AMD Raven Ridge 'Zen APU' Thread

[Ajay]

Raven's CCX design always has 4MB L3.

Zeppelin = 8MB per CCX.
Raven = 4MB per CCX.

It's a huge die even without the additional 4MB of L3.

Do you know the die size?

 

[The Stilt]

Do you know the die size?

Yes?

 

[AtenRa]

Do you know the die size?

Im estimating 200mm2 to 240mm2

Closer to 240 than 200

 

[Ajay]

Yes?

lol!

 

[Ajay]

Im estimating 200mm2 to 240mm2

Closer to 240 than 200

Larger than Zeppelin! Integrated PCH + very large iGPU?

 

[The Stilt]

TN/KV/CZ ish, on half the node size.

 

[Asterox]

Or maybe full fat 8c/16t? That would be interesting....

Yes only in your dreams, 8/16 Raven Ridge APU with 2000 VEGA GCN GPU cores.

If you need "Mobile 8/16 CPU", well Asus has a very good gaming tool for your needs.

 

[LTC8K6]

Yes only in your dreams, 8/16 Raven Ridge APU with 2000 VEGA GCN GPU cores.

If you need "Mobile 8/16 CPU", well Asus has a very good gaming tool for you needs.

65 watt cpu laptop...price, battery life, does it throttle, etc.?

 

[crashtech]

It has to be AM4, right, or is there a mobile socket for Ryzen 7 I don't know about?

 

[cbn]

It has to be AM4, right, or is there a mobile socket for Ryzen 7 I don't know about?

Its AM4, but the dGPU (RX580) is a mobile design.

Pretty amazing in that it has 154W worth of processosing in the form of a CPU (65W) and dGPU (est. 89W).

 

[crashtech]

Its AM4, but the dGPU (RX580) is a mobile design.

Pretty amazing in that it has 154W worth of processosing in the form of a CPU (65W) and dGPU (est. 89W).

Yeah, I'm not sure how they are going to fit the battery in there:

 

[Topweasel]

Do you know the size of it, since you say it isn't huge?

I don't know if you are trying to be difficult for difficulty sake but sure. The die size is roughly a 192mm, The CCX's take up about half of that. So 96mm. Then you look at the L3 Cache. It's about 2/3 the size the CPU's. Which makes it something close to 38mm for the L3.. Cutting that in half would save you roughly 19mm for the two CCX's. Cutting that in half for 4MB would make it roughly 9.5mm^2, on a 200mm die, saved. It's not very scientific and its probably off but look at the die and see for yourself. It's not insignificant but it's also not that large. Not worth redesigning the CCX on top of developing the new die. Me I would call 5% a small amount when talking about how much not having to change the CCX would save.

 

[The Stilt]

I don't know if you are trying to be difficult for difficulty sake but sure. The die size is roughly a 192mm, The CCX's take up about half of that. So 96mm. Then you look at the L3 Cache. It's about 2/3 the size the CPU's. Which makes it something close to 38mm for the L3.. Cutting that in half would save you roughly 19mm for the two CCX's. Cutting that in half for 4MB would make it roughly 9.5mm^2, on a 200mm die, saved. It's not very scientific and its probably off but look at the die and see for yourself. It's not insignificant but it's also not that large. Not worth redesigning the CCX on top of developing the new die. Me I would call 5% a small amount when talking about how much not having to change the CCX would save.

TN/RL = 240.41mm² (32nm)
KV/GV = 244.29mm² (28nm)
CZ/BR = 246.39mm² (28nm)

I already said it's TN/KV/CZ ish in size

 

[AtenRa]

Larger than Zeppelin! Integrated PCH + very large iGPU?

Yes the PCH + iGPU will take a lot of space on the die. Im expectin the iGPU alone to occupy close to 60% of the die (perhaps even more).

 

[coercitiv]

Yeah, I'm not sure how they are going to fit the battery in there:

Don't worry about the battery... worry about the cooling (noise) : i see a lot of "heat pipe" but very little "heat sink".

 

[Topweasel]

TN/RL = 240.41mm² (32nm)
KV/GV = 244.29mm² (28nm)
CZ/BR = 246.39mm² (28nm)

I already said it's TN/KV/CZ ish in size

I thought you were talking about the cache. The 4MB would be about 4% of those die sizes.

 

[nathanddrews]

In its mobile form, I don't expect heat to be a big issue. Ryzen is pretty damned cool/quiet closer to 2GHz, where most mainstream laptops are clocked. With 4C/8T and superior graphics (to Intel), I'm not sure how it could possibly fail. Unless we get single-channel 2133 DDR4 OEM laptops for $899... then it will fail. LOL

 

[Bouowmx]

Current "mainstream" laptops are not clocked close to 2 GHz. Intel Core i5-7200U: 3.1 GHz. Intel Core i5-8250U: 3.4 GHz (sustained: no good data so far). It can fail if it is not given enough power; please no 15 W.

 

[ao_ika_red]

The other reason why AMD use lesser L3 cache is to not directly threaten their current sku (R5 1400).
IIRC, AMD once explained that part of CCX's L3 cache are not connected directly to L2 cache. If AMD manage to cut the less important 4MB cache, I think the performance impact won't be severe enough to notice.

 

[Topweasel]

The other reason why AMD use lesser L3 cache is to not directly threaten their current sku (R5 1400).
IIRC, AMD once explained that part of CCX's L3 cache are not connected directly to L2 cache. If AMD manage to cut the less important 4MB cache, I think the performance impact won't be severe enough to notice.

I think when people talk of the APU's they think of them as the Pentium or Celeron of AMD's lineup. I don't think AMD sees it that way. It's not targeted at that market. Its targeted a a market that isn't impacted by core count that sees value in discreet level GPU in a single package market.

For that reason and many more I doubt AMD cares if it canabalizes the R3, less R3' sold the less good CPU's they have to gimp. On top of that I am guessing a majority of the lineup is between $100-$200.

They desktop versions are only a byproduct of already having the dies and some of the market for what I am talking about being on desktops. But just because it will clock lower or have less resources than an R3 on desktop (it'll be more like a hybrid since many will have SMT). Doesn't mean it will be sold and viewed as a lesser product. Otherwise they would be using a die as large or larger than Zeppelin which they are selling at upwards of $450 per die.

But assuming the retail L3 cache is 4MB. The question is, do the dies still have 8MB in them. I think they do. Even though it's a new die I challenge that they would possibly upset the balance in the CCX's by redesigning the complex for a 4% die savings.

 

[nathanddrews]

Current "mainstream" laptops are not clocked close to 2 GHz. Intel Core i5-7200U: 3.1 GHz. Intel Core i5-8250U: 3.4 GHz (sustained: no good data so far). It can fail if it is not given enough power; please no 15 W.

I've got an i5-7200U laptop and it's only 2.5GHz, but can turbo higher if that's what you mean. For most desktop tasks, it runs much slower than 2.5GHz. I was only referring to base clocks with my 2GHz comment, which is a bit low. Most 4C/8T Intel mobile chips have base clocks between 2.2-2.7GHz, with turbo/single core clocks over 3GHz. If AMD's highest tier Ryzen 4C/8T runs at a 2.2GHz base and 3.xGHz turbo/XFR/whatever combined with its better graphics, then that's going to be impressive if it undercuts Intel prices the way desktop Ryzen did.

 

[AtenRa]

I've got an i5-7200U laptop and it's only 2.5GHz, but can turbo higher if that's what you mean. For most desktop tasks, it runs much slower than 2.5GHz. I was only referring to base clocks with my 2GHz comment, which is a bit low. Most 4C/8T Intel mobile chips have base clocks between 2.2-2.7GHz, with turbo/single core clocks over 3GHz. If AMD's highest tier Ryzen 4C/8T runs at a 2.2GHz base and 3.xGHz turbo/XFR/whatever combined with its better graphics, then that's going to be impressive if it undercuts Intel prices the way desktop Ryzen did.

I dont believe those Mobile APUs will be cheap, they will directly compete against Intel Gen 8 chips. RR mobile CPU performance will be very close to Intel but it will have more than 50% higher iGPU performance than HD620 found on Core i5 8350U and Core i7 8650U.

 

[Bouowmx]

For most desktop tasks, it runs much slower than 2.5GHz.

That's not how Intel Turbo Boost works, unless cTDP was configured down to 7.5 W, but that's a discussion for a different thread.

 

[raghu78]

Raven's CCX design always has 4MB L3.

Zeppelin = 8MB per CCX.
Raven = 4MB per CCX.

It's a huge die even without the additional 4MB of L3.

With 8MB L3 taking up just 16 sq mm I don't think AMD cut L3 cache by half for Raven Ridge. If they did so its not for area benefit but probably for power benefit. AMD might have weighed the perf loss and power reduction and deemed the tradeoff worthwhile.

https://www.pcper.com/news/Processors/AMD-Details-Zen-ISCCC

The other reason I think Raven Ridge has 8MB L3 cache is due to the leaked slides showing Great Horned Owl which is embedded version of Raven Ridge with 8MB L3. There is a possibility that AMD is using partially defective dies for the 15w Raven Ridge chips. Raven Ridge 15W seems to have only 8CU enabled according to few rumours. So it would make sense if AMD is using 4C/8T with 4MB L3 and 8CU for 15w SKUs and 4C/8T with 8MB L3 and 11 CU for 35-45w standard notebook parts.

https://videocardz.com/69428/amd-snowy-owl-naples-starship-grey-hawk-river-hawk-great-horned-owl

TN/KV/CZ ish, on half the node size.

Raven Ridge seems to be around 210 - 220 sq mm. I doubt its as big as 240 sq mm.

http://www.bitsandchips.it/52-english-news/7622-rumor-two-versions-of-raven-ridge-under-development
https://techreport.com/forums/viewtopic.php?t=119682

 

[Arachnotronic]

Raven Ridge seems to be around 210 - 220 sq mm. I doubt its as big as 240 sq mm.

http://www.bitsandchips.it/52-english-news/7622-rumor-two-versions-of-raven-ridge-under-development
https://techreport.com/forums/viewtopic.php?t=119682

The Stilt probably has a chip in hand at this point

But seriously, that's pretty big. Will be interesting to see what it can do.