AMD Ryzen "Picasso" APU Clock Speeds Revealed

[btarunr]

AMD is giving finishing touches to its Ryzen 3000 "Picasso" family of APUs, and Thai PC enthusiast TUM_APISAK has details on their CPU clock speeds. The Ryzen 3 3200G comes with 3.60 GHz nominal clock-speed and 4.00 GHz maximum Precision Boost frequency; while the Ryzen 5(?) 3400G ships with 3.70 GHz clock speeds along with 4.20 GHz max Precision Boost. The "Picasso" silicon is an optical shrink of the 14 nm "Raven Ridge" silicon to the 12 nm FinFET process at GlobalFoundries, the same one on which AMD builds "Pinnacle Ridge" and "Polaris 30."

Besides the shrink to 12 nm, "Picasso" features upgraded "Zen+" CPU cores that have improved Precision Boost algorithm and faster on-die caches, which contribute to a roughly 3% increase in IPC on "Pinnacle Ridge," but significantly improved multi-threaded performance compared to 1st generation Ryzen. Clock speeds of both the CPU cores and the integrated "Vega" iGPU are expected to increase. Both the 3200G and 3400G see a 100 MHz increase in nominal clock-speed, and 300 MHz increase in boost clocks, over the chips they succeed, the 2200G and 2400G, respectively. The iGPU is rumored to receive a similar 100-200 MHz increase in engine clock.



View at TechPowerUp Main Site

 

[Assimilator]

Wait... so the 2x00G are actually Ryzen 1, and 3000G are Ryzen+, so the Ryzen APUs are always a generation behind?

 

[TheLostSwede]

Wait... so the 2x00G are actually Ryzen 1, and 3000G are Ryzen+, so the Ryzen APUs are always a generation behind?

So far, yes.

Also, do we know these are the final clocks?

Right now, it seems like AMD has seeded CPUs/APUs with different clock speeds to different customers/partners, so don't take anything for granted for the time being.

 

[Caring1]

Sounds good to me.

 

[GoldenX]

Boo, it was known that 3000 APUs were going to be Zen+, and with that you can guess that they will go over 4GHz.
I want to know about the IGP specs. Do we get Navi or is it still Polaris? Will they raise the amount of cores or will it still stay at 11 max?

 

[FYFI13]

Not a single word if CU count will increase? If it's 100-200MHz increase across the board only, then I'm really disappointed.

 

[TheLostSwede]

Not a single word if CU count will increase? If it's 100-200MHz increase across the board only, then I'm really disappointed.

Because you were planning on getting one? Or because it doesn't meet your requirement for a system you weren't going to build anyhow?

 

[FYFI13]

Because you were planning on getting one? Or because it doesn't meet your requirement for a system you weren't going to build anyhow?

I'm planning on getting few of these, a good few.

 

[Zubasa]

Boo, it was known that 3000 APUs were going to be Zen+, and with that you can guess that they will go over 4GHz.
I want to know about the IGP specs. Do we get Navi or is it still Polaris? Will they raise the amount of cores or will it still stay at 11 max?

They are still "Vega".
Also increasing the amount of GPU cores wouldn't help much if the GPU is limited by the system ram bandwidth anyway.

 

[Chomiq]

Every time I see this headline I read "Fiasco" instead of Picasso. Something's wrong with my vision.

 

[Valantar]

Wait... so the 2x00G are actually Ryzen 1, and 3000G are Ryzen+, so the Ryzen APUs are always a generation behind?

Yes, and they have always been. Raven Ridge was Zen (1), albeit with some of Zen+'s features added in (the ones that were ready early and could be implemented easily). Picasso is full Zen+. The die was launched at CES and is already available in mobile.

Boo, it was known that 3000 APUs were going to be Zen+, and with that you can guess that they will go over 4GHz.
I want to know about the IGP specs. Do we get Navi or is it still Polaris? Will they raise the amount of cores or will it still stay at 11 max?

Not a single word if CU count will increase? If it's 100-200MHz increase across the board only, then I'm really disappointed.

It's Vega - the die is already being sold for mobile, and AMD has been very clear on Picasso being a 12nm update of Raven Ridge with the full Zen+ feature set (rather than the subset supported by RR). AMD has also confirmed unequivocally that there won't be MCM APUs with Matisse - we'll have to wait for Ryzen 4000 for that. This is the same 11 CUs, same Vega architecture, but a slight clock bump at the same power. And if rumors hold true, a soldered IHS.

 

[Imsochobo]

They are still "Vega".
Also increasing the amount of GPU core's wouldn't help much if the GPU is limited by the system ram bandwidth anyway.

Vega 10 to 12 would make sense, but I do not think much past that.
As memory compatibility improves its easier to feed the gpu.
3466 was only b die but now hynix and micron kits manage it easily at cl16 even in high densities.

 

[Vya Domus]

Not a single word if CU count will increase?

It wont. No point in increasing the core count if these things are already severely memory bandwidth bound.

 

[Darmok N Jalad]

Wait... so the 2x00G are actually Ryzen 1, and 3000G are Ryzen+, so the Ryzen APUs are always a generation behind?

Like Valantar said above, RR was actually somewhere between Zen and Zen+. 14nm, but had better boost.

Not a single word if CU count will increase? If it's 100-200MHz increase across the board only, then I'm really disappointed.

You get better clocks and also an IPC increase. Also consider that with a smaller node, these chips will probably run cooler and stay boosted longer. I bet AMD could also increase the clocks on the Vega IGP as well, as even the old 2200G/2400G could OC another 75-100MHz.

 

[Apocalypsee]

It wont. No point in increasing the core count if these things are already severely memory bandwidth bound.

+1 I'm hoping it have the 1/2 divider on this APU so it would help the iGPU massively, but since it based on Zen+ I doubt it have the option.

 

[R0H1T]

Wait... so the 2x00G are actually Ryzen 1, and 3000G are Ryzen+, so the Ryzen APUs are always a generation behind?

Not exactly, zen 2xxx are slightly tweaked zen cores. They have better latency for instance, the "12nm" shrink improves their power consumption aside from bringing in some of the zen+ improvements. As other have said ~ somewhere between zen & zen+ also AMD APUs are always lagging their standalone products for some reason.

 

[catulitechup]

Meh

Waiting for 7nm APUs

 

[Deleted member 67555]

Wait... so the 2x00G are actually Ryzen 1, and 3000G are Ryzen+, so the Ryzen APUs are always a generation behind?

The first AM4 APU'S were the same as the last generation from AM3+...7870k.
The second was 1st generation Ryzen and was released just prior to the second generation of Ryzen...

I don't think the iGPU has really gotten all that much better... The only thing that seems to really progress is the CPU cores.

Maybe they'll really take off with DDR5

 

[Caqde]

AMD has also confirmed unequivocally that there won't be MCM APUs with Matisse - we'll have to wait for Ryzen 4000 for that.

Matisse is the keyword there. As Matisse is the Zen2 based AM4 CPU codename. Whereas the other codenames are Castlepeak for the Zen 2 based TR4 chips and Renoir for the Zen2/Navi based AMD APU's for AM4 and Notebooks. So AMD could still release Renoir based APU's for the Ryzen 3000 series there won't be any under the Matisse codename. But personally I'm not expecting them to release Renoir until next year and I'd be surprised if they release them as part of the Ryzen 3000 (Desktop) series although it would be a good move if they do.
EDIT:

Wait... so the 2x00G are actually Ryzen 1, and 3000G are Ryzen+, so the Ryzen APUs are always a generation behind?

Yeah so far. The chiplet design they have started using might help them with that issue as they will no longer need to design a seperate AMD APU chip instead with the Zen 2 based APU's we are likely to see AMD have three chips on their AM4 and Notebook APU's. Those being the I/O chip, Zen 2 chip, and a Navi chip. Thanks the this AMD will only need to have the seperate chips ready to make an APU instead of a single monolithic design that incorporates all of those chip designs into one.

 

[TheLaughingMan]

The APUs will always trail behind because AMD doesn't want people to buy them over their desktop parts to save money. If they stay 1 gen behind though, there is enough performance difference for their lower end chips to still have some market presence without eating each other's segment up....for the most part (RIP Ryzen 1200 to 1400 and all their successors).

 

[R0H1T]

The APUs will always trail behind because AMD doesn't want people to buy them over their desktop parts to save money.

Which doesn't make sense because laptops far outsell anything they make on the desktop side. I know this isn't exactly what you said, but this backwards thinking or strategy has let them down in a major way in the OEM space especially notebooks & ULV based systems.

 

[Valantar]

Matisse is the keyword there. As Matisse is the Zen2 based AM4 CPU codename. Whereas the other codenames are Castlepeak for the Zen 2 based TR4 chips and Renoir for the Zen2/Navi based AMD APU's for AM4 and Notebooks. So AMD could still release Renoir based APU's for the Ryzen 3000 series there won't be any under the Matisse codename. But personally I'm not expecting them to release Renoir until next year and I'd be surprised if they release them as part of the Ryzen 3000 (Desktop) series although it would be a good move if they do.

I know this is nitpicking, but Matisse is the code name for first-gen MCM AM4 packages, not a CPU - after all, the CPU chiplets are common across Matisse Rome and whatever 3000-series TR is called. In other words there'd be "room" for an MCM APU within the Matisse code name without breaking with anything significant (though AMD could of course name this something else if they wanted to, given that these are internal code names and their meaning is entirely determined by AMD). Still, that statement alongside the launch of 3000-series 12nm Zen+ APUs is pretty clear evidence that MCM APUs aren't coming to AM4 this go around. Luckily APUs are on a staggered launch cadence compared to CPUs (CES this year, and a few months before the 2000-series last year), so I'm hoping we'll see the first MCM APUs at CES 2020 and that the desktop rollout will be a bit faster than now.

Yeah so far. The chiplet design they have started using might help them with that issue as they will no longer need to design a seperate AMD APU chip instead with the Zen 2 based APU's we are likely to see AMD have three chips on their AM4 and Notebook APU's. Those being the I/O chip, Zen 2 chip, and a Navi chip. Thanks the this AMD will only need to have the seperate chips ready to make an APU instead of a single monolithic design that incorporates all of those chip designs into one.

That's very likely the plan. The issue is that MCM packaging is a lot more expensive than traditional packaging, and laptops (where most APUs go) are a very low margin market. Also, IF uses quite a lot of power (my partner's TR 1920X uses 70-80W for its uncore while under a full CPU load), which they'd need to reduce significantly for a mobile application. Of course there'd be far fewer IF links in an APU than a TR CPU, and they do claim that the Matisse/Rome generation of IF is more efficient, but I don't think the stars of cost and power have quite aligned yet. Hopefully they will the next go around, as I really want a 6-8c 20CU APU for my HTPC.

 

[ironwolf]

Every time I see this headline I read "Fiasco" instead of Picasso. Something's wrong with my vision.

Don't feel bad, I read "Infection" instead of "Inflection" on that roadmap picture.

 

[danbert2000]

This is a great bump for AMD APUs. You're basically getting a pre-Kaby Lake i7 with a GTX 1030 tacked on. I do wonder if they're going to push the GPU speeds up and keep the overclock headroom. If this iGPU could approach 1700 MHz, with fast RAM I could see it flirting with GTX 1050 performance, which would be plenty for the entry level gamer. And the increase in CPU clockspeeds will make the chip less of a bottleneck if a discrete GPU is added later. That was my biggest issue with the 2400G. People were buying them and putting them in systems with GPUs and leaving a lot of performance behind because of the sub-4 GHz clocks.

As you can tell from my specs, I have a i7-5775c. It is very similar to these Ryzen APUs, with 4 cores/8 threads and a beefier than usual iGPU. The Vega 11 GPU part in the Ryzen is 50% faster, but the CPU part of my i7 is 20-30% faster. When these come out, the 3400G will trail my CPU by only a little bit and provide maybe 70% better GPU performance. AMD took 4 years to get there but they are finally on par with the best APU Intel has ever put out, and at a considerably cheaper price.

 

[GoldenX]

They are still "Vega".
Also increasing the amount of GPU core's wouldn't help much if the GPU is limited by the system ram bandwidth anyway.

Yes, and they have always been. Raven Ridge was Zen (1), albeit with some of Zen+'s features added in (the ones that were ready early and could be implemented easily). Picasso is full Zen+. The die was launched at CES and is already available in mobile.


It's Vega - the die is already being sold for mobile, and AMD has been very clear on Picasso being a 12nm update of Raven Ridge with the full Zen+ feature set (rather than the subset supported by RR). AMD has also confirmed unequivocally that there won't be MCM APUs with Matisse - we'll have to wait for Ryzen 4000 for that. This is the same 11 CUs, same Vega architecture, but a slight clock bump at the same power. And if rumors hold true, a soldered IHS.

Yeah, I mean Vega instead of Polaris. Anyway, seems like the next APUs are not very exciting.