AMD preps Ryzen 9 3900XT, Ryzen 7 3800XT, Ryzen 5 3600XT for June 16

[JediFragger]

5 hours ago, Flavio hc 16 said:

yeap, and IMHO all of this is linked to the fact that ddr5 will arrive in 2022 instead of 2021, amd want to  squeeze only 1 architecture instead of 2 before a radical change in platform + 5 nm process, they can say that this year will be without major cpu launches as the coronavirus pandemic "delayed them" 

And they'll certainly want to beat Intel to the DDR5 punch, so makes sense.

[nick name]

I was wondering if AMD would release a 3900X and 3950X with two "good" CCDs instead of the current 1 good and 1 bad.  I assumed better yields would allow for this eventually as the process matured more.  

 

What I'm assuming will happen is better all-core performance and not so much better single-core boost speeds past 4.6GHz.  Unless now that +200MHz bit will actually add more speed than what I see.  Which is an actual +75MHz.  The reason I assume this is because I'm assuming all-core speeds are decided by the capabilities of the weaker of the two CCDs.  So if the weaker CCD can only do ~ 4.15GHz at around 1.3V (with Auto voltages) then all the cores have to do that as they boost to the same speed on both CCDs.  Which is why per CCD overclocking can be so useful for all-core workloads when you overclock.

 

So as it sits now you can do something like 4.4GHz on your good CCD and then 4.25GHz on your bad CCD at 1.26V . . . but if AMD is going to put two "good" CCDs into a CPU then you can just do an all-core (both CCDs) overclock of 4.4GHz at 1.26V.  That would be lovely.  

 

And the 3600XT would simply have a better CCD in it as a result of a matured process and better yields.

 

Essentially no longer having to bin the best CCDs into Epyc CPUs or other skus higher in the product stack.  

 

So while I don't expect much or really any improvement in single-core boost speeds (except for the 3600XT) it does seem like we could see better all-core speeds on all the XT skus.  

[GDRRiley]

this seems like this would only add to the struggle to fit CPUs in bios chips

[.Apex.]

I don't think there will be much difference between the chips but rather a simple higher PBO overclock out of the box and a higher TDP, I doubt they made architecture improvements when Zen 3 is close to being launched, they're only releasing this to get a newer release date title on their chips as most consumers don't understand what any of this means and might look the other way if they saw that Zen 2 was released in 2019 and the Intel chips are new "so that must mean they're better", basically it's a move to shit all over Intel.

 

As for if they expect Zen 3 would be delayed, I'm not sure, maybe they don't wanna take chances as the pandemic is unpredictable and if something happened they're afraid to lose market share because the Intel chips are newer? I think it's a bad decision as it makes all that BIOS fiasco even worse and it's far from the first time AMD has made bad decisions.

[Letgomyleghoe]

I want 12 cores at 5ghz god damnit!

[Dravinian]

We will have to wait and see, but I doubt this is true.

 

Why would they bother, when they are on the verge of releasing Ryzen 4000 anyway?

 

What is the point of the refresh? What would they achieve? Having a CPU that is better than Intel? They already have that.  The difference, even in gaming terms, is utterly pointless and in all other tasks they dominate and are better priced.

 

I would be surprised if AMD bothered to spend the money on doing this, when they simply do not have to do anything, just release Ryzen 4000 series and complete dominate the market place.

[justpoet]

Based on all the BIOS stuff recently, I don't buy it.  If they'd stuck with 500 series boards only for 4000 series, then they might've been able to sneak this into existing, or even say it is a zen2+ type part and also needs 500 series board.  That didn't happen though, so…I highly doubt this is real.

 

If it is though, I agree with the likelyhood of it being mostly binning for both chiplets for better all core boost as well as minor microcode updates, allowing them to push off zen3 a little longer until intel announces whatever comes next, in theory around the start of next year.  It could also allow something like the 3900xt or 3800xt to even more favorably compare to the intel offerings, driving even more of a stake into intel's current generation.

[justpoet]

Actually, come to think of it.  If they just make both chiplets better binned, they may not need to do anything in bios changes to get better performance, thanks to how their boosts tend to work.  They could also call it a 3900X in bios and then allow something like Ryzen Master or the linux kernel updates to adjust how it runs based on a software SN range check or similar.

[Soppro]

It's possible that AMD may be following Intel's naming scheme and putting a 'T' to indicate perhaps a more efficient and lower power variant

[Tellos]

I do not think their doing this, if they do it will have to happen in less than 3 months time as in I would expect an announcement within a week no more. Taking wafers from zen 3 makes no sense. 

[CarlBar]

5 hours ago, Tellos said:

I do not think their doing this, if they do it will have to happen in less than 3 months time as in I would expect an announcement within a week no more. Taking wafers from zen 3 makes no sense. 

 

Zen 3 is 7nm+. Different production line AFAIK.

[derr12]

With reports of all core clocks hitting 4.5ghz on the latest batches of current gen, I’m guessing that the process has matured enough to get some higher clocked variants out there. 

[F.E.A.R.]

 

[GDRRiley]

Well it looks like if this is happening it just because yields and clocks have gotten better over time. toss a few hundred on the base and a tad on the boost. Match intel 

[Tellos]

15 hours ago, CarlBar said:

 

Zen 3 is 7nm+. Different production line AFAIK.

Yes but wafers are not themselves 7nm+ or anything their silcon wafers. Theres a limit on those.

[geokilla]

Some are speculating that while Zen 3 will launch, it will only be for the server side so desktops won't actually get Zen 3 will 2021. I think it makes sense to refresh Zen 2 instead. Zen 2 is already killing Intel so AMD has no reason to launch Zen 3 right now. They might as well keep refining Zen 2 like Intel and the 14nm++++ for another year and focus on what really brings in the most profits, servers and data centres.

Also with Zen, we had Zen+ and then Zen 2. I always thought it was weird AMD didn't announce a Zen 2+ and is instead going straight to Zen 3. If there is a Zen 2 refresh coming, it would explain why AMD and the board partners originally didn't want to support Zen 3 on the B450 chipset.

Edited May 25, 2020 by TheRandomness
Merged cleaning

[straight_stewie]

So AMD reusing old designs is good?
But Intel reusing old designs is bad?

[5x5]

2 minutes ago, straight_stewie said:

So AMD reusing old designs is good?
But Intel reusing old designs is bad?

AMD's design is from mid 2019
Intel's design is from early 2015

Refreshing once is fine. Refreshing every year for the better part of a decade is silliness.

In any case, Zen 3 is still coming. Feel free to mark my words. This is a stop-gap response to the 10700K

[Fasauceome]

SO MANY DIFFERENT RUMORS

 

3900 XT, 3800 XT, 3850X, these are all over the map.

 

I personally believe that no matisse refresh exists and/or is planned, and only AMD marketing will prove otherwise.

[Jumballi]

1 minute ago, straight_stewie said:

So AMD reusing old designs is good?
But Intel reusing old designs is bad?

amd is using old designs within the same generation, these are the same chips but more efficient since the manufacturing is better.
intel is using the same design from from 2015, and even then there were some changes to make them more modern.

It's not a good idea to tell amd to take a break and keep releasing zen 2, especially since intel is still king of the market by overall demand.

[Bombastinator]

58 minutes ago, geokilla said:

https://www.tweaktown.com/news/72709/amd-rumored-to-refresh-matisse-cpus-with-ryzen-7-3850x-and-3750x/index.html?utm_source=ttnewsletter&utm_medium=ttemail&utm_campaign=ttcs?utm_source=ttnewsletter&utm_medium=ttemail&utm_campaign=ttcs

 

https://www.chiphell.com/forum.php?mod=viewthread&tid=2222285&extra=page%3D1&mobile=2

 

 

Some are speculating that while Zen 3 will launch, it will only be for the server side so desktops won't actually get Zen 3 will 2021. I think it makes sense to refresh Zen 2 instead. Zen 2 is already killing Intel so AMD has no reason to launch Zen 3 right now. They might as well keep refining Zen 2 like Intel and the 14nm++++ for another year and focus on what really brings in the most profits, servers and data centres.

Also with Zen, we had Zen+ and then Zen 2. I always thought it was weird AMD didn't announce a Zen 2+ and is instead going straight to Zen 3. If there is a Zen 2 refresh coming, it would explain why AMD and the board partners originally didn't want to support Zen 3 on the B450 chipset.

I suspect what it is going to come down to is fab.  AMD has contracts with TSMC.  At a certain date TSMC is going to make a number chips on a certain node in a specific place, and AMD is going to transport and sell them.  When that happens zen3 will arrive for consumers.  If that can’t happen for a year something else will happen instead.  If that is scheduled to happen in 2 months it will happen.

[straight_stewie]

6 minutes ago, 5x5 said:

AMD's design is from mid 2019
Intel's design is from early 2015

At how many years of refreshing do you draw the line? When exactly does it stop being an acceptable thing to do?

 

 

6 minutes ago, 5x5 said:

In any case, Zen 3 is still coming.

Oh I don't doubt that at all.

My point is that, as enthusiasts and machine builders, we shouldn't care at all about how many times a company has improved an existing design vs. making a new one. Nor should we care why they did it. Those things are of no importance what-so-ever to someone who does not have a financial stake in the companies under question.

We should only care about performance or performance/cost. I'm not even saying that Intel is better at either of those, just that we should stop making arguments about architecture refreshments, because they have absolutely no bearing on anything what-so-ever to those who do not have a financial stake in the company under question.

[Void Master]

3 minutes ago, Fasauceome said:

SO MANY DIFFERENT RUMORS

 

3900 XT, 3800 XT, 3850X, these are all over the map.

 

I personally believe that no matisse refresh exists and/or is planned, and only AMD marketing will prove otherwise.

Maybe that's why they changed their mind about 4th gen ryzen cpu support on x470/b450.....

If they refresh zen 2, then they found a way out.... 

hmmmmmm

who knows ? 

we will wait and see.

[5x5]

3 minutes ago, straight_stewie said:

At how many years of refreshing do you draw the line? When exactly does it stop being an acceptable thing to do?

 

 

Oh I don't doubt that at all.

My point is that, as enthusiasts and machine builders, we shouldn't care at all about how many times a company has improved an existing design vs. making a new one. Nor should we care why they did it. Those things are of no importance what-so-ever to someone who does not have a financial stake in the companies under question.

We should only care about performance or performance/cost. I'm not even saying that Intel is better at either of those, just that we should stop making arguments about architecture refreshments, because they have absolutely no bearing on anything what-so-ever to those who do not have a financial stake in the company under question.

1 refresh per generation. The way Intel did it till 2015 and the way AMD is doing it now. Zen, Zen+, Zen 2, Zen 2+, Zen 3, Zen 3+ (in 2021) and Zen 4 in 2022

[Jumballi]

1 minute ago, Void Master said:

Maybe that's why they changed their mind about 4th gen ryzen cpu support on x470/b450.....

If they refresh zen 2, then they found a way out.... 

hmmmmmm

who knows ? 

we will wait and see.

if these are ryzen 3000, then they should work without any bios updates.