Speculation: Ryzen 3000 series

[beginner99]

They can also raise that percentage up or down, depending on how aggressively they target the frequencies (and my guess is that they will target them aggressively). That means there will be plenty of 4C salvage to go around.

Exactly. The quads don't need to have 4 defect cores but 4 poor cores that clock bad or need high voltage are also undesirable in EPYC or Ryzen >=8 cores simply due to heat and power restriction.

 

[moinmoin]

I wonder what the deal is with all the PRO models; are they just mostly aliases with maybe just the minor difference in whether the coprocessor is active for remote management ability?

They are business only counterparts to the non-PRO consumer chips. Hardware is exactly the same. They are differentiated by validation (ECC is officially supported), availability (chips have guaranteed availability until 3 years from launch) and unlocked security features in the AMD Secure Processor (full ARM TrustZone capabilities, transparently encrypted RAM like on Epyc).

 

[Fanatical Meat]

More rumor but interesting stuff.
Claimed 15% ipc improvement
Better memory support but not as improved as expected
a320 will not support new Ryzen


https://www.google.com/amp/s/wccfte...-uplift-4-5-ghz-x570-pcie-gen-4-40-lanes/amp/

 

[exquisitechar]

More rumor but interesting stuff.
Claimed 15% ipc improvement
Better memory support but not as improved as expected
a320 will not support new Ryzen


https://www.google.com/amp/s/wccfte...-uplift-4-5-ghz-x570-pcie-gen-4-40-lanes/amp/

Is that 15% IPC improvement or per core performance improvement? I know WCCFT says IPC, but I don't really trust them and the machine translated version of the original source is unclear.

And is the 4.5GHz boost all core or single core? Disappointing if it's single core, but not really surprising. There are a lot of things that point to 7nm HPC not bringing a huge bump in clocks, I suppose.

 

[Fanatical Meat]

Is that 15% IPC improvement or per core performance improvement? I know WCCFT says IPC, but I don't really trust them and the machine translated version of the original source is unclear.

And is the 4.5GHz boost all core or single core? Disappointing if it's single core, but not really surprising. There are a lot of things that point to 7nm HPC not bringing a huge bump in clocks, I suppose.

Not sure just interesting stuff.

I’m no AMD fab boy, I’m generally the opposite. AMD historically has had the ability to disappoint consistently.
AMD does appear to have its stuff together since last year but I’m not ready to trust them with a day one purchase *yet*

 

[Karnak]

And is the 4.5GHz boost all core or single core? Disappointing if it's single core, but not really surprising. There are a lot of things that point to 7nm HPC not bringing a huge bump in clocks, I suppose.

Well if it's an ES it's not final. First Bulldozer ES were like 2.8GHz iirc and the first Zen (Summit Ridge) ones were like 3.0GHz and we all know the 1800X with 4.0GHz boost.

If there's anything true about what wccftech is reporting (or the site they're refering to) and it's an ES then I think we'll see a 4.7-4.8GHz boost. Something like that. Just don't think Zen2 will hit 5GHz.

 

[Thunder 57]

Is that 15% IPC improvement or per core performance improvement? I know WCCFT says IPC, but I don't really trust them and the machine translated version of the original source is unclear.

And is the 4.5GHz boost all core or single core? Disappointing if it's single core, but not really surprising. There are a lot of things that point to 7nm HPC not bringing a huge bump in clocks, I suppose.

WCCFTech has actually been pretty good with Ryzen news, if you ignored the stuff that was clearly nonsense like 512MB L3 on Epyc.

15% IPC sounds about right, I think many are expecting 10-15%. My guess is that 4.5 would be single/dual core. Disappointing if true, I was hoping for closer to 4.8. It is their first shot on a new node though, so even if 4.5GHz is the limit at launch they should be able to coax a few hundred more MHz as time goes on.

 

[Thunder 57]

Well if it's an ES it's not final. First Bulldozer ES were like 2.8GHz iirc and the first Zen (Summit Ridge) ones were like 3.0GHz and we all know the 1800X with 4.0GHz boost.

If there's anything true about what wccftech is reporting (or the site they're refering to) and it's an ES then I think we'll see a 4.7-4.8GHz boost. Something like that. Just don't think Zen2 will hit 5GHz.

I guess it depends on how new the sample is. If is a newer ES and it really is launching in a month or two those clock speeds should be pretty representative as to what to expect. I think 4.7-4.8 is very possible though if this is accurate we may not see them at launch. I'd love to see 4.8 boost and all core boost of 4.5, but we shall see.

 

[Karnak]

I guess it depends on how new the sample is. If is a newer ES and it really is launching in a month or two those clock speeds should be pretty representative as to what to expect. I think 4.7-4.8 is very possible though if this is accurate we may not see them at launch. I'd love to see 4.8 boost and all core boost of 4.5, but we shall see.

Well, if it's a "newer" one than it would most likely be a QS (Qualification Sample) instead of an ES.

If there are QS around than we'll know the clocks of the final product since they're most of the time identical.

 

[Abwx]

AMD demoed a chip wich was running at 4GHz if we are to guess its frequency from the displayed power (60W) and it scored 2052 pts in Cinebench R15.

On the other hand Computerbase.de state that their Ryzen 7 score 1815 pts, at seemingly 4075MHz wich is about the higher boost frequency of a good chip, the aggregated maths say that perf/Mhz is apparently 15.1% better for Zen 2 (in CB..)...

 

[piesquared]

WCCFTech has actually been pretty good with Ryzen news, if you ignored the stuff that was clearly nonsense like 512MB L3 on Epyc.

15% IPC sounds about right, I think many are expecting 10-15%. My guess is that 4.5 would be single/dual core. Disappointing if true, I was hoping for closer to 4.8. It is their first shot on a new node though, so even if 4.5GHz is the limit at launch they should be able to coax a few hundred more MHz as time goes on.

Yeah how disappointing. That .9% lower clock speed than you were expecting must be a deal breaker. I know what you mean though, I was expecting 4.5002 GHz so naturally I was also disappointed.

 

[Thunder 57]

Well, if it's a "newer" one than it would most likely be a QS (Qualification Sample) instead of an ES.

If there are QS around than we'll know the clocks of the final product since they're most of the time identical.

Semantics . I was a bit more optimistic since they gained 300MHz from 14nm to 12nm, which is basically 14nm+. I also thought with 7nm they were targeting more for performance than 14nm LPP which was always designed more for low power, even if it served it's purpose rather well. I thought we'd see at least another 300MHz because of that. 4.5 sounds a bit low. But like I said, even if they only launch at 4.5 I'm sure they will ramp it up some by the end of the year or early next near.

No support on A320 seems a bit odd. I doubt anyone actually really bought those, however. Maybe they were more concerned about cheap power delivery than the chipset itself. And did I ready that right, no PCIe 4 on B550? That wouldn't be a good move. Also makes me wonder if any mobo manufacturers would bother trying to enable PCIe 4 on the top x16 slot on current B450/X470 boards if that is the case.

 

[piesquared]

AMD demoed a chip wich was running at 4GHz if we are to guess its frequency from the displayed power (60W) and it scored 2052 pts in Cinebench R15.

On the other hand Computerbase.de state that their Ryzen 7 score 1815 pts, at seemingly 4075MHz wich is about the higher boost frequency of a good chip, the aggregated maths say that perf/Mhz is apparently 15.1% better for Zen 2 (in CB..)...

That 15.1% better IPC is meaningless when you have a .9% lower clock than expected! Plus, the huge improvement in perf/watt is also disqualified due to said disappointment of .9% lower clock.

 

[Fanatical Meat]

AMD demoed a chip wich was running at 4GHz if we are to guess its frequency from the displayed power (60W) and it scored 2052 pts in Cinebench R15.

On the other hand Computerbase.de state that their Ryzen 7 score 1815 pts, at seemingly 4075MHz wich is about the higher boost frequency of a good chip, the aggregated maths say that perf/Mhz is apparently 15.1% better for Zen 2 (in CB..)...

This right here is why I love AT forums. Some big brained people here.
Seriously that is simple & genius

 

[Thunder 57]

Yeah how disappointing. That .9% lower clock speed than you were expecting must be a deal breaker. I know what you mean though, I was expecting 4.5002 GHz so naturally I was also disappointed.

That 15.1% better IPC is meaningless when you have a .9% lower clock than expected! Plus, the huge improvement in perf/watt is also disqualified due to said disappointment of .9% lower clock.

Would you stop putting words in my mouth and instead be productive if you are going to post? I am eagerly awaiting Zen 2, and might actually upgrade from a 2600X for the boost in AVX performance alone. I'll see if the x265 benchmarks are faster enough to warrant it.

Disappointing may sound like a strong word, but I didn't mean it to sound like it will be a failure in any way. I had modest, reasonable expectations (I thought) compared to some out there that expect a 5.1GHz 16 core behemoth. Imagine their disappointment if this leak is true. I just made an argument as to why I was expecting more than 4.5GHz. However, this is their first CPU on 7nm so even if that is the top clock at launch they will surely do better as the process matures, like with Zen+.

 

[Justinbaileyman]

I'm one of those hoping for a 16c/32t cpu and dont expect it to be at a 5ghz frequency.
I'm personally keeping my fingers crossed for say a 3.6Ghz base with a 4 - 4.2 Ghz boost.

 

[Thunder 57]

It will be interesting to see how much was done with IF. The Zen cores themselves are ridiculously efficient, but IF can be a power hog. It does scale up well, but still burns power when not used much. I like to think it's high floor, low ceiling as far as power usage goes. Compare the following graphs for an 8700k vs 2700X:

The 8700k sips power (uncore) then scales up. The 2700X starts off high and then pretty much hits a ceiling.

Also see this power chart that just came out recently comparing server chips:

See what I mean about the "high floor" power usage of IF, whereas Intel's mesh has a low floor. It is worth noting that IF has a much lower ceiling though, using 100+ watts less despite having more cores and way more PCIe.

I guess what I am trying to say is that if AMD can cut the IF power to say 10W going up to 15-20W they could either benefit from having a lower TDP or ideally higher clock speeds. I don't quite think we will see power usage of IF shrink though as it is supposed to be running faster in Zen 2 to begin with. These are just some thoughts. From the Anandtech article:

"After core counts, the next battle will be on the interconnect. Low power, scalable, and high performance: process node scaling will mean nothing if the interconnect becomes 90% of the total chip power. "

Sources:
https://www.anandtech.com/show/13124/the-amd-threadripper-2990wx-and-2950x-review
https://www.tomshardware.com/review...inum-8280-8268-gold-6230-amd-epyc,6058-6.html

 

[Thunder 57]

I'm one of those hoping for a 16c/32t cpu and dont expect it to be at a 5ghz frequency.
I'm personally keeping my fingers crossed for say a 3.6Ghz base with a 4 - 4.2 Ghz boost.

That sounds reasonable, maybe even a bit conservative. However, from what I have heard (and am inclined to believe), we will only see 12 cores at launch, with 16 very likely to follow later this year. It may depend on how successful the launch is and what Intel responds with. I'm certainly no insider and who knows how much has changed in the game of "telephone" before I heard what I did. Will be fun to see.

 

[Fanatical Meat]

That sounds reasonable, maybe even a bit conservative. However, from what I have heard (and am inclined to believe), we will only see 12 cores at launch, with 16 very likely to follow later this year. It may depend on how successful the launch is and what Intel responds with.

I sort of agree with this.

Heard Milton Berle on the Stern Show decades ago say “only take out enough to win”

 

[krumme]

32c server cpu is not bought to run and sit idle. The epyc problem if one can say so must be elsewhere eg. Performance consistency.

 

[Thunder 57]

32c server cpu is not bought to run and sit idle. The epyc problem if one can say so must be elsewhere eg. Performance consistency.

Of course not. What I was getting at is that IF clearly scales well in servers, but can it scale down and sip power like Intel's mesh can? It would be great to try to get the best out of both worlds but I do not think that is possible. I think the boosted AVX capabilities will fix a lot of EPYC's consistency problems.

Actually, looking further into it, it is really only Intel's ring bus that can sip power at lower usage. I probably should have included the mesh data, but here it is:

Also the scaling fooled me a bit but it is clear that IF is actually doing a very good job compared to the mesh. If anything this is bad news for Intel as they will have to move off the ring bus sooner than later as core counts grow.

 

[Justinbaileyman]

That sounds reasonable, maybe even a bit conservative. However, from what I have heard (and am inclined to believe), we will only see 12 cores at launch, with 16 very likely to follow later this year. It may depend on how successful the launch is and what Intel responds with. I'm certainly no insider and who knows how much has changed in the game of "telephone" before I heard what I did. Will be fun to see.

Kinda wish we knew this for certain, because if this is the case of only 12c/24t being released I would just upgrade to a 2950x now.
I need more and faster cores for video editing/rendering and this info that AMD is keeping secret is the only thing holding me back.
We are only like a little over a month away to release anyways so why doesnt AMD just spill the beans on what we're getting and when!!

 

[Tuna-Fish]

32c server cpu is not bought to run and sit idle.

You'd be surprised. For anything where the server running and having capacity is required for revenue generation, the systems are often sized for worst-case (err, best case, because load = revenue) peak load. Meaning that they spend >90% of their life at relatively low loads, simply because it's worth it to have a barely loaded server running for years just to make sure that on the peak days every customer can spend their money at your site.

This is a large part of why AWS is doing so well, btw. Having your own server at a colo facility is much cheaper than purchasing the equivalent server time from AWS if your load is consistent. However, if you cannot predict your load and can set it up so that AWS will spin up new instances on demand, the proper comparison can be the single cheapest AWS instance running 24/7 (+ a few dozen hours per year of running many more instances) to multiple expensive high-end servers.

 

[Thunder 57]

Kinda wish we knew this for certain, because if this is the case of only 12c/24t being released I would just upgrade to a 2950x now.
I need more and faster cores for video editing/rendering and this info that AMD is keeping secret is the only thing holding me back.
We are only like a little over a month away to release anyways so why doesnt AMD just spill the beans on what we're getting and when!!

I am hoping we hear something on May 1st. I wouldn't mind a paper launch with availability in June/Computex.

 

[amd6502]

I am hoping we hear something on May 1st. I wouldn't mind a paper launch with availability in June/Computex.

You're an optimist. I think hopefully end June availability and Computex launch. And hoping we get a look at the product line in a few days May 1. At least a decent outline. Basically: model numbers, cores/threads. Details like final clocks and retail price can wait till the launch.