Speculation: Ryzen 3000 series

[Kenmitch]

I'm not sure where this idea that Ryzen has much lower IPC than 8/9th gen Intel comes from. Intel has a clock advantage, and a very minor ST IPC lead; MT IPC they are behind marginally.

No fun when you spin it that way. /s

 

[Markfw]

I remember when Ryzen launched and how MB manufactures had so much trouble with their drivers and bios. So this time I'm sure they have been working hard on getting it right at launch this time.

Since they are launching a product on the same socket, and it supposed to be compatible, it should be a lot easier than the first time, just a bios update for vcore and to recognize the chip.

 

[realibrad]

Since they are launching a product on the same socket, and it supposed to be compatible, it should be a lot easier than the first time, just a bios update for vcore and to recognize the chip.

I thought they were using a new/modified memory controller? I would imagine that they would have to do some changes given that no?

My knowledge of that is next to nil, so I could be getting that wrong.

 

[Mopetar]

I remember when Ryzen launched and how MB manufactures had so much trouble with their drivers and bios. So this time I'm sure they have been working hard on getting it right at launch this time.

I don't think it should be as much of an issue this time around. Zen was completely new (and I forget if AM4 was also new) so there were bound to be a lot more issues. This time there's a solid existing code base and an existing product pipeline that can be tweaked as necessary. I think that things will go much more smoothly.

 

[realibrad]

I don't think it should be as much of an issue this time around. Zen was completely new (and I forget if AM4 was also new) so there were bound to be a lot more issues. This time there's a solid existing code base and an existing product pipeline that can be tweaked as necessary. I think that things will go much more smoothly.

I hope so. Ryzen came and shook the industry. Its major sticking point was drivers/bios. Considering it was all new it helped take the edge off of the early issues. If this time they focus on giving the MB makers time to refine their bios then we should see a much better launch from AMD.

I have a wedding come up in a few months, but, I really want to pull the trigger on a high frequency chip from AMD.

 

[DrMrLordX]

I remember when Ryzen launched and how MB manufactures had so much trouble with their drivers and bios. So this time I'm sure they have been working hard on getting it right at launch this time.

AMD had a lot less trouble with the X470 release. I wouldn't expect things to be as rocky this time around as they were in 2017.

 

[Topweasel]

AMD had a lot less trouble with the X470 release. I wouldn't expect things to be as rocky this time around as they were in 2017.

Yeah but X470 was just a minor change to X370 as developed by Asmedia. AMD is back to developing their own chipsets for this gen. Time to relearn a lost practice.

 

[Markfw]

Yeah but X470 was just a minor change to X370 as developed by Asmedia. AMD is back to developing their own chipsets for this gen. Time to relearn a lost practice.

These chips are supposed to work in all existing AM4 motherboards (with a BIOS update) so how can this be a big update ?????

 

[Thunder 57]

Yeah but X470 was just a minor change to X370 as developed by Asmedia. AMD is back to developing their own chipsets for this gen. Time to relearn a lost practice.

The most recent rumors suggest x570 will be developed by AMD while ASMedia will handle any others. I would expect AMD to borrow heavily from x470, but I guess we shall see.

 

[Topweasel]

These chips are supposed to work in all existing AM4 motherboards (with a BIOS update) so how can this be a big update ?????

It's designing a chip from the ground up. Socket compatibility doesn't mean drop in support for the chipset itself and even if it did it could be as different as a Pentium vs. K6 vs. Cyrix vs. Centaur for Socket 7. Internal design and drivers can be completely different (because different designers).

 

[Markfw]

It's designing a chip from the ground up. Socket compatibility doesn't mean drop in support for the chipset itself and even if it did it could be as different as a Pentium vs. K6 vs. Cyrix vs. Centaur for Socket 7. Internal design and drivers can be completely different (because different designers).

I don't think so. To use the same socket, all the traces have to go to the same places. Its a minor update. The major update is the chip, not the motherboard.

 

[DrMrLordX]

The most recent rumors suggest x570 will be developed by AMD while ASMedia will handle any others. I would expect AMD to borrow heavily from x470, but I guess we shall see.

I would be more concerned about who did the IMC design than the motherboard. Other than the AGESA firmware, the boards themselves aren't much more than a southbridge these days. X470 is just X370 with a cleaned-up layout. X570 doesn't have to be much different. It's all about the UEFI.

 

[exquisitechar]

Four different ASUS Crosshairs!

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

 

[inf64]

Four different ASUS Crosshairs!

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

Let's see what he will say in July. How much will intel charge for their 18C monster SKUs when AM4 platform will have similar core counts and performance for a fraction of the price.

 

[dacostafilipe]

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

With the performance/power usage we saw vs the 9900k, this tweet is really hard to believe.

 

[PotatoWithEarsOnSide]

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.

 

[beginner99]

X570 doesn't have to be much different. It's all about the UEFI.

I would hope that x570 supports a pcie3 link to cpu and hence faster or more IO options.

 

[DrMrLordX]

I would hope that x570 supports a pcie3 link to cpu and hence faster or more IO options.

It's a nice thought. Not sure what that would do for compatibility with older AM4 boards though, if Zen2 can take advantage of that.

 

[Tuna-Fish]

With the performance/power usage we saw vs the 9900k, this tweet is really hard to believe.

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.

His later tweet was

Let me rephrase ... nobody seems to be able to finish a run of GB at full speed.

This implies that the chips are unstable at target frequencies when running varied loads. This might be caused by one critical path that is only used by some rare instruction being a bit too long and causing instability when the CPU is ran at high performance with code that uses it. This is normal, in the sense that it has happened to both Intel and AMD at times. You could run a good benchmark despite this, if you just pick a benchmark that doesn't happen to stumble on the failing instruction.

In the short term, for the chips already manufactured, the fix is to run at a low enough clock speed that there is time for the critical path to complete. This might be okay for high-density server parts and not okay for desktops. In the longer term, this is typically fixed by redesigning that one offending instruction to have a cycle more latency, relaxing the timing paths. Doing this usually takes a full stepping, so a ~3-5 month delay, depending on how much they need to rework and how much they need to validate the changes. The same issue, but when less severe, is why the second stepping after many major CPU releases has gotten a substantial clock boost.

It might well be that AMD noticed this issue the second they got their first silicon out, and that's precisely why they can already show good benchmarks yet are only releasing in the summer. In the mean time, they have a lot of the faulty/slow chips that are otherwise fine that they can sprinkle around mb manufacturers for testing and validation purposes.

 

[moinmoin]

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

 

[Topweasel]

I don't think so. To use the same socket, all the traces have to go to the same places. Its a minor update. The major update is the chip, not the motherboard.

That's like saying the Nforce 2 and AMD chipsets back in the day we're exactly the same. Having the same socket is one thing doesn't mean the traces are exactly the same, just means that the Mobo makers have to design a new board instead of recycling older designs. What matters is the socket pinout is the same. Chipsets haven't always been drop in compatible.

 

[Asterox]

Let's see what he will say in July. How much will intel charge for their 18C monster SKUs when AM4 platform will have similar core counts and performance for a fraction of the price.

What can you expect, from a "guy that proudly on Twitter shows a sticker and photo with Pentium 4 HT Extreme".

 

[coercitiv]

The most recent rumors suggest x570 will be developed by AMD while ASMedia will handle any others. I would expect AMD to borrow heavily from x470, but I guess we shall see.

The last rumor based on industry sources I saw was AMD will migrate their Epyc chipset towards x570 (with minor tweaks/updates along the way).

 

[dacostafilipe]

His later tweet was

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

Indeed, my point is that there are samples of the Matisse part that are running at or near target frequency.

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

Taiwan is not impressed with what is coming from AMD, and I am told that it is trying very hard to get it going, Frequencies are just nowhere close to be a thread, neither IPC.

Especially in Cloud environment, the high core count drives you to have very pathetic frequencies for Virtual machines, When 8 core VMs, they are not even competitive with a VM from 2012 from Intel ... That is a huge problem. High core count is a big mistake for VMs.

Sooner than later, share holders will understand that, it core count has been taken too far, making all of those parts totally uncompetitive ... especially on the cloud.

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

 

[scannall]

Lots of crazy rumors out there, on both the high and low side. Personally, I'm content to just wait for them to be shipped and benchmarked. Speculation is fun and all, but real numbers are what matters most.