Speculation: Ryzen 3000 series

[Fanatical Meat]

Appears ASRock has updated bios to support new Ryzen processors
Appears a320 is on the update list too

https://www.techpowerup.com/255220/...-cpu-bios-updates-for-socket-am4-motherboards

 

[coercitiv]

Appears a320 is on the update list too

A320 Chipset*
*Supports AMD Next-Gen Ryzen series APU only.

 

[Fanatical Meat]

A320 Chipset*
*Supports AMD Next-Gen Ryzen series APU only.

Ah thank you

I still say nearly no person with an a320 board is looking to get a 3000 cpu.

 

[moinmoin]

You know, the reason why Threadripper was removed from the 2019 lineup may simply be based on timing. For Zen 1 and Zen+ there were 5-6 months each between the launch of the first Ryzen chip and the first Threadripper chip. The 3rd gen Ryzen may now be that late that the 3rd gen Threadripper is naturally pushed into 2020 if AMD keeps the time gap between the launches the same.

 

[JDG1980]

everything seems too optimistic for me- 7nm is the miracle

The 7nm process is the work of TSMC, not AMD.

toms posted that TR from ryzen 2 dropped from the roadmap...wondering what it means https://www.tomshardware.com/news/amd-third-gen-threadripper-roadmap,39254.html

What it probably means is that they're saving it for 2020, hoping that they can get well-heeled buyers to spring for a 12/16-core AM4 chip first and then a 24/32-core TR4 chip the next year.

If their AM4 offerings can beat Intel's mainstream chips (and chances are they can) then why not?

 

[jpiniero]

Could see them holding back TR so that they can sell the remaining 12 nm dies.

 

[amd6502]

Could see them holding back TR so that they can sell the remaining 12 nm dies.

That would be an added side benefit, but I think the big concern is how efficient 7nm could be when clocked below HEDT frequencies. So they could be their own worst competition in the server market.

I do think they will have a TR 3000 but not soon; maybe within a few months of launching zen3/7nm+/6nm; so that's late spring 2020..

 

[Hans de Vries]

toms posted that TR from ryzen 2 dropped from the roadmap...wondering what it means https://www.tomshardware.com/news/amd-third-gen-threadripper-roadmap,39254.html

1) AM4 goes to 16 cores competing from 'below'
2) Threadripper's I/O die has all 8 memory channels and all PCIe lanes. Why would you throw these away?

Why not instead use the workstation EPYC ATX motherboards which are becoming more available and affordable? Compete with higher core counts AND higher memory bandwidth in HEDT.

 

[Thunder 57]

Could see them holding back TR so that they can sell the remaining 12 nm dies.

Could be true. Those probably move slower than Ryzen does. I also think AMD will limit Zen 2 to 12 cores initially for the same reason, to not lose sales on the 2950X.

 

[Shivansps]

Gigabyte 300 boards are getting bios updates now, strangely enoght, the A320 ones got the 0.0.7.2 agesa while they keep BR support, while all others are dropping BR support, considering Gigabyte uses the same chip for all motherboards this to me points out at APU only support as well.
There is really no way to justify this, as A320 boards tends to be the same or slightly modified B350/X370 boards. Also they must re-launch A320 as there is no new chipset.

 

[scannall]

Could be true. Those probably move slower than Ryzen does. I also think AMD will limit Zen 2 to 12 cores initially for the same reason, to not lose sales on the 2950X.

Cores isn't the biggest selling point of Threadripper, to me anyway. It's the 4 channel memory and the huge amount of I/O.

 

[Thunder 57]

Cores isn't the biggest selling point of Threadripper, to me anyway. It's the 4 channel memory and the huge amount of I/O.

While I would agree with you, I can't express how many times I have seen people go on about gaming. It seems like every time someone here mentions upgrading, the response comes down to gaming. It's annoying. Just like how so many B450/X470 boards are described as gaming.

TR has a huge amount of PCIe indeed. It's something I wish Ryzen had more of. Current AMD chipsets are a bit lacking there.

 

[fleshconsumed]

TR has a huge amount of PCIe indeed. It's something I wish Ryzen had more of. Current AMD chipsets are a bit lacking there.

Pretty sure Ryzen still has more PCIe lanes than intel mainstream K processors. Unless you're running 2 videocards and 2 or more NVME drives Ryzen has enough PCIe lanes.

 

[maddie]

Cores isn't the biggest selling point of Threadripper, to me anyway. It's the 4 channel memory and the huge amount of I/O.

Memory channels advantage, yes, PCIe, maybe not.

The I/O die disentangles the # PCIe from the number of cores.

Might it be possible they're thinking of only supporting 2 product lines instead of 3. Server & Desktop/Workstation?

 

[Topweasel]

Memory channels advantage, yes, PCIe, maybe not.

The I/O die disentangles the # PCIe from the number of cores.

Might it be possible they're thinking of only supporting 2 product lines instead of 3. Server & Desktop/Workstation?

No, no it doesn't. TR and Epyc are almost certainly going to use the same IO die and Ryzen is until a new socket going to be limited to the PCIe lanes that are available now. TR is going to be restricted because the specs for TR require a lot less PCIe traces coming out of it.

As for you question at the end. No unless they wanted to work back and do EPYC for the full range of workstation. But there is a certain if kind of week demand for a mid teir HEDT lineup that Ryzen can't fill with just cores. AMD wants to stay in that market for the increased Margin and Halo product statis.

 

[maddie]

No, no it doesn't. TR and Epyc are almost certainly going to use the same IO die and Ryzen is until a new socket going to be limited to the PCIe lanes that are available now. TR is going to be restricted because the specs for TR require a lot less PCIe traces coming out of it.

As for you question at the end. No unless they wanted to work back and do EPYC for the full range of workstation. But there is a certain if kind of week demand for a mid teir HEDT lineup that Ryzen can't fill with just cores. AMD wants to stay in that market for the increased Margin and Halo product statis.

It's amazing what being tired causes to happen.

 

[moinmoin]

Theoretically mainboard manufacturers could do the same as embedded board developers without relying on the Promontory or the upcoming 570 chipset. 3rd gen Ryzen would allow to bifurcate PCIe 4 lanes into double the amount of PCIe 3 lanes etc. But unlikely anybody will do this kind of custom work for the consumer market.

 

[Veradun]

No, no it doesn't. TR and Epyc are almost certainly going to use the same IO die and Ryzen is until a new socket going to be limited to the PCIe lanes that are available now. TR is going to be restricted because the specs for TR require a lot less PCIe traces coming out of it.

As for you question at the end. No unless they wanted to work back and do EPYC for the full range of workstation. But there is a certain if kind of week demand for a mid teir HEDT lineup that Ryzen can't fill with just cores. AMD wants to stay in that market for the increased Margin and Halo product statis.

Zen2 is PCIe4.

On AM4+x570 they can choose to split lanes as PCIe3 lanes effectively doubling the total available as compared to current chipsets

Theoretically mainboard manufacturers could do the same as embedded board developers without relying on the Promontory or the upcoming 570 chipset. 3rd gen Ryzen would allow to bifurcate PCIe 4 lanes into double the amount of PCIe 3 lanes etc. But unlikely anybody will do this kind of custom work for the consumer market.

It can be done at chipset level, and be a selling point for new motherboards

 

[moinmoin]

Zen2 is PCIe4.

On AM4+x570 they can choose to split lanes as PCIe3 lanes effectively doubling the total available as compared to current chipsets

It can be done at chipset level, and be a selling point for new motherboards

Will be really interesting which manufacturers will opt for pushing PCIe 4 and which for bifurcating all lanes (if that's possible) and offer respectively more connectors.

 

[scannall]

Will be really interesting which manufacturers will opt for pushing PCIe 4 and which for bifurcating all lanes (if that's possible) and offer respectively more connectors.

As a guess, most will likely have a PCIe 4 slot for video cards. And maybe one for NVME drives. Splitting the rest into PCI3 3 for the rest of the slots and M2 connectors. NVME raids on a consumer board might be fun. ;-)

 

[Veradun]

Will be really interesting which manufacturers will opt for pushing PCIe 4 and which for bifurcating all lanes (if that's possible) and offer respectively more connectors.

PCIe4 and the option to bifurcate also opens up to a lot of segmentation options for board manufacturers.

And we also already know they have far bigger lineups than with x470.

Not a coincidence, imho :>

 

[kapulek]

Some ZEN 3 rumors.

 

[moinmoin]

Some ZEN 3 rumors.

No way Threadripper and Xbox are getting Zen 3 earlier than Epyc. If that were to happen it's more likely a repeat of Zen+ with Epyc skipping it altogether.

 

[coercitiv]

https://twitter.com/x/status/1126466183740321798

 

[dnavas]

No way Threadripper and Xbox are getting Zen 3 earlier than Epyc. If that were to happen it's more likely a repeat of Zen+ with Epyc skipping it altogether.

Yeah, I assume Epyc would be sampling first, but the server market is a long sales cycle. There may be some fudging in the term "earlier". That said, I was already wondering if TR3 was going to be pushed out because of memory bandwidth/DDR5. If TR could get both DDR5 and local stacked memory/cache, there might be some benefit in waiting. If Zen2 doesn't deliver higher clocks, there's even less of a reason to upgrade to TR3 -- assuming memory bandwidth constrains the ability to take advantage of the greater fp throughput on hedt workloads (big if), and you also don't get significant per thread boost...? [looks ahead to next post .. hmm :| ] I've been planning an upgrade, but I've been starting to have concerns anyway....

I don't see SM3|4 being attractive on anything other than Epyc. And questionable even there. Even on servers, latency matters. If I could have 20% higher clocks (lower latency) or 20% greater throughput (higher SMT), I'd take the lower latency on TR and desktop every time. I could go either way on servers, depending on what I ran on them. I/O tasks might do well with more lower latency threads. But I'm not putting business logic on those cores. Even then, a lot of data access is computation (serde).

As for the 16-core "leak", as long as I'm referencing the post.... I'm not going to lie, 3.3/4.2 would be a miss for me. I'd never expect 16 cores to boost too high, but single core boost needs to be higher, and the base is under 2950x's, so that pair in a TR3 isn't going to be too attractive. Of course, I've been assuming that Zen2 would disappoint the 5Ghz expectations about as soon as they started floating around, and that we'd get 24 and 48 core models in TR3 because of tdp constraints; 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.