Speculation: Ryzen 3000 series

[Abwx]

Storage? Do consumers really read/write that much from multiple NVMe drives at once?

Something else I'm not aware of? Answers on a post(card) please!

Bandwith of a PCIe4 nvme is up to 60% higher...

https://www.computerbase.de/2019-05/erste-pcie-4.0-ssds-corsair-gigabyte-phison-e16/

That s released opportunistically..

 

[B-Riz]

I suspect there is no 16-core because all full dies go to epyc. they can offer epyc in 8-core increments eg. chiplets. Ryzen gets the broken dies 6-core and 12-core and the high leakage once as the 8-core.
Since only 1 core needs to be defunct to make it a 6-core (same number of cores per ccx) such an allocation makes sense. I supect also a 6-core sku (or even quad) with less cache.

I don't think there are that main broken dies that the can offer Epyc with partially active chiplets and my doubt chiplets can have different number of active cores, that would be akkward. Since then they would need to disable 2 cores per chiplet with 8 chiplets it adds up quickly. Easier to just remove entire chiplets.

I seriously do not understand the heated angst about not getting 16/32 on AM4. What we get are still just server cores that were not good enough to be server chips.

I do not think it would be an ideal setup, too many compromises, as TR3 16/32 at a lower price point with more features and quad channel RAM seems like a better fit.

 

[Topweasel]

I suspect there is no 16-core because all full dies go to epyc. they can offer epyc in 8-core increments eg. chiplets. Ryzen gets the broken dies 6-core and 12-core and the high leakage once as the 8-core.
Since only 1 core needs to be defunct to make it a 6-core (same number of cores per ccx) such an allocation makes sense. I supect also a 6-core sku (or even quad) with less cache.

That obviously not the case there is a $330 and $380 products that use full spec dies. Inevitably they will sell more then the 12c chips. At even just $600 they get better margins just because No extra CPU package (pins/substrate/installation), no extra IO Die, no extra retail packaging. Sales would be a lot lower over the 12c and 8c chips anyways. It's probably because of several things. 1. Power usage (getting good clocks at reasonable power usage maybe be hard), 2. For that reason Binning (need to get sufficient stock of dies that can run at desired clocks within desired power) 3. ASP. AMD for whatever reason desided that they didn't want to push 8c pricing much lower then what we were already paying. They have 3 choices for a 16c chip. Push everything down to slot into $500 price point and lower ASPs. Another would be to push into Threadripper pricing with it ($600+). The final and what they seemed to have decided is to cap AM4 to ~$500. So cut out the 16c option while they build stock on the binned chips. This gives them a CPU in the pocket for Comet Lakes release or any movement on competitive pricing from Intel. When it comes out they can slide the 12c solution down to $400 and slide everything else down $30 or so.

Now personally I said and still think withholding on the 16c part is a bad idea. But there is wiggle room there. But I don't think they are in a great position to be withholding and should still be going down the "disruptive" war path. But this might be a good move for the start of 7nm when everything is at its most expensive and keep the 6c parts in better position financially (like it might be a bad idea to come out of the gate with those guys sub $200).

 

[zinfamous]

I'm not super enthused about the chipset fan thing. I've had pretty much every one of those I've ever had die on me even with the most careful dust maintenance lol.

~11W, seems like you would be able to have Cooler Master or somebody make a little vertical heatpipe/fin design (almost like a baby 212), and cool it passively if you have good case airflow. It will be interesting to see what kind of mods come about. It's still really early, I'll stick with 470 for now since I already have it, and don't have any 4.0 gear anyway.

But I bet we see some clever workarounds for it. I hope, lol. At least it would have been nice if they made the damn things standard across every 570 board so the fans would be common and easy to swap. But no, they look vastly different across makes/models, bleh.

From some post earlier in this thread or perhaps another one, it was suggested/reported (supposedly from MSI?), that the fan spools up only under high M.2 SSD use. That heatsink covers the 2x PCIe M.2 slots (PCIe 4.0?), so it seems logical that this is the primary use for the fan. If you're the type that needs very many cores, and the professional bits in a 570 class board, it's already known from earlier generations that those pro NVMe SSDs can throttle pretty heavily under persistent load and a generous amount of heat.

I wondered about this before reading that comment (though my thought was that this would be a detriment to the chipset--sharing a heatspreader with the 2x M.2 slots, but it seems to be the opposite?), and it does seem to be tied to the M.2 slots more than the chipset. Still, one wonders if separating the chipset heatsink from the heatspreader/sinks for the M.2 slots would be a better design, especially if it can remove the fan entirely. As Markfw also mentioned, there has to be a better way to do this with shared heatpipes (and possibly a copper spreader over each M.2 slot.). Leave the chipset with a solid heatsink to be cooled with case fans.

 

[beginner99]

That obviously not the case there is a $330 and $380 products that use full spec dies.

Yeah but "high leakage" eg high clocking and epyc wants low leakage as clocks are less important and power use a lot more important.
AMD must have planned for a 16-core or else making the IO die be capable of 2 chiplets for just 1 sku seems kind of a bad trade-off.

 

[jpiniero]

AMD must have planned for a 16-core or else making the IO die be capable of 2 chiplets for just 1 sku seems kind of a bad trade-off.

Even looking at the Epyc package, a CPU chiplet link was designed for a pair of CPU chiplets.

 

[zinfamous]

no its not,

it lacks pci-e 4

wont buy a new board without it, well I personally don't need it but I would like to have it

apparently not. news was out last week or the one before that many mainstream and pro x470 boards (I think also 450 boards?) have bios flashes out now or upcoming that will allow pci-e 4.

 

[Panino Manino]

Will we ever see CPUs with odd core numbers again?
Wouldn't make sense, a Ryzen 7 with 7 cores in 7nm?

 

[Rifter]

Just on PCIe4 - what are the use cases that justify (require) it over PCIe3?

It was shown a couple of years back that PCIe3 x8 had a negligble difference compared to PCIe3 x16; I cannot see the fundamental change that would drive the need for PCIe4 in graphics.

Storage? Do consumers really read/write that much from multiple NVMe drives at once?

Something else I'm not aware of? Answers on a post(card) please!

newer yet to be released GPU's, NVMe drives, and maybe large HBA cards would be the only thing i can think of to utilize pcie4.

 

[moinmoin]

I don't remember AMD even mentioning XFR during the Zen 2 presentation, and this is a feature that is bound to stay, especially considering Intel introduced their own version with Thermal Velocity.

Will be interesting whether they go back to the approach in 1st gen Ryzen where XFR added up to 200MHz on top of the boost clock spec and essentially enforced a OC headroom that way.

I suspect there is no 16-core because all full dies go to epyc. they can offer epyc in 8-core increments eg. chiplets. Ryzen gets the broken dies 6-core and 12-core and the high leakage once as the 8-core.
Since only 1 core needs to be defunct to make it a 6-core (same number of cores per ccx) such an allocation makes sense. I supect also a 6-core sku (or even quad) with less cache.

I don't think there are that main broken dies that the can offer Epyc with partially active chiplets and my doubt chiplets can have different number of active cores, that would be akkward. Since then they would need to disable 2 cores per chiplet with 8 chiplets it adds up quickly. Easier to just remove entire chiplets.

The only reason there's no 16 core yet is because 12 cores are sufficient for now.
Looking at the layout I highly doubt Epyc/Threadripper will come in chiplets+IOC configurations different from 8+1 and 4+1.

 

[Shivansps]

AMD is waiting for Intel to launch its 10C to launch the 16C one... a bit of pointless for Intel to launch 10C now.

 

[Paul98]

I am quite impressed especially with this being the first iteration and on a new process. I am really looking forward to benchmarks. Going to have to debate if I get the 12core now or wait for a few months for when I expect to see the 16 core. That single thread performance and 12 cores is very tempting.

 

[Fanatical Meat]

AMD is waiting for Intel to launch its 10C to launch the 16C one... a bit of pointless for Intel to launch 10C now.

As I said much earlier in the thread.
I once heard Milton Berle say “only take out enough to win”
That statement applies here too.

 

[jpiniero]

AMD is waiting for Intel to launch its 10C to launch the 16C one... a bit of pointless for Intel to launch 10C now.

The point of the 10 is more so they can give the i3/i5/i7 HT.

I don't expect Intel to go more than 10 while still on the Ring, regardless of what AMD does.

 

[fleshconsumed]

You can see a fan if you look closely

Unfortunately, if you look at the head-on image in the gallery, it appears that there is a fan lurking behind the heatsink next to the PCIe slots.

Hopefully someone (Thermalright?) will release a passive sink suitable for retrofitting on X570 boards.

Yeah, you guys are right. I found a better picture from an expo and after adjusting brightness it does appear to have a fan. I want to buy a new xX70 motherboard, and it looks like I'm going to have to choose between x570 with fan or x470 without full PCIe4 support. I hate noise with a passion and seeing how PCIe4 native support is the only selling point of x570 I'll try to find x470 on sale. Shame.

 

[IEC]

I'm thinking those chipset fans won't spool up except under extreme load (e.g. extreme nVME RAID). As long as it's silent or close to it under normal operation I don't think a fan is a negative. I have a X399 board that has a small fan near the VRMs, for example. Almost never hear it spool up except at boot.

 

[JasonLD]

That obviously not the case there is a $330 and $380 products that use full spec dies. Inevitably they will sell more then the 12c chips. At even just $600 they get better margins just because No extra CPU package (pins/substrate/installation), no extra IO Die, no extra retail packaging. Sales would be a lot lower over the 12c and 8c chips anyways. It's probably because of several things. 1. Power usage (getting good clocks at reasonable power usage maybe be hard), 2. For that reason Binning (need to get sufficient stock of dies that can run at desired clocks within desired power) 3. ASP. AMD for whatever reason desided that they didn't want to push 8c pricing much lower then what we were already paying. They have 3 choices for a 16c chip. Push everything down to slot into $500 price point and lower ASPs. Another would be to push into Threadripper pricing with it ($600+). The final and what they seemed to have decided is to cap AM4 to ~$500. So cut out the 16c option while they build stock on the binned chips. This gives them a CPU in the pocket for Comet Lakes release or any movement on competitive pricing from Intel. When it comes out they can slide the 12c solution down to $400 and slide everything else down $30 or so.

Now personally I said and still think withholding on the 16c part is a bad idea. But there is wiggle room there. But I don't think they are in a great position to be withholding and should still be going down the "disruptive" war path. But this might be a good move for the start of 7nm when everything is at its most expensive and keep the 6c parts in better position financially (like it might be a bad idea to come out of the gate with those guys sub $200).

If this was 2005, it would make sense to go "disruptive" on desktop lineup, and Intel would have been panic mode already and would find a way to squeeze 16 cores on mainstream lineup by any means possible. But this is 2019, Desktop sales have fallen in half compared its peak years decade ago, and will keep declining. It has gotten to the point where Intel seems like they are about to skip 10nm for desktop and it doesn't look as devastating as it would have been 10 years ago. Basically, going down on "distuptive" war path on desktop market would be pointless and would only cut into their potential profits. AMD is probably smart to use their best 8-core dies on EPYC since that is where the growth lies.

 

[ozzy702]

Yeah, you guys are right. I found a better picture from an expo and after adjusting brightness it does appear to have a fan. I want to buy a new xX70 motherboard, and it looks like I'm going to have to choose between x570 with fan or x470 without full PCIe4 support. I hate noise with a passion and seeing how PCIe4 native support is the only selling point of x570 I'll try to find x470 on sale. Shame.

Why not wait until there are reviews out. Yes those fans are small, but they may not have to spin very fast to provide enough cooling. I don't like the idea of fans on a mobo either, but hopefully the execution is solid and they aren't noisy.

 

[amd6502]

He didnt fool me, its just annoying he fooled so many.

Not just annoying, but completely mind boggling. The cult of Jim had their day or reckoning 5/27/2019 and are dissolved.

Will we ever see CPUs with odd core numbers again?
Wouldn't make sense, a Ryzen 7 with 7 cores in 7nm?

Odd cores were much anticipated.

https://www.reddit.com/r/Amd/comments/bso26r/_/eop4oce

Maybe for Ryzen 4000 series. To be more serious, I could see 3300g being 3c/6t (maybe with half L3) now that there is a native dual core APU.

So with the 7nm launch they actually have a full range. A basic line, that would more or less be okay and complete by itself, with 12nm second gen APU quads covering from Ryzen 3 to Ryzen 5 (with 3400g and 3200g), and 7nm covering from Ryzen 5 (3600) to Ryzen 9 12c.

If they did nothing further it would be a complete line in itself. But about everyone expects they're going to improve on it and fill some of the small gaps, and augment the top end to 16c with a new flagship high bin.

So what are the gaps. Well, there is a big perf gap between 3400g and the low bin 7nm 3600. Also big gap between the 3200g and 3400g. So 3300 and 3500 is what's missing.

3300g lower freq bin APU with 6 to 8t.

3500 line could be filled well with Pinnacles 8c/16t low freq bin, as well as non-SMT variety of Pinnacles 8c/8t (with freqs slightly above 2600's) targeted at gamers, as well as the relatively rare 7nm 4c/8t die salvage, once enough stock gets accumulated (~November?).

This would make the 3000 line diverse and have something for everybody.

Well, the big picture of the 3000 gen is here on the table. What comes in Q4 will just be refinements and small details.

 

[Topweasel]

I am quite impressed especially with this being the first iteration and on a new process. I am really looking forward to benchmarks. Going to have to debate if I get the 12core now or wait for a few months for when I expect to see the 16 core. That single thread performance and 12 cores is very tempting.

This has become my debate. Was thinking of replacing my old 4770k and R 270x ITX setup. Had 3 different avenues to go, 1 Upgrade my Main R7 1700 system, move the 1700 to new small PC. But if I did that I would want a 16c CPU. The other would have been a new in this case 3800x in the ITX system and the 16c part in the desktop. But if that's not coming any time soon I can just wait till it and TR3 is released. So with this I have 2 systems one I would love more cores in and one I would like more speed in. Do I get the 3800x or the 3900x or do I get both. Right now I am leaning towards completely new build in the ITX system with a 3900x. Hold out till next year for the 16c part or maybe even 32c depending on how TR3 shows and if I go TR3 it would be a more of a full blown refit of the system.

 

[amd6502]

Do I get the 3800x or the 3900x or do I get both. Right now I am leaning towards completely new build in the ITX system with a 3900x. Hold out till next year for the 16c part or maybe even 32c depending on how TR3 shows and if I go TR3 it would be a more of a full blown refit of the system.

Will these be water cooled?

3700x looks like it's run semi-close-enough to peak perf/watt while 3800x is being pushed to the minimum perf/watt. With the dual chiplet 3900x at least the wattage is split up, so maybe 2x ~42 watts, which is close to what the 3700 chiplet is seeing. The 3800x seems like a significantly lesser binned chiplet that's being overvolted and with high amps gushing through.

 

[Topweasel]

Will these be water cooled?

3700x looks like it's run semi-close-enough to peak perf/watt while 3800x is being pushed to the minimum perf/watt. With the dual chiplet 3900x at least the wattage is split up, so maybe 2x ~42 watts, which is close to what the 3700 chiplet is seeing. The 3800x seems like a significantly lesser binned chiplet that's being overvolted and with high amps gushing through.

Actually I got an H100i unused off of someone really cheap that I have been waiting to use. It will be a more open setup then my current one, but it will be on top of desk to I need to control everything but deepness. Otherwise I would be going with the 3700x which is really great clocks for the power usage. Case I am currently looking at is the Define Nano S.

 

[amd6502]

Nice. I think the 3700x looks better than the 3800x (and my guess is it might overclock better too, but will have to wait a month or two for reviews and concensus on how the 3700x overclocks).

 

[B-Riz]

OMG!!!! More Zen2 coming at E3!!!!

 

[Topweasel]

Nice. I think the 3700x looks better than the 3800x (and my guess is it might overclock better too, but will have to wait a month or two for reviews and concensus on how the 3700x overclocks).

Getting kind of lazy when it comes to overclock. Specially with how XFR2 worked on Ryzen 2k, I am not sure it will be worth it and 40w at load means nothing. But thats the tough spot if I am using a 105w CPU in that system I might as well pack in the cores, but if I was getting a CPU as expensive as the 3900x I would rather get a 16c and put that into my main desktop. Guess I will wait for the reviews and see what the multicore clocks will be for these guys. 50/50 says that the 3800x will keep the clocks up really high and at 105w is right there with a stock 9900k. It might be why Intel felt they needed to start stockpiling chips that run 5GHz all core (lord knows what the power usage will be like) not to get throttled with AMD's 8c choices in multi-threaded benches.