Speculation: Ryzen 3000 series

[dnavas]

This, for me, clearly only talks about frequency, not stability. And it's still hard to believe ...

The rumors on sampled Epyc frequencies have always been low. I don't think it's so hard to believe that bringing up a brand new node will encounter a few bumps. The issue is that we've had a lot of hyped hopes wrt desktop frequencies, which makes looking at a 1.6G chip hard to believe.

 

[Topweasel]

That out of context and a response to Ashraf's:



If you look at the tweets after the one I quoted, the context is an other one:

This, for me, clearly only talks about frequency, not stability. And it's still hard to believe ...

It's super idiotic. Per core even the ES clocks are nearly on par with Intel on their highend CPU's. But in the end we are talking about a platform that is supporting 64core CPU's. If anyone thought they would be getting 3GHz clocks on those they were insane. Where AMD will have a real advantage is in the 4 chiplet up to 32c chips. Those are going to have crazy clocks with the power they will be allotted.

But this is Tiawan he is talking about. This isn't OEM's looking for shipping products to validate. This is platform designers with ES chips designing motherboards and platforms. They won't get the cream of ES chips. They will get low clocked chips there to test the platform in general.

 

[Zucker2k]

It's super idiotic. Per core even the ES clocks are nearly on par with Intel on their highend CPU's. But in the end we are talking about a platform that is supporting 64core CPU's. If anyone thought they would be getting 3GHz clocks on those they were insane. Where AMD will have a real advantage is in the 4 chiplet up to 32c chips. Those are going to have crazy clocks with the power they will be allotted.

But this is Tiawan he is talking about. This isn't OEM's looking for shipping products to validate. This is platform designers with ES chips designing motherboards and platforms. They won't get the cream of ES chips. They will get low clocked chips there to test the platform in general.

Won't it be a bit disingenuous to test on subpar ES chips? I mean, how do you guarantee functionality of your products running high clocked chips?

 

[exquisitechar]

I'd love to know what fairy dust AMD sprinkled on their ES at CES if neither their clocks nor IPC were on a par with the 9900K.
Unless someone is saying that Lisa Su blatantly lied about the Ryzen ES being an 8c CPU, either the clocks or IPC have to be ahead.

True, but we need more testing than just Cinebench. Maybe Francois doesn't consider slightly beating the 9900k in that benchmark or even in most workloads to be good enough to be a threat. He seems to think AMD won't stand a chance once Intel starts rolling out everything they've been forced to hold back because of the 10nm disaster, with the information he has access to.

Not saying I agree with him, but it will be interesting to look back at this at launch...

 

[Abwx]

And Francois Piednoel's take on Zen 2...hmm.

Panic at the disco...

He seems not aware of what AMD demonstrated not so long ago, a Ryzen at about 4GHz, but for some reasons AMD could make a 4GHz chip but not a 2GHz one...

Notice that their Epyc demo was at 1.8-2GHz and that the Blender test duration was about 30s.

I also learned that less cores is better when it comes to VM, does Piednoel really read what he writes..?

 

[PotatoWithEarsOnSide]

Less cores is great for VMs if you don't value security.

 

[Topweasel]

Won't it be a bit disingenuous to test on subpar ES chips? I mean, how do you guarantee functionality of your products running high clocked chips?

Not really. AMD has had a long long list of launches with sub par clocks on ES. Even in the days where clockspeed was the biggest thing in the world. AMD was sending test K8 Opeteron's out clocked at 1.4-1.8GHz and all the rumors were that AMD was struggling to get competitive clocks and yet they launched with 2.2GHz CPU's. AMD provides the specs needed for correctly running the CPU and gives them CPU's that fit comfortably within their requirements. It probably puts a little pressure on the OC orriented boards, but AMD's specs will leave a lot of leyway and these board designers have decades of experience working on performance boards. Why do you think we have a billion BIOS settings, it's to cover all ground that users may need to use to push limits after the fact.

 

[krumme]

Sounds like the stuff that's usually first fixed through microcode updates.

I would think its just the usual metal layer revisions. Its clearly not on transistor level or it would be delayed to 2020. Perhaps they need an extra metal level revision and that's the reason for mid 2019 release and not spring.

 

[Atari2600]

And Francois Piednoel's take on Zen 2...hmm.

Sure he is completely discredited given his opinion that Zen1 would be nowhere.

 

[Kocicak]

Taiwan is not impressed with what is coming from AMD

WOW. Whole Taiwan. What a disaster....

 

[Fanatical Meat]

WOW. Whole Taiwan. What a disaster....

I didn't get that either, makes no sense without saying whom or what in Taiwan doesn't like it.
I'd assume Intel's head of sales for Taiwan wouldn't like it, whom else is there to mention?

 

[teejee]

Francois Piednoel is the guy who managed to convince the tech world that he was like the most important engineer at Intel or something. This is very very far away from the truth.
He was a principal engineer at Intel, that is level 10 out of 13, not bad at all but far away from being in the top league. And there is not even a single patent mentioning his name so he has not been part of any important innovation at Intel.

Compare this to for example Per Hammarlund, former fellow at Intel (grade 12) who left for Apple in 2015. He was among other things responsible for Intels first Hyperthreading (in P4), the soul chief architect for Haswell and then chief architect of future cores. And he is part of a ridicoulus amount of patents as well. That was a major loss for Intel.


Intel engineering grades:
https://www.google.com/amp/s/levelsfyi.com/intel-levels-salary/amp/

 

[beginner99]

I didn't get that either,

because that is where TSMC is located that is fabbing the chips plus other tech companies that may or may not have zen2 ES for testing.

 

[DrMrLordX]

Let Francois hoist himself on his own petard. The idea that Rome can't compete with Intel CPUs from 2012 in VM tasks is ridiculous.

 

[Kocicak]

When I read some of his comments on Zen and AMD, seeing that he cannot check a very short message for typos before he posts it and that his announcement about leaving Intel has a grammar nonsense in it, I just do not take this person too seriously.

 

[Fanatical Meat]

because that is where TSMC is located that is fabbing the chips plus other tech companies that may or may not have zen2 ES for testing.

So who in Taiwan? The chip fab, some guy who works for the chip fab, every single person in Taiwan? The claim needs citation.

 

[Zucker2k]

So who in Taiwan? The chip fab, some guy who works for the chip fab, every single person in Taiwan? The claim needs citation.

More than likely, the motherboard makers.

 

[Markfw]

More than likely, the motherboard makers.

Motherboard makers will make more from socket AM4 than they ever have before. So why would they not like it ?

I think this guy is full of it.

 

[Topweasel]

Motherboard makers will make more from socket AM4 than they ever have before. So why would they not like it ?

I think this guy is full of it.

Unless it is MSI. They have made several public statements giving AMD a hard time in what looks like an attempt to gain favoritism by their primary partners in Intel and Nvidia in case markets dictate limited allocation. They would throw AMD AM4 sales out so quickly if they thought Intel would think of them more favorably.

 

[Fanatical Meat]

Motherboard makers will make more from socket AM4 than they ever have before. So why would they not like it ?

I think this guy is full of it.

Agreed it’s a weird claim without say who...

I’m no AMD fan boy, I was hugely disappointed with the first zen release, I’ve griped multiple times about my crappy FX6300 system, I’ve regretted building that FX6300 system instead of spending $100 more for a proper i3 system at the time.

 

[Thunder 57]

Agreed it’s a weird claim without say who...

I’m no AMD fan boy, I was hugely disappointed with the first zen release, I’ve griped multiple times about my crappy FX6300 system, I’ve regretted building that FX6300 system instead of spending $100 more for a proper i3 system at the time.

I thought the only one hugely disappointed with the release of Zen was that guy juanrga. Just out of curiosity, were you disappointed with the performance or the smoothness of it? Or something else, maybe pricing?

 

[Fanatical Meat]

I thought the only one hugely disappointed with the release of Zen was that guy juanrga. Just out of curiosity, were you disappointed with the performance or the smoothness of it? Or something else, maybe pricing?

Just the reviews at the time and I’ll accept some of the blame. I expected it to at minimal equal intels offering not equal them in *some* benchmarks. Just wasn’t what I was looking for at the time.
I still feel Ryzen needs work, the memory scene with it is confusing, irritating and expensive is you want the only memory that seems to work top tier with it.
This isn’t saying I think Ryzen sucks today, it’s just an irritating thing to deal with. Needing fast low latency memory to get the best performance and that memory is either no longer available because motherboard manufacturers haven’t kept up their compatibility matrixes or finding some good fast low latency memory and it costs like $100+ more. Stuff like that irritates me.

 

[IEC]

You don't *need* low latency memory for Ryzen. ADATA now makes (XPG) 2x8GB DDR4-3000 CL16 kits that are plug and play/certified on Ryzen. They have gone as low as $80 shipped directly from ADATA on Rakuten. That's like 95%+ of the performance at <50% of the price. They usually OC to 3133 or 3200 speeds as well...

Since Ryzen got more popular, memory and motherboard makers have improved compatibility of their products. I fully expect the 3000 series to be better versus the 2000 series, just as the 2000 series was better versus the 1000 series. It's an open question whether or not it will close the gap completely, however.

 

[Fanatical Meat]

You don't *need* low latency memory for Ryzen. ADATA now makes (XPG) 2x8GB DDR4-3000 CL16 kits that are plug and play/certified on Ryzen. They have gone as low as $80 shipped directly from ADATA on Rakuten. That's like 95%+ of the performance at <50% of the price. They usually OC to 3133 or 3200 speeds as well...

Since Ryzen got more popular, memory and motherboard makers have improved compatibility of their products. I fully expect the 3000 series to be better versus the 2000 series, just as the 2000 series was better versus the 1000 series. It's an open question whether or not it will close the gap completely, however.

That’s exactly what I’m hoping for Ryzen 3000 being more established
*Im obsessed about having 32GBd for my next machine. I know it’s illogical
It’s a wants vs needs thing

 

[Thunder 57]

Just the reviews at the time and I’ll accept some of the blame. I expected it to at minimal equal intels offering not equal them in *some* benchmarks. Just wasn’t what I was looking for at the time.
I still feel Ryzen needs work, the memory scene with it is confusing, irritating and expensive is you want the only memory that seems to work top tier with it.
This isn’t saying I think Ryzen sucks today, it’s just an irritating thing to deal with. Needing fast low latency memory to get the best performance and that memory is either no longer available because motherboard manufacturers haven’t kept up their compatibility matrixes or finding some good fast low latency memory and it costs like $100+ more. Stuff like that irritates me.

I think a lot of the memory stuff worked itself out with Zen+. I got 16GB of G.Skill RAM that is supposedly Samsung D die rated for 3200MHz. I put it in and tried XMP and it worked without any problems. Maybe I just got lucky, but memory support with Zen+ seems to be much less of an issue.

Zen was a clean slate design on a fresh node. That is tough to pull off. It also had a lot of catching up to do. It would have been great, but maybe impossible to have launched something like Zen+ on the first try. Zen+ is what the first Zen should have been, particularly with the smooth XFR/boost clocks. But I think it is a bit disingenuous to think of the Zen launch and being "hugely disappointed".