AMD Zen 2 (Ryzen 3000) - *** NO COMPETITOR HINTING ***

[Murphy]

Zen hasn't been any super low power design any time.
It's GloFo's manufacturing node which has been holding it down.
Because of their inability to develop working node themselves, GloFo bought Samsung's 14nm mobile-SOC node!
That's why many people were originally expecting some max ~3,5GHz clocks for Ryzen.

Not really...

Zen is manufactured on Global Foundries' 14 nm process Low Power Plus (14LPP).

And no Global Foundries' manufacturing node was not holding them down, Global Foundries' and Samsung collaborated on the 14nm node, 12nm was the first FinFet node that was all of Global Foundries' own doing.

 

[AMDPower]

is there any news/rumours on whether the new X570 platform will allow more than 2 bandwidth restricted Nvme drives? Or would i have to wait for thread ripper.

I don't need more than 16cores but do need space for 4nvme drives without all the speed restriction, want to build a raid array of 4 drives. I would possibly go dual gpu next build so populating pci slots to accommodate NVMe drives is a no go.

my X58 system is really showing its age, cant wait for new AMD cpu/mobo

 

[CuriousTomCat]

is there any news/rumours on whether the new X570 platform will allow more than 2 bandwidth restricted Nvme drives? Or would i have to wait for thread ripper.

I don't need more than 16cores but do need space for 4nvme drives without all the speed restriction, want to build a raid array of 4 drives. I would possibly go dual gpu next build so populating pci slots to accommodate NVMe drives is a no go.

my X58 system is really showing its age, cant wait for new AMD cpu/mobo

X570 should unleash the full bandwidth of PCIe 4.0

 

[hominid]

X570 should unleash the full bandwidth of PCIe 4.0

I've been reading stories saying that PCIe 4.0 should be possible, for the first PCIe slot at least, for the X370 and X470 motherboards with a bios update. That would be nice if possible!

 

[Sargatanas2511]

PCIe 4.0 is not going to make much or any difference to a normal PC any time soon. It's not even like it'll help with SLI/Xfire because they barely exist now.

 

[LePhuronn]

PCIe 4.0 is not going to make much or any difference to a normal PC any time soon. It's not even like it'll help with SLI/Xfire because they barely exist now.

It depends on how it's implemented. You could take 8 PCIe 4 lanes and feed a GPU with the equivalent of 16 lanes of PCIe 3 - same bandwidth. That gives you either another 8 lanes to feed a 2nd GPU at full PCIe 3 bandwidth, or spread those lanes out to support more NVMe drives and whatnot that use up lanes.

Essentially, because PCIe 4 is double the bandwidth, you could bifurcate and distribute the lanes in such a way you're emulating twice the PCIe 3 lanes you had before. And that would make a big difference to the higher-end consumer systems.

 

[Journey]

is there any news/rumours on whether the new X570 platform will allow more than 2 bandwidth restricted Nvme drives? Or would i have to wait for thread ripper.

I don't need more than 16cores but do need space for 4nvme drives without all the speed restriction, want to build a raid array of 4 drives. I would possibly go dual gpu next build so populating pci slots to accommodate NVMe drives is a no go.

my X58 system is really showing its age, cant wait for new AMD cpu/mobo

If you are very lucky someone like MSI/ASRock might bring out a board that supports the PCI-E 4.0 being spit off into 3.0 lanes, so they could have an arrangement of two slots with 16x PCI-E 3.0 and one slot with 16x slot with only 8x PCI-E 3,0, totalling 40 lanes of 3.0 or 20 lanes of 4.0.

This would allow you to install two GPU's with on running at 8x lanes, and a 16x PCI-E M.2 riser card, assuming the board has bifurcation on the 16x slot as well..

 

[Sargatanas2511]

It depends on how it's implemented. You could take 8 PCIe 4 lanes and feed a GPU with the equivalent of 16 lanes of PCIe 3 - same bandwidth. That gives you either another 8 lanes to feed a 2nd GPU at full PCIe 3 bandwidth, or spread those lanes out to support more NVMe drives and whatnot that use up lanes.

Essentially, because PCIe 4 is double the bandwidth, you could bifurcate and distribute the lanes in such a way you're emulating twice the PCIe 3 lanes you had before. And that would make a big difference to the higher-end consumer systems.

I am still of the impression it won't really. I mean how many systems are PCIe lane starved just now? It may make a difference in a few years when things like NVMe start being more mainstream though I suppose.

 

[CuriousTomCat]

I am still of the impression it won't really. I mean how many systems are PCIe lane starved just now? It may make a difference in a few years when things like NVMe start being more mainstream though I suppose.

Here's an example of one

is there any news/rumours on whether the new X570 platform will allow more than 2 bandwidth restricted Nvme drives? Or would i have to wait for thread ripper.

I don't need more than 16cores but do need space for 4nvme drives without all the speed restriction, want to build a raid array of 4 drives. I would possibly go dual gpu next build so populating pci slots to accommodate NVMe drives is a no go.

my X58 system is really showing its age, cant wait for new AMD cpu/mobo

 

[LePhuronn]

I am still of the impression it won't really. I mean how many systems are PCIe lane starved just now? It may make a difference in a few years when things like NVMe start being more mainstream though I suppose.

Some existing X470 systems completely lose their 2nd GPU slot if you use more than 2 NMVe drives, or you can't use 2 at a full x4 speed. No matter how you slice it, right now there are limitations as to how you can distribute PCIe lanes, something which could be alleviated by bifurcating PCIe 4 lanes.

 

[Sargatanas2511]

The examples are good ones, but they do seem to be uncommon at best. More is never a bad thing though I will say that .

 

[LePhuronn]

Uncommon examples yes, but examples nonetheless. To be honest I don't subscribe to the hoohah about not being able to run a bazillion NVMe drives in RAID because if you have the need for that much throughput then you'd buy something like X299 or Threadripper. But certainly we have the potential to get around these limitations - however niche they may be - by playing fast and loose with how PCIe 4 can be implemented.

Of course it would set a tricky precedent though. We could get away with it because SLI/Crossfire is essentially dead and existing gaming cards aren't limited by PCIe 3, let alone anything else, but what happens if Turing 2 and Arcturus actually do max out a 16x PCIe 4 connection? All that lane trickery which was giving us multiple NVMe SSDs, 10Gb ethernet controllers and multiple GPU slots suddenly goes away, unless the 2020 CPUs start giving us a boatload more lanes to play with.

Realistically I think it's for this very reason of tempering expectations going forward we won't see too much PCIe 4 trickery on X570 and BIOS updates on X470.

 

[EsaT]

Not really...


And no Global Foundries' manufacturing node was not holding them down, Global Foundries' and Samsung collaborated on the 14nm node, 12nm was the first FinFet node that was all of Global Foundries' own doing.

Zen is high performance architecture, not some tablet level Jaguar like in current consoles, or Intel's Atom.

And GloFo's 7nm failure was just another in their row of failures to develop modern high performance node.
Their marketing may have sticked it with that name, but 14nm was designed and developed by Samsung for making mobile SOCs and then licensed by GloFo to get actually working 14nm something node.
Again "12nm" should have been named 14nm+...
While still falling short of clock speed increases inside Intel's 14nm.

Voltage scaling clearly puts optimal frequency low.
https://www.gamersnexus.net/guides/3290-exponential-ryzen-voltage-frequency-curve

GloFo Simply hasn't had any proper high performance node in long time.
AMD was just stuck with them, because of WSA agreements from spinning off AMD's factories when forming Global Foundries.
First when AMD was pretty equal share holder.
And then WSA was likely further reinforced as condition for AMD being able to sell most of their share.
AMD was literally obliged to buy at least certain amount of wafers from GloFo and also pay penalty for using other fabs.
https://www.overclockersclub.com/news/41948/

 

[4K8KW10]

Zen is high performance architecture, not some tablet level Jaguar like in current consoles, or Intel's Atom.

Here, I'd like to correct you and say that Zen behaves extremely well or even better at lower wattages - a Ryzen 5 2500U at only 15W is a beast. The architecture can be scaled down to 6-7W, no problem, while maintaining huge performance, multiple times that of an Atom.

 

[Murphy]

Zen is high performance architecture, not some tablet level Jaguar like in current consoles, or Intel's Atom.

Architecture and manufacturing process are two entirely different things.

Also...

Zen was designed to succeed both Excavator (High-performance) and Puma (Low-power) covering the entire range in one architecture

• Cover the entire spectrum from fanless notebooks to high-performance desktops
• More aggressive clock gating with multi-level regions
• Power focus from design, employs low-power design methodologies
  • >15% switching capacitance (CAC) improvement

And GloFo's 7nm failure was just another in their row of failures to develop modern high performance node.

I wouldn't exactly call it a failure, it was a choice of economics, going 7nm means large amounts of capital investment, something Global Foundries' didn't think was economically viable.

Their marketing may have sticked it with that name, but 14nm was designed and developed by Samsung for making mobile SOCs and then licensed by GloFo to get actually working 14nm something node.
Again "12nm" should have been named 14nm+...
While still falling short of clock speed increases inside Intel's 14nm.

While Samsung may have licensed a working 14nm to Global Foundries' that doesn't mean Global Foundries' didn't or wouldn't have their own working 14nm, in fact they were working on their own 14XM process, the only reason they licensed Samsung's 14nm LPP was because people weren't that interested in 14XM and it was late, they needed a working 14nm process and they need it fast so they could compete with the likes of TSMC.

Voltage scaling clearly puts optimal frequency low.

So it's a low power part then.

GloFo Simply hasn't had any proper high performance node in long time.
AMD was just stuck with them, because of WSA agreements from spinning off AMD's factories when forming Global Foundries.
First when AMD was pretty equal share holder.
And then WSA was likely further reinforced as condition for AMD being able to sell most of their share.
AMD was literally obliged to buy at least certain amount of wafers from GloFo and also pay penalty for using other fabs.

AMD and their WSA have nothing to do with the point your original made, that Zen hasn't been any super low power design any time, and that Global Foundries' manufacturing node has been holding it down.

Having said that what exactly is your definition of a "proper high performance node" as it's unclear what you mean by that.[/QUOTE]

 

[4K8KW10]

Just ahead of the weekend, prolific computer components leaker TUM_APISAK shared an intriguing graphic. The image is a breakdown analysis of a computer processor codename that had captured TUM's gaze, an AMD APU designated 2G16002CE8JA2_32/10/10_13E9. Thankfully, there was a catchier name for the component: AMD Gonzalo.

Above you can see the image shared by TUM and the decoded AMD codename. This image can be clicked on for a clearer, larger image. Using known AMD naming conventions it can be determined that 2G16002CE8JA2_32/10/10_13E9 is a second generation octa-core Zen processor. However, it isn't certain this is a Zen+ or Zen 2 architecture. The second character, G, suggests this is a gaming part, and elsewhere within the digits it is indicated that the CPU has a base/boost of 1.6/3.2GHz. For reference, the PS4 Pro tops out at 2.13GHz, and the processor in the Xbox One X at 2.3GHz.

Further along the character string is the designation 10_13E9 which TUM says represents the inclusion of Navi 10 Lite graphics cores. We are on less certain ground here but the GPU could come with a base clock of 1GHz.

Adding in a sprinkling of other rumours (and more salt) – for example that AMD Navi will launch at E3 in June, it is possible that we might hear about AMD Gonzalo at that time and we might get the first teases regarding the upcoming PlayStation 5 and Xbox Scarlett.

https://hexus.net/tech/news/cpu/126623-amd-gonzalo-apu-spotted-zen-2-plus-navi-10-lite/

 

[0h_no_not_again]

And GloFo's 7nm failure was just another in their row of failures to develop modern high performance node.

GloFo Simply hasn't had any proper high performance node in long time.

That is completely incorrect.

The GloFo 7nm was performing well in comparison to TSMC 7nm and was going to hit the market at a similar time.

Lack of investment killed GloFo 7nm, where they needed several Billion USD to hit economically viable scales. When 14nm/12nm demand is so high, why would short sighted investors chase thin 7nm margins?

I suspect the promise of EUV lithography being just around the corner helped put the nail in the 7nm coffin, as it is the last planned non-EUV node. Once EUV is more mature, it will make more sense to spend Billions tooling up for big volumes.

 

[AMDPower]

thanks for all the help gents, looks like i will have to sit tight and see what motherboard manufacturers will do in regards to Nvme drives. But in all likelihood thread ripper is a safer bet for me.

 

[Dirk Diggler]

What's the TLDR, is it gonna be upgrade time for my 2700x?

 

[Zeeflyboy]

Most likely a significant performance increase over your 2700X, but whether it’s worth replacing an already very capable processor is another question which only you can answer!