Ryzen 3000 Leaks

[The Benjamins]

22 minutes ago, GoldenLag said:

The best we will get om CES is a teaser, dont expect any extra. Probably going to announche mass production, would be surprised if we got anything else.

It looks like we will get a teaser Dec 18th. This is a Offical AMD contest to Guess the 3700X & 3600X C15 scores, ends 14th winner notification? 18th.

 

[GoldenLag]

2 hours ago, maartendc said:

I am going to say this is fake.

 

I think they will not up the core count so dramatically. As Hardware Unboxed has said

Some of Hardware unboxed critisisms were good, others not so much. It seems he took more from what people who were hyping said rather than what was said in the video. 

 

Its all still valid critisism though, the sources for these leaks werent the same as the more trusted sources Adore uses. Something he states very early im the video

 

Spoiler

I will take some critism myself as i might have taken these rumours a bit lightly. We will see how these mature over time and how thing will work out.

 

[AngryBeaver]

4 minutes ago, GoldenLag said:

Some of Hardware unboxed critisisms were good, others not so much. It seems he took more from what people who were hyping said rather than what was said in the video. 

 

Its all still valid critisism though, the sources for these leaks werent the same as the more trusted sources Adore uses. Something he states very early im the video

 

  Hide contents

I will take some critism myself as i might have taken these rumours a bit lightly. We will see how these mature over time and how thing will work out.

 

I am going to lean towards most of that information if not ALL being fake. Just too many pieces that don't add up. You know what they say about something that seems too good to be true.

[CarlBar]

4 minutes ago, GoldenLag said:

Some of Hardware unboxed critisisms were good, others not so much. It seems he took more from what people who were hyping said rather than what was said in the video. 

 

Its all still valid critisism though, the sources for these leaks werent the same as the more trusted sources Adore uses. Something he states very early im the video

 

  Reveal hidden contents

I will take some critism myself as i might have taken these rumours a bit lightly. We will see how these mature over time and how thing will work out.

 

 

Honestly the only things he was on point about where the TDP's, (counterargument: 9900K @95W TDP), and Pricing (Counter argument we've had past accurate leaks on specs include pricing, probably extrapolation on the lakers part or outright internal partner company estimates for forecasting purposes.

 

The rest range from poor to outright steaming pile of drivel that did some pretty serious harm to their credibility for me.

 

Their complaint about the RX3080 was especially stupid, (even though i consider the RX Navi stuff the sketchiest part). Just because the leaker claims competition with the RTX 2070 or it has similar performance  dosen;ty mean it actually is competing. Unless AMD pulla  suprise move and announce an equivalent to Tensor and Rt cores in Navi it;s going to have to undercut the RTX 2070 by a fair margin just to be equal value for money because of the reduced feature set. To actually undercut it it;s going to have to be cheaper still. 50% still is pretty high but it's within reason.

 

But most of their arguments had issues IMO.

 

I'm still cautious, (if slightly optimistic), about these leaks you understand but i don't think the criticisms where really mostly valid.

[Agosto]

4 hours ago, GoldenLag said:

The VRM at the price is still pretty good. Good enough for 8 core parts and a pretty OK overclock

3 phases with no doublers (just double the components, quite "meh" ones)

It also has no LLC as far as I can tell from owners' posts.

[GoldenLag]

5 minutes ago, Agost said:

3 phases with no doublers (just double the components, quite "meh" ones)

It also has no LLC as far as I can tell from owners' posts.

They arent great boards. But for the price their pretty good. These are sub 75$ motherboards afterall

[Taf the Ghost]

2 hours ago, GoldenLag said:

The best we will get om CES is a teaser, dont expect any extra. Probably going to announche mass production, would be surprised if we got anything else.

The Gigabyte Product Stack leak, which was from a presentation in Sept/Oct puts the X570 launch as Computex. So that's in line for May, which is about when we should expect these launches. It was March 2017 and April 2018, for Ryzen, so far.

 

I think the two things catching Tim off is #1, Jim didn't say these were the exact prices & are subject to change. The numbers look fine, if you assume Navi, especially, came back better than they expected. And, #2 and forgotten by nearly everyone, this isn't a single generation Node Shrink for AMD. This is much closer to Intel going from 22nm to 14nm++. It's about a 1.5 generation Node Shrink. The clocks are fully doable, especially if the efficiency curve is 400 Mhz higher for TSMC's 7nm process. (Also, early discussion of 7nm by AMD had 1.35x performance at the same power. Which means they binned down their parts a bit so they could expand the tiers.)

 

I would expect prices to move around a bit, but those are the categories and segments AMD can sell, on the Consumer side, into. AMD can't really sell high-end Gaming GPUs. That's been the problem they've had for years. If Navi came back really awesome, they'll move things up a tier, but they have to meet the segments they can sell into. AMD will make more money selling a potential 3080 at 250USD than at 400USD. However, I would expect AMD to bump that up to 299USD and give the retailers a better cut, like they did with the 1800X and Threadripper 1st Gen.

 

That's why I would agree with Jim's take on the leaks he got over Tim's analysis, but that's because of the economics of the situation that AMD is in. Though AMD is probably only going to announce release dates for the next line of products at CES, though maybe they'll bring Linus on stage to drop a GPU.

[Simon771]

I would prefer 8c/16t CPU with 5Ghz boost instead of 12c or even 16c.

[ravenshrike]

Thing is, everyone saying that it makes no sense for AMD to go to 16 cores at these prices is missing one salient point. Intel will have no response at all till Icelake, and that assumes significant IPC improvement. Without it AMD will have completely outplayed Intel in performance and price across the entire consumer product stack until Intel gets their 7nm EUV chips out at the earliest, and that assumes no further leaps with TSMCs 7nm+. This isn't like the early 2000s where Intel will be able to sweep that sort of thing under the rug. It will allow AMD to overwhelm the otherwise somnolescent OEM consumer market's sensibilities.

[Sharp_3yE]

The only thing I care about with this news is that fact that AMD is coming out with new processors that will be even better. Good for them. It helps everyone when AMD is doing good. I've even been contemplating whether or not to have a Ryzen CPU. It's a good price. Keep going AMD. You're doing pretty dang good.

[Phentos]

If the rumored specs for the "Ryzen 9 3850X" are accurate, I'm buying it day one and selling my current new rig to a buddy who needs a new rig for esports gaming.

 

The 9700K I have serves me well in Premiere Pro, but 16 cores at 5Ghz would be way more important.

[CarlBar]

6 hours ago, Taf the Ghost said:

The Gigabyte Product Stack leak, which was from a presentation in Sept/Oct puts the X570 launch as Computex. So that's in line for May, which is about when we should expect these launches. It was March 2017 and April 2018, for Ryzen, so far.

 

I think the two things catching Tim off is #1, Jim didn't say these were the exact prices & are subject to change. The numbers look fine, if you assume Navi, especially, came back better than they expected. And, #2 and forgotten by nearly everyone, this isn't a single generation Node Shrink for AMD. This is much closer to Intel going from 22nm to 14nm++. It's about a 1.5 generation Node Shrink. The clocks are fully doable, especially if the efficiency curve is 400 Mhz higher for TSMC's 7nm process. (Also, early discussion of 7nm by AMD had 1.35x performance at the same power. Which means they binned down their parts a bit so they could expand the tiers.)

 

I would expect prices to move around a bit, but those are the categories and segments AMD can sell, on the Consumer side, into. AMD can't really sell high-end Gaming GPUs. That's been the problem they've had for years. If Navi came back really awesome, they'll move things up a tier, but they have to meet the segments they can sell into. AMD will make more money selling a potential 3080 at 250USD than at 400USD. However, I would expect AMD to bump that up to 299USD and give the retailers a better cut, like they did with the 1800X and Threadripper 1st Gen.

 

That's why I would agree with Jim's take on the leaks he got over Tim's analysis, but that's because of the economics of the situation that AMD is in. Though AMD is probably only going to announce release dates for the next line of products at CES, though maybe they'll bring Linus on stage to drop a GPU.

 

Not only that, but he goes on about how they don't finalize a lot of stuff until much closer to launch when they get the silicon back. Except since these are using the same CPU dies as EPYZ and since they're about to start shipping EPYC they've allready got the dies back, they allready know how the silicon performance and yields are. In fact for the Ryzen 3, R5 top end and R7's they're almost certainly binning the silicon for those right now as they're all using chiplets with cores disabled, all of which they'll be culling from EPYC stuff where all 8 cores don't work/meet specs.

 

The only thing stopping AMD launching Ryzen 3 and Threadripper 3 right now is lack of CPU chiplets and maybe the IO dies not being complete.

[Taf the Ghost]

5 minutes ago, CarlBar said:

 

Not only that, but he goes on about how they don't finalize a lot of stuff until much closer to launch when they get the silicon back. Except since these are using the same CPU dies as EPYZ and since they're about to start shipping EPYC they've allready got the dies back, they allready know how the silicon performance and yields are. In fact for the Ryzen 3, R5 top end and R7's they're almost certainly binning the silicon for those right now as they're all using chiplets with cores disabled, all of which they'll be culling from EPYC stuff where all 8 cores don't work/meet specs.

 

The only thing stopping AMD launching Ryzen 3 and Threadripper 3 right now is lack of CPU chiplets and maybe the IO dies not being complete.

AMD most likely had the first test silicon for 7nm Zen2 and Vega 20 in December 2017. That's part of why they can go Server First on this node jump. Main thing is the motherboards, especially the PCIe 4.0 ones, won't be ready until further into 2019. It's going to be a design jump for the motherboard manufacturers, so they can't just redo a lot of previous designs. AMD probably could have launched Zen2 Desktop for Holiday 2018 if they had been really pressed to get something out. However, they aren't. 

 

As a subtle note, generally products don't make public roadmaps until they've got at least some test silicon back. I wonder how many actually functioning Icelake-SP dies Intel has actually gotten back. Must not be a lot.

[CarlBar]

12 minutes ago, Taf the Ghost said:

AMD most likely had the first test silicon for 7nm Zen2 and Vega 20 in December 2017. That's part of why they can go Server First on this node jump. Main thing is the motherboards, especially the PCIe 4.0 ones, won't be ready until further into 2019. It's going to be a design jump for the motherboard manufacturers, so they can't just redo a lot of previous designs. AMD probably could have launched Zen2 Desktop for Holiday 2018 if they had been really pressed to get something out. However, they aren't. 

 

As a subtle note, generally products don't make public roadmaps until they've got at least some test silicon back. I wonder how many actually functioning Icelake-SP dies Intel has actually gotten back. Must not be a lot.

 

ah we know from various partners that even as recently as the EPYC announcement they didn't have production grade silicon back, but they were in the home stretch of getting there, and should be there by now.

[schwellmo92]

12 hours ago, maartendc said:

I am going to say this is fake.

 

I think they will not up the core count so dramatically. As Hardware Unboxed has said, AMD claimed 14% improvements at the same power. So they can likely increase the clockspeeds by 14% and call it a day. Together with some other improvements in architecture.

 

This is a revision of the same ZEN architecture. If they would so drastically increase both clockspeed and core counts, they would need a total overhaul of the architecture.

 

And it is DEFINITELY fake because they are listing prices. Even AMD does not know prices so far away from launch I would presume.

The slides I saw were 25% performance increase at same power - can you provide proof of this 14%?

[Taf the Ghost]

6 minutes ago, CarlBar said:

 

ah we know from various partners that even as recently as the EPYC announcement they didn't have production grade silicon back, but they were in the home stretch of getting there, and should be there by now.

Given just the amount of packaging going into Rome, they'd have a huge supply of Zen2 dies already, but the testing cycle was always going to be pretty long for this type of product. Though I think the main take away from this cycle is that, unlike Zen1, AMD isn't rushing. They're making good money & margin on Zen products, so they can let those keep rolling until they drop the big bomb in Intel's lap.

[CarlBar]

1 hour ago, Taf the Ghost said:

Given just the amount of packaging going into Rome, they'd have a huge supply of Zen2 dies already, but the testing cycle was always going to be pretty long for this type of product. Though I think the main take away from this cycle is that, unlike Zen1, AMD isn't rushing. They're making good money & margin on Zen products, so they can let those keep rolling until they drop the big bomb in Intel's lap.

 

What i meant was the silicon they'd sent ot for basic review was full of flaws and they where going to have to do some major revishions, but that silicon wasn't out in the wild yet for testing by select end users. They've got to have good silicon back by now as those same select customers are actually putting their orders in, so they have to have been given final specs by this point. 

 

As far as Zen2 Die's, oh sure now they're presumably into production i don't doubt they've got a lot of dies, but aside from parts where they'd have to disable some of the CPU's on the chiplet none of the stuff is usable for Ryzen because EPYC gets priority and they're still ramping up availability just to meet initial select partners demand, let alone get ready for full general release after that. They might have a lot of chiplets, but it's going to be nowhere near as many as they actually need to get EPYC out the door, never mind Ryzen or Threadripper.

[MikeSK]

Realistically, if this is not fake news, when should we expect this technology to be for sale for Threadripper? Early 2019, mid 2019, 2020? Also, can we expect a massive market inflation over MSRP like the GTX 1080 ti?

[IceCold008]

My next build was going to be an intel build but if these specs for AMD new CPU's are true. I might have to jump from blue to red lol. Benchies first like. 

[Crunchy Dragon]

1 hour ago, MikeSK said:

can we expect a massive market inflation over MSRP like the GTX 1080 ti?

The problem with the 1080Ti and GPU prices as a whole was due to mining and simple supply and demand.

[AngryBeaver]

Anyways just wanted to add this. While I have provided lots of reasons why these leaks are likely fake.

 

If they happen to be even remotely true I will be ordering parts for a new build about 15 minutes after they are available. I have a 2080ti that I still haven't done anything with and I am itching to put it in something. I will need ideas on a nice theme for my build. I normally don't do anything theme related, but I think me going back to AMD after so long might warrant it.

[ravenshrike]

RGTs source was separate from AdTV's, and specifically doesn't work for AMD directly, but a third party. He didn't come out with names or clockspeeds, just that the minimum numbers of cores for the top end was going to be 12, and they were shooting for 16. This leads me to believe one of three things. AdTV' s list is an internal mockup of how they want to bin the lineup based on early returns, someone else heard the same rumors and created their own binning list, or something finally clicked in the 7nm production process and supply is building a lot faster than it was a month ago, giving them a sufficient supply of chips in order to start nailing this stuff down.

[Unixsystem]

7 hours ago, ravenshrike said:

RGTs source was separate from AdTV's, and specifically doesn't work for AMD directly, but a third party. He didn't come out with names or clockspeeds, just that the minimum numbers of cores for the top end was going to be 12, and they were shooting for 16. This leads me to believe one of three things. AdTV' s list is an internal mockup of how they want to bin the lineup based on early returns, someone else heard the same rumors and created their own binning list, or something finally clicked in the 7nm production process and supply is building a lot faster than it was a month ago, giving them a sufficient supply of chips in order to start nailing this stuff down.

The Adored leak specifically stated that the 16 core parts wouldn't even be announced until May, and who knows when they'd release. So both could technically be true - they wanted 16 at launch, but now they are keeping them on the back burner until yields/stock are high enough for it not to eat into their highly binned Epyc chiplets.

[Mihle]

 

Wait and see how correct these leaks ends up being

[Prox1ma]

I'll believe it when I see it. It's getting overhyped and I personally think its too good to be true. We all saw how the Vega hypetrain ended. Set your expectations low and you will not be disappointed. Ryzen 3000 maybe at best on par with CL refresh in terms of IPC, maybe a tad bit higher in core count. Navi seems unbelievable. I doubt they would even be able to match up to 2070 level performance at best.