Speculation: Ryzen 3000 series

[DrMrLordX]

Hah, I've got some of those on my enermax. At full speed, they are definitely NOT quiet!

Indeed. Mine are going into retirement on my Matisse system (going wc, probably). They have served me well though.

Noise is all a matter of perspective. Some of those little rinky-dink fans can be kind of annoying. I hope the X570 board manufacturers are mindful of such. Either that, or a lot of us are going to be running Matisse on good X470 boards . . .

 

[scannall]

Yeah I figured the grease pencils could amount to a problem if they fragmented. They still saved a lot of money just using those upfront. I'm surprised they actually got ahold of the Fisher pens eventually.

Fisher developed the pens on their own dime. No cost to Nasa. Then sold them for $6.00 each to Nasa and the Russian Space Agency.
Fisher Space Pen

 

[amd6502]

386sx to 3600x, how much of an upgrade would it be. OC'd at 25MHz right now.

 

[Thala]

386sx to 3600x, how much of an upgrade would it be. OC'd at 25MHz right now.

That really depends on the workload. Having no FPU in the 386 means that you are going to feel the speedup much more with floating point heavy applications

 

[amd6502]

That really depends on the workload. Having no FPU in the 386 means that you are going to feel the speedup much more with floating point heavy applications

Yeah, just joking; I sadly don't have access to this machine anymore (no idea where it went). It was my first build, took it to college in '91, and it wasn't long till I installed BSD on it and also tested linux on it. It did soon get a 387 coprocessor and actually eventually was used for some number crunching. Anyway, looking at this, and then at something like the 3000 series over 25 years later is quite something. Many order of magnitude increase; can't even do the math off the top of my head.

 

[jamescox]

With all the cpu power at hand at lowish tdp I think the time has come to help the climate - lol
With a 3700x 8c at 65tdp that's what my boy teen gets next time. He simply burns electricity like crazy as he plays 144hz 1440 and never accepts anything going below 120Hz while he watch a YouTube video or 2 and keep contact on social media with 50 tabs.
Perhaps a change of his Vega 56 too but I am a bit more sceptical here for the efficiency of the new gpu as ddr6 power usage vs hbm 2 is a lot to compensate for.

Yes, they have doubled the cache, which should also cut memory accesses significantly. Previously the L3 caches couldn't really share data between the two CCX. Geekbench seems to indicate (for a while now) that there are still 2 separate instances, but if they can snoop data from each other with low latency, we could be talking about up to 4x the effective L3 cache size.

Regardless, even 2x cache increase will diminish the need for memory accesses significantly

Interesting. 4GHz base, 3.9GHz average turbo?
X370 mobo used.
Memory latency 60ns.

It is great if they got the memory latency down, but there are so many other things in play that we can’t tell to much without just getting the test results in real applications. The cache hit rate should be significantly higher with double sized caches and the improved hardware prefetch can make big differences also. They have made many other improvements.

The CCX to CCX latency should be massively improved over Zen 1. In Zen 1, CCX to CCX communication has to go through the IF switch at memory clock. With Zen 2, there is no memory clock on the cpu chiplet, so I would expect the CCX to CCX communication on die to happen at core clock, possibly with the switch being double the internal width compared to Zen 1.

Going to the other chiplet in the 12 core should be slightly higher latency but it should be much better than Zen 1. The IF serdes clock is more than double. The clock speed of the switch in the IO die could be decoupled from the memory clock and it could be significantly higher clocked than in Zen 1. The side effect, if that is true, is that Zen 2 will be significantly less sensitive to memory clock compared to Zen 1.

It will be interesting to see the results of 12 core Zen 1+ ThreadRipper vs. 12 core Ryzen 3000. It is double the L3 cache, but with Zen 2 possibly having significantly lower latency to remote L3, it could make an even larger difference than would be expected just from the size. The 12 core has 64MB L3. The top single die Xeon is 38.5 MB, and most of those cost thousands of dollars.

 

[Abwx]

"This is an error we are correcting. Pre-X570 boards will not support PCIe Gen 4. There's no guarantee that older motherboards can reliably run the more stringent signaling requirements of Gen4, and we simply cannot have a mix of "yes, no, maybe" in the market for all the older motherboards. The potential for confusion is too high.
When final BIOSes are released for 3rd Gen Ryzen (AGESA 1000+), Gen4 will not be an option anymore. We wish we could've enabled this backwards, but the risk is too great."​

AMD​

https://www.computerbase.de/2019-06/pcie-4.0-amd-mainboard/

 

[beginner99]

"This is an error we are correcting. Pre-X570 boards will not support PCIe Gen 4.

​

Or simply a reaction to the critical reception of the chipset fans which could have impacted sales of x570 boards if PCIE4 worked on x470.

 

[DrMrLordX]

​

Or simply a reaction to the critical reception of the chipset fans which could have impacted sales of x570 boards if PCIE4 worked on x470.

Even if X470 had supported PCIe 4.0, it still would have only worked for the PCIe slot fed from the CPU. Of course now we'll never know for sure how well it would actually work.

Makes you wonder how PCIe 5.0 will pan out on consumer hardware.

 

[Abwx]

​

Or simply a reaction to the critical reception of the chipset fans which could have impacted sales of x570 boards if PCIE4 worked on x470.

The risk they are talking about is that if they state that PCIe4 can be enabled on previous MBs they (AMD and OEMs) would be held responsible for any MB that wouldnt work accordingly, and thus could be sued if they do not either provide a replacement or pay a compensation.

Methink that it s far more concerning than eventual lower sales of X570, it s not that they wouldnt cash something from X470s sold instead.

 

[beginner99]

Makes you wonder how PCIe 5.0 will pan out on consumer hardware.

I doubt it will find it's way there anytime soon because of cost and lack of need.

 

[Despoiler]

PCIe4 doubles the signalling rate compared to 3. To do this the max trace length had to be halved. Nothing is for free. This has nothing to do with selling boards and everything to do with customer experience. How would you like to buy a Navi GPU only to find out your graphics are unstable because the link keeps failing? Do you think the average consumer is going to think highly of AMD given they have no idea how the technology works?

 

[bsp2020]

AMD should not have mentioned that they may enable PCIe4 in older platform at all. It just confused people.
I guess they are new to milking their customers. They should stop over-delivering to customers who already paid...

 

[Topweasel]

Yeah this isn't to keep people from buying X470 boards to avoid the chipset fan. This is a "well it's feasible but not nearly as reliable as we hoped". Add in certain manufacturers like MSI that is forgoing a lot Zen 2 and PCIe 4.0 support on 4** and older you add a significant amount of confusion. This might be a compromise to level the playing field to keep people from getting confused. Now that MSI doesn't have to work on PCIe support in these boards maybe they will open some of their older stuff to Ryzen 3k.

 

[PotatoWithEarsOnSide]

I bet they don't.

I can understand why AMD backtracked here. As mentioned above, it is all about liability. If they endorse enabling PCIE4 on a PCIE3 spec board, they will open themselves to potential lawsuits for any number of reasons.
It is probably the right call, but a little disappointing nonetheless.

 

[Atari2600]

Yeah - pity they don't make an unofficial unsupported bios available.

Like overclocking. Here it is, run at your own risk.

 

[joesiv]

Yeah - pity they don't make an unofficial unsupported bios available.

Any board that has BIOS updates available prior to this new decision being rolled out might have the now unsupported feature. Go look for your Beta BIOS's now before they're pulled! It's almost like they did make unofficial support available with the early announcement.

 

[Shivansps]

Nah, as i say AMD is getting greedy, they know they are getting the upper hand now so there is no need to play as the good guy anymore.

They KNOW they had to support PCI-E 4.0 ON THE M2 because it is really close to the CPU socket and in some cases it is even closer than the main pcie, and PCIE 4 M2 is the main selling point of PCI-E 4.0 as no much people cares about the other ports so X570 with its active fan would be a hard sell.

No to mention the artificial limitation of 1st gen Ryzen and Raven on X570 and they went back on their promise for 300 chipsets.
This is the result of AMD being on top now, and i expected this.

 

[Kenmitch]

And so the conspiracy theory begins..../s

 

[Snarf Snarf]

My money is on the OEM Board partners getting upset at AMD for promising things they either won't or can't deliver. They will likely do QC testing on their top tier x370 and x470 motherboards but the rest of the lineup is going to be SOL I feel.

 

[Shivansps]

And so the conspiracy theory begins..../s

There is no conspiracy here, just business practices, most likely, AMD wanted to do that, but then OEMs told AMD that supporting PCI-E 4.0 is a bad idea because that was the main selling point of X570.

The "they cant have 'yes, no, maybe'" is a excuse, they are doing exactly that for 300 series support, why they cant do the same for PCI-E 4?

 

[fleshconsumed]

Nah, as i say AMD is getting greedy, they know they are getting the upper hand now so there is no need to play as the good guy anymore.

They KNOW they had to support PCI-E 4.0 ON THE M2 because it is really close to the CPU socket and in some cases it is even closer than the main pcie, and PCIE 4 M2 is the main selling point of PCI-E 4.0 as no much people cares about the other ports so X570 with its active fan would be a hard sell.

No to mention the artificial limitation of 1st gen Ryzen and Raven on X570 and they went back on their promise for 300 chipsets.
This is the result of AMD being on top now, and i expected this.

I don't think it has anything to do with AMD being on top now. Most likely it's about limited resources. AMD is a small company, chances are resources they would need to allocate to work with OEM manufacturers to support PCIe4 on legacy motherboards are better used elsewhere working on new products.

It would have been nice to have PCIe4 on x470 motherboards, but realistically speaking PCIe3 is plenty fast for everything that I do nowadays to the point where I'd actually prefer cheaper fanless X470 over more expensive fanned X570.

 

[Thunder 57]

I don't really see a problem with this. They probably thought it could work and would be good PR. Now that they have final silicon and Navi, maybe they found it it didn't work as well as they had hoped. Now they have to deal with the bad PR. They probably should have kept quiet until they knew for sure.

What I think is more problematic is the possibility that Zen 2 might not work in X/B300 boards. It looks like that may not be true, or it may be on a board by board basis. That to me is more damning than not gaining a feature that did not exist when X/B400 came out.

 

[jpesk2]

Looks like a new userbenchmark for a ryzen 5 3600: https://www.userbenchmark.com/UserRun/17389094 DDR2400

Here it is compared to the 8700 (non-k) https://cpu.userbenchmark.com/Compare/Intel-Core-i7-8700-vs-AMD-Ryzen-5-3600/3940vsm810675 . Looks like a great performer. It will be nice to see what shakes out in the reviews. I think AMD did a bang up job -- good for them, good for all of us!!

 

[PPB]

I said that 200cm2 is enough for the chipset while all you did was to display a cooler that has barely this surface yet are claiming that it s enough for 13.5W albeit at 87°C, that s all you stated so far, i see no formal evaluation of yours, hence your tendency to use a derogative language as a pathetic mean to compensate for this lack.

Now getting back to the matter and say 200cm2 sized plate, wich is not as much as what Gygabyte used, and check how much it can dissipate at an ambiant of 300°K (27°C).

The plate has two sides], so its total radiative surface is 400cm2
[/QUOTE]

Strike one. You calculation starts wrong assuming air renovation for the surface facing the PCB is the same as the surface not facing the PCB. So you dont really have 400cm2 when employing 200cm2 surface on the PCB.

Besides that, do you really realize how big is 200cm2 worth of surface area? A 60x80mm heatsink like the one I showed is already big in ATX standards, and that is less than 1/4 of that.

If you cant fit 200cm2 in a motherboard for starters, all your following napkin math goes to the drain.

The Gigabyte motherboard you like to talk so much about also needs to house at least 2 PCI-E 4 NVME SSDs under that big PCH heatsink. So it's not really only a PCH heatsink only, as the inner surface will eventually get heated by those 3 NVME SSDs whose power consumption is at least the same of current PCI-E 3 NVME SSDs, which is 6-8w each, but judging by the oversized ssd controller heatsinks they are showcasing the new pci-e 4 models, it will likely be more than that.

So we are not talking about only 11w to dissipate, more like +25W because if the PCH gets heated for starters, is because it's running heavy I/O on the PCI-E 4.0 SSDs to begin with. Not even mentioning the fact that we are talking about a 500+ EATX motherboard, plain ATX, mATX and ITX be damned fitting 200cm2 worth of a heatsink so i dont hear a fan, am i right?.

200cm2 heatsink surface area just so you dont hear a fan is plain stupid and we can only thank you are not designing these motherboards as you neglect worst case scenarios regarding air renovation and inner case temperature in your calculations.

Anyway we ll soon see the MB production tested, methink that this chipset TDP frenzy will end as it begun, abruptly...

The chipset tdp frenzy was started by the nostalgics and or lunatics that got triggered by seeing a 50mm fan and went as far as to suggest everyone should copy the Gigabyte's design I linked above like it's a good design practice to trap the heat of your NVME SSDs plus other pcb components under such a massive PCH heatsink plate design.

I certainly couldn't care less, as neither i'm buying a x570 board because ITX x570 seems as cramped as never before and even if it were, the fan spinning is the least of my worries as I know how better small fans got in the noise department the last 10 years and it's MTBF will well beyond what my use case for that motherboard would be.