Speculation: Ryzen 3000 series

[TheGiant]

PCIe 4.0 NVMe RAID drives setups are closing in on that Optane advantage fast. Also Intel is not going to be the only one offering persistent memory for much longer https://blocksandfiles.com/2019/04/...memory-extension-tech-for-epyc-optane-battle/

As regards PCIe 5.0 and DDR5 roadmaps, Intel has been churning away at roadmaps for some while and changing them so fast that nobody knows what's really going on engineering wise. Intel has already lost the node advantage and is late to offer PCIe 4.0.
They still have a monolithic design which is less flexible and much slower to develop than AMD's IO+chiplets design.
Those Intel roadmaps are really banking on everything going right and that is overly optimistic considering Intel's recent history.

there is nothing comparable with optane atm on the market ( especially 4K files)

but that doesnt matter with ryzen 3k series

do we have any other leaks on x570 boards? some without fan?

 

[darkswordsman17]

The vendor rep state the fan doesn't always run and implied that there is a lot of "fast stuff when all the fast stuff is being powered the fan runs"
This has lead people to speculate the fan is mainly for M.2 raid setups.

That would actually be a surprisingly forward thinking design consideration (maybe I'm wrong and there's Intel boards that do similar) if they were looking at elaborate heatsink/fan setups as a way of dealing with multiple fast SSDs knowing that will be something that is likely to happen. I wonder if they might even integrate it with the backplate as a means to shunt as much of the heat away from the socket as possible in order to try and deal with the dense heat output of 16core chips.

I could easily see it offering everything from 16 cores to 64 cores in stair steps of 16 cores, or 8 cores if they have a 4 chiplet mode for the IO chip and 4 cores if they have a 2 chiplet mode for that same IO die. I suspect that the 4 chiplet mode may be the smallest that they bother with as 4 chiplets with 4 cores each would likely be the least cost configuration that has any real market relevance.

I really hope they'd enable the extra memory channels. Even if they just topped out at 48 cores, they'd absolutely need more than 4 channel memory as 32 core chips are already pretty memory constrained. 6-8 8core chiplets with each being able to have its own dedicated memory channel could make for a solid mini-server, where you could isolate individual users (for security/privacy). Which I'm sure AMD would prefer to push anyone wanting that capability to EPYC, I think its a potential avenue where Threadripper would make a good amount of sense. If Zen 2 is as good as it seems, I have a hunch AMD will have no problem selling as many of them to major companies, and with the consumer Ryzen going possibly 16 core, they'd need something to differentiate Threadripper, and I think it would please enthusiasts, many of whom are already doing things like that.

I hope that the I/O is fairly flexible (so that if your workload is very memory intensive that you could just have one or two chiplets with the full memory bandwidth).

 

[Atari2600]

Is there any impression as to what they'll do with TR? It seems to me it would make little sense to offer a 16/32 chip, if the Ryzen 39xx already will provide that at a lower cost with higher single-core performance to boot.

On X399 you still have:

- Big power budget
- Much larger memory capacities
- Much more PCIe

For any kind of CAE work, X399 is very useful.
For FEA like ANSYS/Hyperworks/Nastran you can make use of the PCIe connectivity to set up a massive RAID0 NVMe based drive for your scratch area.
For CFD like Fluent/CFX/StarCD you will need more memory than AM4 is able to offer. 128 GB is still a bit stingy, so it'd be nice to see X399 (officially) capable of taking 256 GB.

Intel currently has the lead in FP operations, but with Zen2 doubling up to 256b FPUs, that should largely go away.

AVX512 is still so niche that you'd have to get custom compiled installs of the commercial. Which is going to be hard to come by.


edit: Actually, I take that back. AMD is already competitive for (at lease some) commercial code:

https://www.cfd-online.com/Forums/hardware/196400-amd-epyc-cfd-benchmarks-ansys-fluent.html

 

[lobz]

I'm sure they very much are interested in rome.
Right now they have to compete against servers with persistent memory that get to skip a lot of the CPUs load store cycles from and to the main storage and thus has a huge benefit over anything else on the market especially in anything that has to do with lots of data.
https://www.storagereview.com/intel_optane_dc_persistent_memory_module_pmm
In a couple of years they will have to compete against PCIe5 and DDR6 as well.
https://www.top7buy.com/intels-server-route-map-shows-ddr-5-and-pcie-5-0-memory-in-2021/

The first one is quite a niche right now (even if it's not as ridiculously narrow as avx512) because the client base that needs optane so desperately that it would be worth to them paying the exorbitant prices having optane means, is not that big at the moment. With the second one you are trying to compare something that comes within 2 months (well it's already shipping) with something that would come in 2 years at best - and basically nothing happened at intel at its best in the past 4 years regarding product and process node launches. We'll see what intel will have to compete with at the time they get there. Yes, it still sounds very funny, but things have turned around a lot, that's why one can write something as atrocious as intel having to compete with anything on the server market

 

[lobz]

With Old Jim, it's impossible to have polite disagreements.

That narcissist would lob you with personal attacks and tell you to commit suicide.

Another straight up lie from you, as this case was shown and explained already, this particular accusation was also thoroughly debunked in one of his videos, but that just shows how some arguments go here on this forum, unfortunately...
Thankfully I got the much needed help for ignoring someone (I tend to miss a lot of things on websites when browsing on a phone), so we won't bother each other anymore anyway.

 

[AtenRa]

If the bellow diagram is legit, then Ryzen 3 has so much I/O capabilities that it doesn't even need a motherboard chipset.

I would like a Mini-iTX board with no chipset that can take a 16C Ryzen 3 please

 

[jpiniero]

Even if they just topped out at 48 cores, they'd absolutely need more than 4 channel memory as 32 core chips are already pretty memory constrained

If you think about it, Threadripper doesn't make much sense now to AMD except to sell the partially defective IO dies. They'd make a ton more money if it's on the Epyc platform.

 

[TheGiant]

if AMD releases 16C AM4 ryzen it will destroy the TR ecosystem and pretty much give a signal that previous HEDT is obsolete
they should give 16C desktop more time, 12C is enough

 

[.vodka]

if AMD releases 16C AM4 ryzen it will destroy the TR ecosystem and pretty much give a signal that previous HEDT is obsolete
they should give 16C desktop more time, 12C is enough

Nah. Progress has to go on. AMD is at the helm of progress right now. Their platforms are upgradable:

AM4 gets an upgrade to 16 cores, TR4 benefits from Zen2's new chiplet topology eliminating NUMA related drawbacks on the future lineup (this alone plus TR4's extra connectivity/memory BW/capacity justifies HEDT) while also probably getting an upgrade to more than 32 cores.

AMD gets to charge more for 3xxx parts since they're bound to have the halo parts AND most secure designs so far for the foreseeable future on both AM4/HEDT while the chiplet paradigm scales nicely, is cheap and lets them reuse every chiplet out there save for the ones that go straight to the trash bin.

Win/win for everyone.

 

[krumme]

if AMD releases 16C AM4 ryzen it will destroy the TR ecosystem and pretty much give a signal that previous HEDT is obsolete
they should give 16C desktop more time, 12C is enough

That's some reactive defensive tactics in light of upcoming 10c Intel cpu imo.
Their roadmap is now a yearly tock tock tock and 7nm euv and 5nm is safe. New fast cpu is coming.
TR is pretty irrelevant from an economic perspective but aside from that what really matters is imo to get back brand value and revenue. A 16c will add to that. It will establish amd as the leader. No need to keep it back. TR can just establish a higher end hedt segment. 2500 usd 32c stuff. If they want to cater to that segment that is.
Consumers were kind of lucky that epyc had this inconsistent memory latency because it hindered a lot of server sales. That gave us the dies and kept prices low. I hope amd have turned up production radically because otherwise we will get shafted and just get expensive cpu from the trash bin. Rome unfortunately looks way to attractive. But hey consumers buying less of the expensive phones might help to keep tsmc capacity up!

 

[Atari2600]

if AMD releases 16C AM4 ryzen it will destroy the TR ecosystem and pretty much give a signal that previous HEDT is obsolete
they should give 16C desktop more time, 12C is enough

Ach now.

You know as well as me memory bandwidth is still a key, if not the key bottleneck in many CAE workloads.

X399 has its uses at 16C and 32C.

It won't be as cheap or common as AM4, but that is fine. Folks who need the capability won't blink at paying the extra few hundred $/€/£ for it.

Even renderers who wouldn't benefit from the 4 channels of memory might see enough benefit from the added PCIe/NVM connectivity to justify the platform.


[and is 64GB really enough for HEDT/prosumer workloads? Certainly not for me]

 

[Shivansps]

What a bs move that msi board hidding the active fan.

 

[Karnak]

View this content on Instagram

View this content on Instagram

Maybe hard to tell but looks to me like there is no chipset fan.

 

[TheGiant]

Ach now.

You know as well as me memory bandwidth is still a key, if not the key bottleneck in many CAE workloads.

X399 has its uses at 16C and 32C.

It won't be as cheap or common as AM4, but that is fine. Folks who need the capability won't blink at paying the extra few hundred $/€/£ for it.

Even renderers who wouldn't benefit from the 4 channels of memory might see enough benefit from the added PCIe/NVM connectivity to justify the platform.


[and is 64GB really enough for HEDT/prosumer workloads? Certainly not for me]

OFC
but what you say is WorkStation workload, look at what people do on their threadrippers/i9....16C with 16/32GB RAM?really a WS workload..more like the cinebench runners

I am saying that if you have 16C 2CH ram desktop with lower power and higher freq and IPC suddenly 4CH and more PCI-E lanes becomes a noargument unless you really know what you are doing..

AMD isnt good at keeping price levels and so their profits look like

I see 16C on the AM4 as engineering decision, 12C as a balance engineering/marketing/finance decision with potential to release a 16C if Intel has something in their pocket (10C comet lake isnt a competitor for 16C ryzen 3k)

 

[mattiasnyc]

I would think the primary problem for x399 moving forward if they release an AM4 16 core cpu is that it puts into question the future of the TR4 platform. I think potential buyers of a TR would ask themselves just where that platform is heading. Sure, there's the 32 core current CPU, but what about future CPUs? Will there be faster CPUs for TR4 or CPUs with more cores?

It would seem to me that if AMD says nothing about TR when AM4 is announced and finally released then the TR platform is a bit in limbo. Why buy into it if there won't be future CPU upgrades? It's not about the advantages that already exist - memory channel count, lane count etc - it's about the fact that there's competition with Intel and that AMD will need to match/surpass Intel also in the future.

If AMD announces that new Threadrippers are coming, based on Zen 2, even if it's "only" up to 32 cores, then I think the x399 platform will do fine (or a x499 if there's a platform update). But I really do think timing will be difficult here and that there likely will be a slump in sales until such an announcement is made. I could be wrong of course but...

 

[maddie]

OFC
but what you say is WorkStation workload, look at what people do on their threadrippers/i9....16C with 16/32GB RAM?really a WS workload..more like the cinebench runners

I am saying that if you have 16C 2CH ram desktop with lower power and higher freq and IPC suddenly 4CH and more PCI-E lanes becomes a noargument unless you really know what you are doing..

AMD isnt good at keeping price levels and so their profits look like

I see 16C on the AM4 as engineering decision, 12C as a balance engineering/marketing/finance decision with potential to release a 16C if Intel has something in their pocket (10C comet lake isnt a competitor for 16C ryzen 3k)

What exactly defines a competitor? Price, cores, GHz? If 2 products are similarly priced, are they competitors even if one far surpasses the other? I say yes.

 

[moinmoin]

PCIe 4.0 NVMe RAID drives setups are closing in on that Optane advantage fast. Also Intel is not going to be the only one offering persistent memory for much longer https://blocksandfiles.com/2019/04/...memory-extension-tech-for-epyc-optane-battle/

As for a direct competitor for Optane, shouldn't Micron be able to offer products based on 3D XPoint not limited to Intel platforms? Since Micron bought the fab I can't imagine them doing nothing (or only Intel exclusive products) with it.

 

[Abwx]

I would think the primary problem for x399 moving forward if they release an AM4 16 core cpu is that it puts into question the future of the TR4 platform. I think potential buyers of a TR would ask themselves just where that platform is heading. Sure, there's the 32 core current CPU, but what about future CPUs? Will there be faster CPUs for TR4 or CPUs with more cores?

For TR 32/48C will still make sense, FI a 32C at 250W@4GHz has not only 100% better perf than the alleged AM4 16C but has also somewhat better perf/watt at the plateform level.

A 24C is unlikely because it require 4 chiplets, unless yields are such that they have a massive inventory of dies with only 6/7C being functional.

 

[dnavas]

I would think the primary problem for x399 moving forward if they release an AM4 16 core cpu is that it puts into question the future of the TR4 platform.

Not for me, but for those who worry about the limits of cores, if TR4 only uses half of their area for cores, I can imagine all kinds of heterogenous compute opportunities for the rest.

 

[Fanatical Meat]

View this content on Instagram

View this content on Instagram

Maybe hard to tell but looks to me like there is no chipset fan.

I’m going to soeculate there is a fan and it’s being concealed by the weird cuts in the picture. Scale is zoomed in on some parts.
Every X570 board has had a fan, safe to assume this will have a fan.
Never owned a PC with a chipset fan but the idea doesn’t bother me much.

 

[Fanatical Meat]

Good news is we are around 60 hours away from more info
Bad news is we have no idea what that info will be.

 

[.vodka]

View this content on Instagram

View this content on Instagram

Maybe hard to tell but looks to me like there is no chipset fan.

Look closely, there's a mesh covering the chipset heatsink.

That's a hidden fan.

 

[zinfamous]

nevermind, read the wrong thing.

 

[crashtech]

The suspense is killing me at this point. No word on AVX2/AVX512 at this point yet either, from what I can tell.

 

[Mockingbird]

The suspense is killing me at this point. No word on AVX2/AVX512 at this point yet either, from what I can tell.

AMD talked about AVX2 during EPYC 2's unveiling