Speculation: Ryzen 3000 series

[Fanatical Meat]

yes exactly, it looks excellent because it has different cooling designer for it, you wont have a passive cooled cpu that powerful as r3000 and adding pci-4 to cpu with some minor power makes sense
imagine your new ipad gets a fan for not that much needed features....you wont like it
about that nvme raid, sorry I dont need one, nvme current drives are imo performing well on the desktop without raid
I dont know about your opinion, but those x570 boards look "over" cooled, overengineered and expensive
but yes I am waiting for the reviews, actually I am very interested in what is needed to get that r3000 above 4,5GHz - boards, cooling, power etc...
I am pretty sure about 4-4,2 GHz it will have massive efficiency, lets see in the reviews

Well said especially the over engineered part.
I’m thinking in a similar way. I have no need or desire I really likely any future need for nve raid, honestly I’ll probably do a big ssd for my next build and leave a super fast nve drive for a future upgrade.
I was originally going to build something on the first day or week of release, now I’m thinking waiting until August simply to see more reviews.
I’m not totally against a 570 board I just want an honest comparison between let’s say a good 470 board like the tiachi and a good 570 board. How much noise does the fan add, is the fan needed for standard(ish) operation and is the faster pci lanes perceivable better.
Paying an extra $30-$50 for a 570 board isn’t the concern.

 

[DrMrLordX]

I always wondered why theStilt left.

It seemed to be people arguing with him over stuff that, to him, wasn't arguable. Can't really blame him since he is an "in the trenches" kind of guy instead of a forum warrior arguing from extrapolations of seemingly-disparate data points. Pity we don't have him around for analysis of Matisse.

 

[Thunder 57]

It seemed to be people arguing with him over stuff that, to him, wasn't arguable. Can't really blame him since he is an "in the trenches" kind of guy instead of a forum warrior arguing from extrapolations of seemingly-disparate data points. Pity we don't have him around for analysis of Matisse.

Well he seemed to make a name for himself, considering he is mentioned in my BIOS/UEFI.

 

[lixlax]

Decided to run few benchmarks before and after the may Windows 10 update to see if perfomance really changed on my Ryzen 1700, videocard is Vega 56. Updated the chipset and graphics drivers in the process.

Firestrike:
Saw a noticeable increase in physics, combined and overall score.
https://www.3dmark.com/3dm/36715719
https://www.3dmark.com/3dm/36731041

3dmark06 (for single threaded cpu limited comparison): everything was within margin of error.

Shadow of the Tomb Raider (1440p, low preset, ingame benchmark): Both DX11 and 12 results were within margin of error.

I guess the results are dependent of the application/game used or maybe I did something wrong (ofcourse 3 tests in total isn't enough either to get a clear picture).

 

[Hans de Vries]

16 core GB result.

Very impressive at only 4.3GHz.

Best AMD windows 10 GB4 multi core score ever by far:
http://browser.geekbench.com/v4/cpu/search?dir=desc&q=amd+windows&sort=multicore_score

 

[zinfamous]

Could be a case of WSA terms and them needing to buy a certain number of wafers. That seems the most likely to me, but there could be other reasons as well.

probably the main reason. They continue to make good on the agreement, efficiently (cost) purchasing wafers at whatever makes the most sense for each die--maybe 14nm for the Ryzen 3, maybe 5, and "RR" class dies that eventually will be released (do those actually use the IOC?) and the 12nm for Ryzen 7, 9 (will it be 9?) and maybe 5.

 

[Markfw]

Best AMD windows 10 GB4 multi core score ever by far:
http://browser.geekbench.com/v4/cpu/search?dir=desc&q=amd+windows&sort=multicore_score

There is something wrong here. A series one threadripper 1950x 16 cores@4 ghz has 47k , while my second gen TR 24 cores has 31k ? And a 3rd gen 16 core has 61k ? whats wrong with this picture ?

 

[Kenmitch]

There is something wrong here. A series one threadripper 1950x 16 cores@4 ghz has 47k , while my second gen TR 24 cores has 31k ? And a 3rd gen 16 core has 61k ? whats wrong with this picture ?

Maybe 16c/32t is the sweet spot. What are the clocks on the TR 24? Could be the clocks and 16c32t sweetspot tainting the results.

 

[Hans de Vries]

There is something wrong here. A series one threadripper 1950x 16 cores@4 ghz has 47k , while my second gen TR 24 cores has 31k ? And a 3rd gen 16 core has 61k ? whats wrong with this picture ?

Mostly a Windows problem:
http://browser.geekbench.com/v4/cpu/search?dir=desc&q=amd+2970wx&sort=multicore_score

 

[Markfw]

Mostly a Windows problem:
http://browser.geekbench.com/v4/cpu/search?dir=desc&q=amd+2970wx&sort=multicore_score

I downloaded the linux version on my 2990wx, and I unzipped the files. There are 3 files, but none of them execute. (not a real linux person)

 

[Hitman928]

I downloaded the linux version on my 2990wx, and I unzipped the files. There are 3 files, but none of them execute. (not a real linux person)

What are the 3 files?

 

[Rifter]

It seemed to be people arguing with him over stuff that, to him, wasn't arguable. Can't really blame him since he is an "in the trenches" kind of guy instead of a forum warrior arguing from extrapolations of seemingly-disparate data points. Pity we don't have him around for analysis of Matisse.

Yeah its really to bad he's gone.

 

[guachi]

It's pretty easy to understand why they picked it when you look at the results it gives. The Intel product ends up looking like hot garbage and the AMD product ends up looking almost impeccable.

Sure it's pointless for the vast majority of people, but the whole point of these presentations is marketing your product.

It's basically a visual representation that AMD's chip has a lot of horsepower leftover to do other things while you game. That's a great selling point.

 

[loafbred]

It does alright in most games but I can get hitches sometimes during games like battlefield especially.

This is common on later versions of Windows 10, and it will continue with a new cpu. If you create the script described in the link below, it will disappear:

https://www.reddit.com/r/Windows10/comments/7a0763

Make sure you update Windows to version 1903 first. If not, you'll have to delete the script and remake it after updating.

 

[DownTheSky]

https://twitter.com/x/status/1138789078986502145

 

[realibrad]

https://twitter.com/x/status/1138789078986502145

That seems strange to me. I would think that there are times when you might want a very high frequency single core. If all cores boost to their max given the listed constraints, then you will see a lower clocked single core as the other active cores will create heat that will lower the max frequency of all cores. I have to believe that there is some way that the chip sees the workloads and then will do something workload specific. Why would you want to boost all cores when only one or 2 cores are needed.

 

[DrMrLordX]

There is something wrong here. A series one threadripper 1950x 16 cores@4 ghz has 47k , while my second gen TR 24 cores has 31k ? And a 3rd gen 16 core has 61k ? whats wrong with this picture ?

Might be a memory speed/performance issue? GB4 is funny like that. I try not to take those scores too seriously unless the memory configs are nearly identical.

That seems strange to me. I would think that there are times when you might want a very high frequency single core.

We'll have to do some SuperPi runs on Matisse and see what it does. In theory it should be able to boost a bunch of cores or just one at the same time, and it won't matter since only one core will actually be doing anything. You can achieve some pretty exceptional clockspeed on cores that are at or near 0% utilization.

 

[naukkis]

That seems strange to me. I would think that there are times when you might want a very high frequency single core. If all cores boost to their max given the listed constraints, then you will see a lower clocked single core as the other active cores will create heat that will lower the max frequency of all cores. I have to believe that there is some way that the chip sees the workloads and then will do something workload specific. Why would you want to boost all cores when only one or 2 cores are needed.

No more cores are boosted than what is needed for workload. AMD just don't have strict limits of clocks per cores used which are needed for limit max power peaks -if they can handle them other way there's no need to limit clocks per cores used.

If only 1-2-cores are needed all power headroom can be given to those cores and boost them as high as they could, limit is how much voltage can be used to chase ultra-high clocks. Same max voltage can also be given to boost more cores, only when temperature or power limits are met they have to start scaling voltage and clocks down.

 

[moinmoin]

The I/O coupled with the chiplet in the SKU use 12nm while the X570 chipset is using 14nm.

That doesn't answer the question: why?

I always wondered why theStilt left. I figured it had to do with backlash with his reported Zen+ power numbers, but I guess it was something else.

It was my fault. He put the blame for Windows not being prepared for Threadripper 2 WX topologies solely on AMD, and I called him out on that referring to the zero issues Linux has with that topology. He then referred to the issues Windows has/had with Bulldozer's CMT topology, and I spoke my thoughts that he's distracting from the initial discussion.

No more cores are boosted than what is needed for workload. AMD just don't have strict limits of clocks per cores used which are needed for limit max power peaks -if they can handle them other way there's no need to limit clocks per cores used.

If only 1-2-cores are needed all power headroom can be given to those cores and boost them as high as they could, limit is how much voltage can be used to chase ultra-high clocks. Same max voltage can also be given to boost more cores, only when temperature or power limits are met they have to start scaling voltage and clocks down.

Indeed. And the Windows 10/UEFI optimization of allowing Zen cores to change speed within 1-2ms (instead ~30ms) now should increase the amount of scenarios where actually very few cores are actively in use.

 

[zinfamous]

That seems strange to me. I would think that there are times when you might want a very high frequency single core. If all cores boost to their max given the listed constraints, then you will see a lower clocked single core as the other active cores will create heat that will lower the max frequency of all cores. I have to believe that there is some way that the chip sees the workloads and then will do something workload specific. Why would you want to boost all cores when only one or 2 cores are needed.

well, AMD now plays 4, no 6, no 8, no 16, no 64 dimensional Chess, really. Get with the times and your 1 dimensional Chess, bro.

 

[.vodka]

That seems strange to me. I would think that there are times when you might want a very high frequency single core. If all cores boost to their max given the listed constraints, then you will see a lower clocked single core as the other active cores will create heat that will lower the max frequency of all cores. I have to believe that there is some way that the chip sees the workloads and then will do something workload specific. Why would you want to boost all cores when only one or 2 cores are needed.

Maybe this will clear it up for you.

See that tweet translated to a picture, XFR2's magic:

This is what differentiates the 3700x and the 3800x, one is a 65w part while the other's ceiling is 105w. Same as 2700 (65w) and 2700x (105w). See how beautiful the boost algorithm is? It doesn't matter if you have a dual core or a 16 core part, low threaded workloads will always run at or near the advertised turbo speed if the operating conditions allow it.

Keep it cool, give it a good motherboard with a decent VRM and let it do its thing. Expect it to be even more polished and better at making the most out of the designated TDP bracket it's working in on 3rd gen parts.


1st gen parts and their behavior, in comparison, are from the stone age. This, and the fact that max clock speed is ~4GHz anyways makes them more suited to an all core overclock with p-states since you don't have any control over XFR behavior as you do on second gen parts with PBO.

 

[Ajay]

16 core GB result.

Very impressive at only 4.3GHz.

Yeah, but can it run Crysis?!

 

[Fanatical Meat]

Yeah, but can it run Crysis?!

I can’t believe we’ve made it so far without a Crisis reference.

 

[realibrad]

So then is it left to windows and the application on effectively requkest

Maybe this will clear it up for you.

See that tweet translated to a picture, XFR2's magic:

This is what differentiates the 3700x and the 3800x, one is a 65w part while the other's ceiling is 105w. Same as 2700 (65w) and 2700x (105w). See how beautiful the boost algorithm is? It doesn't matter if you have a dual core or a 16 core part, low threaded workloads will always run at or near the advertised turbo speed if the operating conditions allow it.

Keep it cool, give it a good motherboard with a decent VRM and let it do its thing. Expect it to be even more polished and better at making the most out of the designated TDP bracket it's working in on 3rd gen parts.


1st gen parts and their behavior, in comparison, are from the stone age.

So then it looks like there is a sort of single core max. I get that the chip will boost core frequency of the cores individually, but, there comes a point where the effective result is a single core max boost.

For example, if the workload is a single thread, then only a single core need be boosted. That core will max out in terms of boost. Lets say its 5 If somehow the workload were duplicated perfectly, the two cores doing work could not both boost to 5 as the things listed previously would inhibit the frequencies of the cores. So, I get that there is nothing that explicitly states a max single core frequency, but, the reality is that there is likely a average max single core frequency boost. That said, given the variation between chips, I doubt its consistent and thus becomes much easier to give all core boost numbers.

 

[.vodka]

So then it looks like there is a sort of single core max. I get that the chip will boost core frequency of the cores individually, but, there comes a point where the effective result is a single core max boost.

Yes, that's the idea. It then tapers off depending on many conditions.

https://www.reddit.com/r/Amd/comments/b8a5ft

https://www.reddit.com/r/Amd/comments/bzt50i

Have a look at Robert's explanation and replies in both of these links, he goes in depth what the algorithm does and how it behaves.

---------------------------

I like the new PBO on 3rd gen, it's more flexible and the best way to overclock these chips.

So that's 4.9GHz max boost clock for the 3950x with PBO on, a suitable motherboard, and an adequate cooling solution. A little motherboard trickery as with PBO parameters on second gen parts gets that to 5GHz for lightly threaded workloads.

+200MHz across the board for the max boost clock and a similar boost for the average clock speed, simple as toggling a BIOS switch. Gotta love that, and the fact that now we have control over the max clock. No bclk overclocking needed...

 

[realibrad]

Yes, that's the idea. It then tapers off depending on many conditions.

https://www.reddit.com/r/Amd/comments/b8a5ft

https://www.reddit.com/r/Amd/comments/bzt50i

Have a look at Robert's explanation and replies in both of these links, he goes in depth what the algorithm does and how it behaves.

---------------------------

I like the new PBO on 3rd gen, it's more flexible and the best way to overclock these chips.

So that's 4.9GHz max boost clock for the 3950x with PBO on, a suitable motherboard, and an adequate cooling solution. A little motherboard trickery as with PBO parameters on second gen parts gets that to 5GHz for lightly threaded workloads.

+200MHz across the board for the max boost clock and a similar boost for the average clock speed, simple as toggling a BIOS switch. Gotta love that, and the fact that now we have control over the max clock. No bclk overclocking needed...

I'm actually super excited for this chip. I'm looking at the 3800x or 3900x. I'm looking for a gaming machine, and am under the baseless assumption that a single chiplet will be better for gaming but...

This is going to be a fun system for me.