Speculation: Ryzen 3000 series

[Shivansps]

if i had to bet ill say the will do exactly what worked before, just with more cores.

ex:
R3 3200 6C/6T 11CU @ u$s100
R5 3400 6C/12T 15CU @ u$s 140

an R3 3300 witout IGP will probably be in some point in the midle.

IF the APU are really TBA Q3 2019 there is NO WAY that they already have the specs, naming, prices... So i would forget those two APUs from the adored leak.

I also think that R3 3300 @ 100 is not real or at least it has some specs wrong, and the $130 R3 3300X is the real thing.

 

[moinmoin]

The PCIe lanes. The quad channel memory controller. Anything else? What would the wattage be there?

Feeding 16 cores with the right data in a timely manner still requires an awful lot of uncore power, that simply the nature of it. Perhaps the I/O controller of Zen2 will improve this. Perhaps Zen2 will be able to completely shut off modules to pull down idle and low thread power - perhaps not.

IMC and PCIe lanes are indeed the biggest part of the uncore's power usage, and unlike with cores those can't be easily gated when idle. I expect the power consumption for the AM4 I/O chip to be within the ballpark to that of current Ryzen chips at idle, possibly slightly above for supporting PCIe 4.

 

[Kenmitch]

Not sure how it'll play out in the end as far as the skus go. It seems like the most logical solution would be chiplets for all future products. The ability to mix and match seems like it would outway the downsides of going chiplets for all. I'd imagine it would also be of a huge benefit to it's custom products and greatly speed up the process of getting them to market.

Some of those skus (if true) could really bring on the pain for Intel.

Looking forward to seeing how it all pans out in the end.

 

[Atari2600]

IMC and PCIe lanes are indeed the biggest part of the uncore's power usage, and unlike with cores those can't be easily gated when idle

Do you consider Infinity Fabric as part of that or separate?

Having a modular architecture does come with a disadvantage of having to shuttle information around a more complex distribution network than *can be ideal in some situations*.

 

[PhonakV30]

all I can say is No Way . looks like Those leaks are super Saiyan.

 

[Kedas]

They did double the L3 cache to reduce the impact of latency increase.
An increased IF frequency also reduces this latency.
Also chiplets do allow for higher max. clocks that you couldn't get on any 12C CPU die now, at least not for the same price.

What I don't see happening is a Ryzen 3000 launch on CES though.
An announcement with launch date, yes.
They did say Epyc first later Ryzen, and Epyc doesn't launch before CES.
Everyone expects Ryzen 3000 in March/April

 

[PotatoWithEarsOnSide]

Consensus seems to be 14 months between each Ryzen generation. That implies June (?)

 

[Fanatical Meat]

Consensus seems to be 14 months between each Ryzen generation. That implies June (?)

Dammit don’t ruin my fantasy of being able to build one in April

 

[Zapetu]

Here's some rumors that Next Gen Xbox will feature Zen2 and AMD next-gen GPU:
https://wccftech.com/xbox-scarlett-4k-60fps/

It's based on a video made by Brad Sams:

Spoiler: The Next Gen Xbox Strategy...Zen 2, Streaming, and IP

I didn't see this posted in here yet.
Take it as you will. It doesn't really say anything about chiplets, just that both next gen consoles would be released in 2020 and would use Zen2 CPUs (If MS does then why would Sony use anything else?).

 

[whm1974]

Personally with 7nm I wouldn't bother producing pure CPUs with less then 8 cores.

This is what I'll do if possible and profitable. Adjust accordingly especially pricing.

3000Mhz APU 4c/8t w/ 8CU Navi 1000mhz iGPU. TDP 35W 4MB L3. Replaces Ryzen 2200 . Come with decent low profile fan less heatsink

3500Mhz APU 6c/12t w/ 16CU Navi 1200mzh iGPU. TDP 65W 8MB L3. Replaces Ryzen 2400 and 2600 Comes with Wraith Stealth

3500mhz+ CPU 8c/16t. TDP 65W 16MB L3 replaces Ryzen 2700, Comes the above HSF.

A higher clocked CPU 8c/16t TDP 95W Same as above.

Of this is dependent on binning and what could be had with 1 gen 7nm nodes. and I forgot about the chiplets... Memory speeds will be higher and all SKUs will be unlocked.

I wondering if video memory can be made in chiplets? And how dense?

 

[moinmoin]

Do you consider Infinity Fabric as part of that or separate?

Having a modular architecture does come with a disadvantage of having to shuttle information around a more complex distribution network than *can be ideal in some situations*.

Imo IF (which is not only the data links aka SDF but also all the controls and sensors as part of the SCF) unquestionably is a part of the uncore. Regarding having to shuttle data around, the complexity actually will be lower with the Zen 2 system organisation, not having links between each and all CCX's but just one link to the IOC each. Being able to gate whole CCX's and links to them for power saving is one further advantage I saw a full L3$ copy on the IOC giving. Without it (which is the currently prevailing assumption) the L3$ on each chiplet as well as the links to them need to be powered at all time.

 

[dlerious]

Consensus seems to be 14 months between each Ryzen generation. That implies June (?)

There's only been 2 generations. R7 was Mar. 2 to Apr. 19, R5 was Apr. 11 to Apr. 19, and Threadripper was Aug. 10 to Aug. 31. There was no Zen+ for Epyc.

 

[VirtualLarry]

The R3-3300 is way too much for $100, it makes sence to a point were im thinking they are probably leaving rebranded RR as Athlons at sub <$100... but why they are skipping 4C/8T and 6C/6T on $100 and going directly to 6C/12T? its too much.

If the source of all this is AMD, they are probably testing reactions.

With the Ryzen R5 1600 selling for $129.99 new on sale, a next-gen CPU lineup that includes the lowest SKU at 6C/12T, and $100, does kind of make sense, in a "Core Wars" kind of way. Think of how many existing Intel rigs will be rendered immediately obsolete by that move? It's a move aimed straight at Intel's jugular, and I applaud it. Then again, I kind of see what you're saying. Without more "killer apps", some people may find that 6C/12T is all that they need, for the next 5-10 years.

I can kind of see where you're coming from, if you expect a steady stream of upgrade customers, for the next 5 years, and no-one shows up after buying a $100 6C/12T CPU.

 

[maddie]

With the Ryzen R5 1600 selling for $129.99 new on sale, a next-gen CPU lineup that includes the lowest SKU at 6C/12T, and $100, does kind of make sense, in a "Core Wars" kind of way. Think of how many existing Intel rigs will be rendered immediately obsolete by that move? It's a move aimed straight at Intel's jugular, and I applaud it. Then again, I kind of see what you're saying. Without more "killer apps", some people may find that 6C/12T is all that they need, for the next 5-10 years.

I can kind of see where you're coming from, if you expect a steady stream of upgrade customers, for the next 5 years, and no-one shows up after buying a $100 6C/12T CPU.

All well and good, but if someone can afford it, how many really buy the product they need versus the one they want? That is why we have marketing departments.

 

[OTG]

The frequency of upgrades may not matter so much if this gets them enough marketshare.
50% with infrequent upgrades has gotta be better than 10% with frequent upgrades.

 

[Fanatical Meat]

With the Ryzen R5 1600 selling for $129.99 new on sale, a next-gen CPU lineup that includes the lowest SKU at 6C/12T, and $100, does kind of make sense, in a "Core Wars" kind of way. Think of how many existing Intel rigs will be rendered immediately obsolete by that move? It's a move aimed straight at Intel's jugular, and I applaud it. Then again, I kind of see what you're saying. Without more "killer apps", some people may find that 6C/12T is all that they need, for the next 5-10 years.

I can kind of see where you're coming from, if you expect a steady stream of upgrade customers, for the next 5 years, and no-one shows up after buying a $100 6C/12T CPU.

I’m thinking that chip will be sold to system builders only like Dell. Explains the low price and gives the big players something to float their margins with.
Purely a guess from someone who is not an expert on the subject.

 

[PotatoWithEarsOnSide]

It's a strange situation. Even if yields are very good, I don't see how AMD can have enough spare 8c chiplets (including defective ones) to meet the needs of the lower and mid-ranged consumers unless the top AM4 is 16c.
We're potentially looking at 3 fully functional 8c chiplets for every 1 defective (and likely salvageable) 8c chiplet. We know that Epyc 2 won't only be 8x 8c chiplets, but also that the volume isn't massive either. Same goes for Threadripper, though with fewer chiplets and all fully functional.
I honestly don't see them disabling fully functional 8c chiplets in order to meet that demand.
If yields are lower, thus more defective chiplets, then maybe it is possible. However, that screws with the financial cost per fully functional chiplet, which I assume would be the best way to allocate costings.

 

[H T C]

It's a strange situation. Even if yields are very good, I don't see how AMD can have enough spare 8c chiplets (including defective ones) to meet the needs of the lower and mid-ranged consumers unless the top AM4 is 16c.
We're potentially looking at 3 fully functional 8c chiplets for every 1 defective (and likely salvageable) 8c chiplet. We know that Epyc 2 won't only be 8x 8c chiplets, but also that the volume isn't massive either. Same goes for Threadripper, though with fewer chiplets and all fully functional.
I honestly don't see them disabling fully functional 8c chiplets in order to meet that demand.
If yields are lower, thus more defective chiplets, then maybe it is possible. However, that screws with the financial cost per fully functional chiplet, which I assume would be the best way to allocate costings.

AMD is going to use a huge amount of cream-of-the-crop Epyc chips for this bad boy alone and, if we apply the "rule" of only the best 5% of chiplets go to Epyc and only the best of those goes to the top chip, than that's a heck of allot "not good enough" 8c chiplets after the binning process.
I'm pretty sure initial yields are much less ... shall we say ... favorable ... then AMD would have us believe, simply because it's a new process. This will ofc improve with time and then it's quite possible AMD will end up sacrificing perfectly good chips for lower solutions, depending on the needs.

 

[PhonakV30]

a Ryzen 2400G (or 6C/12T / 16 CU )with 8 GB HBM at price $150 is enough to kill all Low-end CPUs/Cards.

 

[itsmydamnation]

AMD is going to use a huge amount of cream-of-the-crop Epyc chips for this bad boy alone and, if we apply the "rule" of only the best 5% of chiplets go to Epyc and only the best of those goes to the top chip, than that's a heck of allot "not good enough" 8c chiplets after the binning process.
I'm pretty sure initial yields are much less ... shall we say ... favorable ... then AMD would have us believe, simply because it's a new process. This will ofc improve with time and then it's quite possible AMD will end up sacrificing perfectly good chips for lower solutions, depending on the needs.

1. The chips that EPYC will like , low leakage , low clocking are the polar opposite of the chips we want, high leakage , high clocking chips
2. A chiplet is smaller then the A12 and it has been in production for 1/2 a year right now, so 1 full year next june
3. much like the 6 core ryzen i think there will be plenty of "sacrificing" of fully functional chips for lower end product right off the bat.

 

[H T C]

1. The chips that EPYC will like , low leakage , low clocking are the polar opposite of the chips we want, high leakage , high clocking chips
2. A chiplet is smaller then the A12 and it has been in production for 1/2 a year right now, so 1 full year next june
3. much like the 6 core ryzen i think there will be plenty of "sacrificing" of fully functional chips for lower end product right off the bat.

Yes but to get the required amount for that bad boy i linked to, AMD will have to make a huge amount of chips so that, after the binning process, they achieve the required amount of Epyc chis and that's for this one super-computer: it will use 80K 8c chiplets.

If only 5% of all chiplets even get to be Epyc chips and out of those, few make the top-of-the-line chip, which is the one the super-computer will use, then the amount of total chips available for selection is already huge to begin with: should be more than enough for segmentation, initially, no?

 

[Shivansps]

With the Ryzen R5 1600 selling for $129.99 new on sale, a next-gen CPU lineup that includes the lowest SKU at 6C/12T, and $100, does kind of make sense, in a "Core Wars" kind of way. Think of how many existing Intel rigs will be rendered immediately obsolete by that move? It's a move aimed straight at Intel's jugular, and I applaud it. Then again, I kind of see what you're saying. Without more "killer apps", some people may find that 6C/12T is all that they need, for the next 5-10 years.

I can kind of see where you're coming from, if you expect a steady stream of upgrade customers, for the next 5 years, and no-one shows up after buying a $100 6C/12T CPU.

More likely, they be all be buying the 6C/12T at $100 and dont bother with anything over that, who is going to pay almost twice that for just 2 more cores? Not me, thats for sure.

And if there is no $100 APU you can bet any sale that is right now going to the 2200G its gona go the Athlons.
Thats bad for AMD.

But i dont fully belive in that leak, there are missing skus there, and APU having names, specs and prices when they are TBA Q3 2019 is completely fake. Remember we are still waiting on specs for Athlons that should be going out this month.

 

[Tuna-Fish]

More likely, they be all be buying the 6C/12T at $100 and dont bother with anything over that, who is going to pay almost twice that for just 2 more cores? Not me, thats for sure.

If there is a $100 6-core part, it would almost certainly have a locked multiplier, and likely a substantially lower peak frequency than the more expensive parts, giving people a reason to go for the more expensive ones.

And if there is no $100 APU you can bet any sale that is right now going to the 2200G its gona go the Athlons.

Based on the earlier leaked roadmap, the APUs for 2019 will be the existing 12nm parts, likely rebranded. This sucks, but that's life: AMD likely has a limited ability to tape out new designs, and they chose to go with the high-margin ones first.

 

[DarthKyrie]

More likely, they be all be buying the 6C/12T at $100 and dont bother with anything over that, who is going to pay almost twice that for just 2 more cores? Not me, thats for sure.

And if there is no $100 APU you can bet any sale that is right now going to the 2200G its gona go the Athlons.
Thats bad for AMD.

But i dont fully belive in that leak, there are missing skus there, and APU having names, specs and prices when they are TBA Q3 2019 is completely fake. Remember we are still waiting on specs for Athlons that should be going out this month.

6c/12t CPUs might be needed to play the new games coming with the next-gen console ports which AMD is also building for Sony and MS. All of these products make sense when you take the new consoles into consideration using Zen2.

 

[realibrad]

6c/12t CPUs might be needed to play the new games coming with the next-gen console ports which AMD is also building for Sony and MS. All of these products make sense when you take the new consoles into consideration using Zen2.

Maybe. The Xbox One and PS4 both had 8 cores. I believe 1 core was set for the OS, so the games had 6-7 cores to access. We still saw very little progress for multi core gaming.