~2 GHz.

Smaller dies however seem to be worse than bigger dies. With thermals having less horizontal volume to leak into.

 

[Arkaign]

The simple question to ask; has anyone else achieved 5 GHz at a smaller budget than AMD?

TSMC had Cortex A72 at ~4 GHz, but the changes in A76 pushed it down to ~3 GHz. Tachyum has 4 GHz for its Prodigy, it's however a more simple design than x86.
T216 => 2.8 GHz
T432 => 3.2 GHz
T864 => 4 GHz // 180W?
T16128 => 4 GHz // 360W?

They don't have a need for high GHz desktop users. A64FX is pretty much locked at ~2 GHz.

Smaller dies however seem to be worse than bigger dies. With thermals having less horizontal volume to leak into.

Yeah sometimes the move to a new node is pretty disappointing for sure. I'll never forget the Prescott debacle lol.

As for 7nm TSMC, Vega 7nm didn't really seem to fully live up to what you might expect from such a move. The amount of stream processors / CUs basically remained unchanged, and frequency bumped up only somewhat mildly, all at the same basic power consumption (actually a shade higher than Vega64 air). I know comparing GPUs to CPUs may as well be like comparing apple jacks to hand grenades, but we don't exactly have a wealth of other things to look at right now directly comparing a direct architecture 'evolution' on a new node. Vega 64 to Vega 7 didn't reinvent the wheel. Zen 2 is an evolution of Zen/Zen+ as well. So, what will 7nm do for it?

The point about the square mm area of the die for heat dissipation and other such complications arising from cramming more stuff into a smaller area is valid.

I don't expect Intel to be off the hook either in regards to further die shrinks. We just may be in for an era where we are going to see massively diminished returns compared to the glory days of predictably large steps forward.

We also get into the murky depths of what exactly IS "7nm", etc. Because it's not so simple as that single number, a TON of other factors make it far more complex when you read the other variables at play.