Speculation: Ryzen 3000 series

[NobleX13]

I wonder how these compare to the quad-core Matisse Engineering Sample I have sitting here on my bench. This was an early production sample, similar to the one that showed up in some of the early Ryzen 3000-series leaks. It's got a much lower clock speed than these. If I remember correctly this chip boosts to a maximum of 3.8Ghz.

I would throw up a CPU-z screenshot but I forgot to nab one when I had this chip in a functional system. All I have on hand for motherboards is an X470 Gaming Plus from MSI, and it will only boot this CPU on the original BIOS that first supported Matisse CPUs.

To be clear, this is nothing more than a distant cousin to the new quad-core parts.

 

[DrMrLordX]

To me it sounds more like that stock is pilling up closer to next CPU launch, rather than a good thing. The fact they were unable to launch B550 until this point and A320 still dares to live also suggest they had far too many chipsets produced and stored.

This is not a suprise to me, they had far too many unsold Ryzen 1000 and 2000 around once the Zen 2 launched. This is also happening while Picasso stock is drying up and Renoir is nowhere to be found.

I wouldn't be that pessimistic. AMD has sold enormous numbers of Matisse CPUs. Binning through that many dice is going to produce defects which can be later sold as quads. As for B550, there's no indication that AMD has been "unable" to launch this chipset at all. They made more money forcing everyone to go x570, and it gave mobo OEMs the chance to recycle old chipsets from previous generations that they may have still had in stock (see: MSI "Max" motherboards). Nobody's been able to articulate exactly what it is that B550 would offer over B450 with the possible exception of guaranteed UEFI support for Matisse. You don't really think AMD would put PCIe 4.0 on B550 at launch in July 2019, do you? Or even a few months later?

 

[DarthKyrie]

I wouldn't be that pessimistic. AMD has sold enormous numbers of Matisse CPUs. Binning through that many dice is going to produce defects which can be later sold as quads. As for B550, there's no indication that AMD has been "unable" to launch this chipset at all. They made more money forcing everyone to go x570, and it gave mobo OEMs the chance to recycle old chipsets from previous generations that they may have still had in stock (see: MSI "Max" motherboards). Nobody's been able to articulate exactly what it is that B550 would offer over B450 with the possible exception of guaranteed UEFI support for Matisse. You don't really think AMD would put PCIe 4.0 on B550 at launch in July 2019, do you? Or even a few months later?

I have heard that the B550 will have PCIe 4 support for the first M.2 slot and the first PCIe slot which should be an x8 4.0 lanes.

 

[DrMrLordX]

I have heard that the B550 will have PCIe 4 support for the first M.2 slot and the first PCIe slot which should be an x8 4.0 lanes.

Sounds about right. It's just that AMD wouldn't have done that last year. It would have undercut x570.

 

[DarthKyrie]

Sounds about right. It's just that AMD wouldn't have done that last year. It would have undercut x570.

The B550 boards are basically beefed up B450 boards with PCIe 4.0 from the CPU.

 

[Thunder 57]

I wouldn't be that pessimistic. AMD has sold enormous numbers of Matisse CPUs. Binning through that many dice is going to produce defects which can be later sold as quads. As for B550, there's no indication that AMD has been "unable" to launch this chipset at all. They made more money forcing everyone to go x570, and it gave mobo OEMs the chance to recycle old chipsets from previous generations that they may have still had in stock (see: MSI "Max" motherboards). Nobody's been able to articulate exactly what it is that B550 would offer over B450 with the possible exception of guaranteed UEFI support for Matisse. You don't really think AMD would put PCIe 4.0 on B550 at launch in July 2019, do you? Or even a few months later?

How very Intel of them. But hey, at least they kept the same socket.

 

[DrMrLordX]

How very Intel of them.

They're learning! Hopefully they don't learn all the wrong lessons . . .

 

[Steltek]

I had to look, and the cheapest Comet Lake i3 (10100) has a 4.3 SCT and 4.1 ACT and would in theory be slightly more than $125. Who knows that Intel will do with the actual pricing but it's clear the 3300X is priced to compete with that but you aren't getting much of a discount by going with AMD. Doesn't seem like there will be anything 4C8T cheaper than the 10100.

I want to see how high the 3100 and 3300X can clock.

Not that it means much (benchmarks being mostly worthless), but

Overclocked AMD Ryzen 3 3300X Beats Intel Core i7-7700K in Geekbench

Benchmarks for AMD's upcoming budget Ryzen CPUs.

www.tomshardware.com

 

[DrMrLordX]

That's interesting. Would be fun to see a comprehensive 3300x vs 7700k bench-off at similar clocks. GB4 isn't that great on its own.

 

[Shivansps]

I wouldn't be that pessimistic. AMD has sold enormous numbers of Matisse CPUs. Binning through that many dice is going to produce defects which can be later sold as quads. As for B550, there's no indication that AMD has been "unable" to launch this chipset at all. They made more money forcing everyone to go x570, and it gave mobo OEMs the chance to recycle old chipsets from previous generations that they may have still had in stock (see: MSI "Max" motherboards). Nobody's been able to articulate exactly what it is that B550 would offer over B450 with the possible exception of guaranteed UEFI support for Matisse. You don't really think AMD would put PCIe 4.0 on B550 at launch in July 2019, do you? Or even a few months later?

I would not be so sure, im still getting A320 and B450 boards whiout Ryzen 3000 bios in high quantities (around 100-200) every now and then. it is a nightmare that i dont seem to wake up. This means they are still dealing with unsold motherboards produced over a year ago somewhere in the supply chain. So is not a suprise to me that B550 was delayed this long and A520 is still MIA. Both OEM and AMD wants fresh chipset money with a new motherboard series launch, there is no way AMD planned for a year of x570 PCI-E 4.0 exclusivity. This was probably OEM pressure to clear stock first.

Now with the CPUs remember how they had to give the Ryzen 1600 for almost free for stock clearing after Ryzen 2000 launched, and now after almost a year of Ryzen 3000, the Ryzen 2600 and even 2700 stock is still holding strong, and what happened with the Ryzen 1600AF and now 1200AF is just bizzare. So im really not suprised about what happened with these 3100... I do wonder if they are trying to cover holes until Renoir finally launch at desktop, because APU stock IS drying up.

One thing is certain, these 3100 along Renoir makes zero sence, so something is up here. Maybe Renoir really dosent launch until the end of the year, and APU stock is disapearing fast.

They're learning! Hopefully they don't learn all the wrong lessons . . .

Too late, remember how they blocked OEM to provide bios with PCI-E 4.0 support for 300 and 400 motherboards?

 

[DrMrLordX]

there is no way AMD planned for a year of x570 PCI-E 4.0 exclusivity. This was probably OEM pressure to clear stock first.

You don't think AMD knew about the supply chain issue with B450? It made a lot of sense for them to just let it ride. Try giving them a little credit.

Too late, remember how they blocked OEM to provide bios with PCI-E 4.0 support for 300 and 400 motherboards?

You still think that worked properly, don't you?

 

[AtenRa]

Now with the CPUs remember how they had to give the Ryzen 1600 for almost free for stock clearing after Ryzen 2000 launched, and now after almost a year of Ryzen 3000, the Ryzen 2600 and even 2700 stock is still holding strong, and what happened with the Ryzen 1600AF and now 1200AF is just bizzare. So im really not suprised about what happened with these 3100... I do wonder if they are trying to cover holes until Renoir finally launch at desktop, because APU stock IS drying up.

One thing is certain, these 3100 along Renoir makes zero sence, so something is up here. Maybe Renoir really dosent launch until the end of the year, and APU stock is disapearing fast.

AMD still producing at GloFo 14nm and 12nm.
Those Ryzen 1000 and 2000 CPUs are not sold for free, AMD always launch a product in the market with a high price and gradually lowers prices as RND cost gets amortized and manufacturing costs is getting lower.

Two reasons why AMD still using 14/12nm at GloFo.
1. WSA agreement still active , so they have to use those 14/12nm wafer capacity
2. Not enough 7nm capacity in 2019 and H1 2020.

So AMD introduced 1600F and 1200F to use 12nm GloFo capacity because of the WSA and in order not to use 7nm capacity for low margin products.

As for Ryzen 3300X/3100, I believe those are coming to counter Intel's new 4C 8T Core i3s Gen 10 coming in May.

 

[Shivansps]

You don't think AMD knew about the supply chain issue with B450? It made a lot of sense for them to just let it ride. Try giving them a little credit.

it really matters who realised that?

You still think that worked properly, don't you?

Well i did try a PCI-E 4.0 NVME on a A320 board with a PCI-E 4.0 enabled bios and it was working. There is really no way to know for sure, one way or the other. What i do know is that pci-e 4.0 was one of the selling points for x570 and the only reason for B550 to exist so it gets suspicious to block it in favor of "stability".

AMD still producing at GloFo 14nm and 12nm.
Those Ryzen 1000 and 2000 CPUs are not sold for free, AMD always launch a product in the market with a high price and gradually lowers prices as RND cost gets amortized and manufacturing costs is getting lower.

Two reasons why AMD still using 14/12nm at GloFo.
1. WSA agreement still active , so they have to use those 14/12nm wafer capacity
2. Not enough 7nm capacity in 2019 and H1 2020.

So AMD introduced 1600F and 1200F to use 12nm GloFo capacity because of the WSA and in order not to use 7nm capacity for low margin products.

As for Ryzen 3300X/3100, I believe those are coming to counter Intel's new 4C 8T Core i3s Gen 10 coming in May.

When i said for free i didnt mean it literally, but the 1600 ended up selling way below $100 at some point, hurting their own Ryzen 2000 sales. Thats stock clearing, you really dont want to do that intencionally.
At that time they were using 14nm production to produce APUs and 12nm to produce Zen+, so im petty sure the fabs were busy.

And what you said about 12nm production and 1600AF/1200AF production stops to make sence the moment you realise that it would have been better to put the fabs into making more Picassos that they are now dissapearing before Renoir arrival.

 

[amrnuke]

A little OT but since it's come up:

Out of curiosity, have we yet verified the true 1200AF and 1600AF uarch / specs? I know CPU-Z shows it is Pinnacle Ridge, but I've also heard others says that it's just ported 14nm -> 12nm with no significant updates outside of microcode.

So I'm unsure if 1600AF is essentially the old 1600AE but ported to 12nm, or if it's a badly-binned 2600? (In which case, how do we explain reports that show on 1600AF the IMC may actually be old Zen because it seems fairly limited compared to proper Zen+ chips?)

Do we have any official data on this or at least a technical breakdown?

 

[moinmoin]

I know CPU-Z shows it is Pinnacle Ridge, but I've also heard others says that it's just ported 14nm -> 12nm with no significant updates outside of microcode.

Zen+ is just Zen ported to 12nm with updated microcode.

 

[DarthKyrie]

I would not be so sure, im still getting A320 and B450 boards whiout Ryzen 3000 bios in high quantities (around 100-200) every now and then. it is a nightmare that i dont seem to wake up. This means they are still dealing with unsold motherboards produced over a year ago somewhere in the supply chain. So is not a suprise to me that B550 was delayed this long and A520 is still MIA. Both OEM and AMD wants fresh chipset money with a new motherboard series launch, there is no way AMD planned for a year of x570 PCI-E 4.0 exclusivity. This was probably OEM pressure to clear stock first.

Now with the CPUs remember how they had to give the Ryzen 1600 for almost free for stock clearing after Ryzen 2000 launched, and now after almost a year of Ryzen 3000, the Ryzen 2600 and even 2700 stock is still holding strong, and what happened with the Ryzen 1600AF and now 1200AF is just bizzare. So im really not suprised about what happened with these 3100... I do wonder if they are trying to cover holes until Renoir finally launch at desktop, because APU stock IS drying up.

One thing is certain, these 3100 along Renoir makes zero sence, so something is up here. Maybe Renoir really dosent launch until the end of the year, and APU stock is disapearing fast.



Too late, remember how they blocked OEM to provide bios with PCI-E 4.0 support for 300 and 400 motherboards?

The mobo guys were still cranking out B450 boards because ASMedia was late with the new chipset design. The new chips were supposed to be ready in time for the x570 release but they were having issues with their PCIe 4.0 which is why AMD ended up using the I/O die as a chipset chip for x570 boards. We would have had the B550 and A520 boards earlier this year but COVID-19 caused delays in manufacturing those chips (ended up being delayed until after the new year instead of being done Q4 2019).

Should AMD have had their partners stop making the lower tier boards for budget builders when they knew that the new stuff wouldn't be ready for another 6 months?

 

[amrnuke]

Zen+ is just Zen ported to 12nm with updated microcode.

That's true of just the core, but they have different IMCs and cache (including much lower latency in many places) which go beyond microcode, correct?

If that's the case, can we verify which IMC and cache is present on 1200AF and 1600AF?

It has been a running curiosity of mine.

 

[DrMrLordX]

it really matters who realised that?

Absolutely.

Well i did try a PCI-E 4.0 NVME on a A320 board with a PCI-E 4.0 enabled bios and it was working. There is really no way to know for sure, one way or the other. What i do know is that pci-e 4.0 was one of the selling points for x570 and the only reason for B550 to exist so it gets suspicious to block it in favor of "stability".

B550 != A320. If you will recall, one of the major things AMD had to change moving to PCIe4.0 was trace routing. Thin PCBs with traces routed for PCIe 3.0 were AT BEST working only for the first PCIe slot typically reserved for dGPUs. You weren't getting PCIe 4.0 to any of your m.2 slots or to any of the other PCIe slots further away from the CPU socket. B550 would have been designed from the ground-up to be compliant with the guidelines PCISIG established for PCIe 4.0 wrt proper routing of traces and meeting the minimum standards for motherboard layers (which supposedly was 6-layer, though some x570s are only 4-layer boards, hmm). It would have supported multiple PCIe slots and m.2 slots at PCIe 4.0 speeds. Unlike any of the 400 and 300-series chipsets. You can't infer anything about B550 from observations about A320.

Fact is that 300 and 400-series chipset boards were in no way, shape, or form compliant with the standards established by PCI SIG. Advertising them as PCIe 4.0 boards would have been dishonest, even if you could (maybe )get one PCIe slot working at those speeds as a hack. That it drove additional sales of x570 was a nice bonus, and probably motivated AMD to just pull the plug anyway because, why fret over it? It's not like anyone in the consumer market would benefit from PCIe 4.0 compliance in one PCIe slot anyway.

the 1600 ended up selling way below $100 at some point, hurting their own Ryzen 2000 sales.

Can you prove that? Or show us some evidence, other than conjecture or limited anecdotal evidence? AMD might have pushed the 1600AF as a replacement for 2000-series chips and stopped production of hexcores like the 2600 and 2600X.

The mobo guys were still cranking out B450 boards because ASMedia was late with the new chipset design.

First I've heard of that. Interesting if true.

 

[Thibsie]

Read it months ago already (B550 delayed because of (ASMedia) and that was on this forum.

 

[DrMrLordX]

Read it months ago already (B550 delayed because of (ASMedia) and that was on this forum.

Not saying you're wrong. It's just that I've seen so much different info on B550 in the past that it's hard to figure out what was really going on.

 

[moinmoin]

That's true of just the core, but they have different IMCs and cache (including much lower latency in many places) which go beyond microcode, correct?

If that's the case, can we verify which IMC and cache is present on 1200AF and 1600AF?

It has been a running curiosity of mine.

No, the silicon is essentially the same (the slightly denser 12nm allows for better spacing thus the higher reachable frequencies). The higher latency on consumer Zen was actually a bug and not by design, they were fixed in Epyc and Threadripper, but not in Ryzen until Zen+.

 

[Thibsie]

Not saying you're wrong. It's just that I've seen so much different info on B550 in the past that it's hard to figure out what was really going on.

No offense taken
A lot has been said indeed.

 

[Shivansps]

Can you prove that? Or show us some evidence, other than conjecture or limited anecdotal evidence? AMD might have pushed the 1600AF as a replacement for 2000-series chips and stopped production of hexcores like the 2600 and 2600X.

AMD is the only one that can prove it since no one else has the number of the expected Ryzen 2600 sales compared to what they got.

But im sure you will understand what having the 1600 selling for $80-$100 and the 2600 at $160-200 was a problem.
And the fact the 2600 stock is still holding strong after a year of being replaced and they had to create addicional SKUs (1600/1200AF) to keep producing on 12nm is a important clue. I dont think those were AMD plans, i think the original 1600 created a lag in the middle area once the Ryzen 2600 launched and it is still a problem today.

The only thing that i dont understand is why they allowed Picasso stock to run so low, they could have produced more Picassos instead of the 1600AF/1200AF. The APUs are produced at a diferent fab or something? The 3400G is gone and at this rate the 3200G will follow, why that happened when they are wasting 12nm production for the 1600AF/1200AF is beyond me.

 

[DrMrLordX]

AMD is the only one that can prove it since no one else has the number of the expected Ryzen 2600 sales compared to what they got.

So, conjecture.

But im sure you will understand what having the 1600 selling for $80-$100 and the 2600 at $160-200 was a problem.

Not really. If they EOL the 2600 and push everyone to the 3600 while producing the 1600AF for pennies and selling it to the downmarket, it just expands their marketshare.

No offense taken
A lot has been said indeed.

I will say this much: it would not surprise me at all that ASMedia had struggled with the chipset. I'm not their #1 biggest fan.

 

[gorobei]

re: b550 mooreslawisdead says the impression/rumblings he is getting is that it was held back in order to get the pcie4 working properly as they have a new process/technique/part to enable the bus speeds at reasonable cost.