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

[JPB]

AMD rumored to refresh Matisse CPUs with Ryzen 7 3850X and 3750X

A big juicy rumor has just hit the internet, with it teasing that AMD would be refreshing its current-gen Ryzen 3000 series CPUs with two new entries.



These two new processors are the Ryzen 7 3850X and the Ryzen 7 3750X, which would succeed the Ryzen 7 3800X and Ryzen 7 3700X processors that are currently on the market. The tweet comes from 'HXL' who published the SKUs of the two new processors and made sure to tell the world they are "Not APU".

The tweet itself was a reply his own tweet of a 'Big day coming...' and that the release would be on 6/16 which after that AMD would launch on 7/7. An interesting timeline of release for the Matisse refresh, as 7/7 was the date that the Zen 2-based Ryzen 3000 series launched in 2019.

If you didn't believe just that single tweet, then how about GIGABYTE kinda confirming the Matisse Refresh in their B550 presentation slide, as noted by VideoCardz. The GIGABYTE motherboards roadmap for support shows "Matisse & Refresh",

"Vermeer", and "Renoir" -- all of which aren't released yet.

AMD seems to be responding to the just-launched Intel Core i9-10900K and Core i7-10700K processors, as these new Ryzen 7 3850X and Ryzen 7 3750X processors would directly compete against Intel's new 10th Gen CPU offerings.

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

We first heard wind of AMD's upcoming Matisse Refresh in some leaked news barely 24 hours ago, but now we have some even juicier stuff to talk about: the Ryzen 9 3900XT, Ryzen 7 3800XT, Ryzen 5 3600XT.

That's right, according to our friends at Wccftech, AMD is about to launch the Matisse Refresh with the Ryzen 9 3900XT, Ryzen 7 3800XT, Ryzen 5 3600XT processors. The new Ryzen 3000 series CPUs would adopt the "XT" branding, something that Radeon fans will remember and love to see back in the ring.

The new rumor has AMD replacing the Ryzen 9 3900X, Ryzen 7 3800X and Ryzen 5 3600X with the new Ryzen 9 3900XT, Ryzen 7 3800XT, Ryzen 5 3600XT processors -- but increasing the CPU clocks, having enhanced overclocking support, and dropping the price on the current-gen CPUs that they replace.

Ryzen 9 3900XT - 12C/24T + faster clocks + enhanced OC support
Ryzen 7 3800XT - 8C/16T + faster clocks + enhanced OC support
Ryzen 5 3600XT - 4C/8T + faster clocks + enhanced OC support

These new CPUs would compete directly against Intel's latest Comet Lake processors, in the just-launched Core i9-10900K, Core i7-10700K, and Core i5-10600K.

 

[Rigg]

---------------------------
Looks like these newer 3600 batches are doing some previously impossible clocks at ~1.2v. If 3600 level quality chiplets are hitting that high, then it's completely plausible for 3800x-3900x tier chiplets to do the quoted speeds.

2000MHz fabric also means 1:1 4000MHz memory.

Those nice b-die and rev. E sticks we're all using will now be able to get closer to their actual abilities... and games will love the lower latency.

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

Looks like 7nm got a nice refinement lately... bodes well for Zen3 and 7nm+ or whatever the enhanced version it's built on is called now

The 3600 (7th week 2020) I picked up last Thursday is holding stable @ 4.55/4.5 @1.28v (load SVI2). Passed an hour of P95 SFFT no AVX. I'm an 1:45 into an 8 hour realbench test as we speak. I'll start a thread on it tomorrow with some screenshots. No wonder AMD is planning a refresh. This silicon is amazing. I bought 5 3600's and 2 3600x's in 2019. The one's I had on launch day took 1.4v to get to 4.3. My good 3900x CCD (30th week 2019) is significantly worse than this new 3600 CCD. That chiplet was far and away better than anything I got in an R5 last year. This one is stable approximately 75 mv lower at a given clock vs the 3900x chiplet . Craziness.

It's also doing 1900 FLCK stable.

 

[DrMrLordX]

That's unexpected.

 

[.vodka]

Some more insane samples.

AMD Ryzen 5 3600 @ 4702.27 MHz - CPU-Z VALIDATOR

[b8mgd8] Validated Dump by Abilio (2020-05-22 11:27:06) - MB: ASRock B450M Steel Legend - RAM: 16384 MB

valid.x86.fr

This one is stable enough at 4.7GHz 1.4v all core to do a CPU-z validation and a cinebench run. This is 3950x tier silicon... at the 3600 quality level.

Just imagine that 3600 in a 3950x operating at its PB2 parameters + PBO, maybe this one hits 4.9GHz ~1.5v single core and finally justifies PBO's existence, lol

The 3600 (7th week 2020) I picked up last Thursday is holding stable @ 4.55/4.5 @1.28v (load SVI2). Passed an hour of P95 SFFT no AVX. I'm an 1:45 into an 8 hour realbench test as we speak. I'll start a thread on it tomorrow with some screenshots. No wonder AMD is planning a refresh. This silicon is amazing. I bought 5 3600's and 2 3600x's in 2019. The one's I had on launch day took 1.4v to get to 4.3. My good 3900x CCD (30th week 2019) is significantly worse than this new 3600 CCD. That chiplet was far and away better than anything I got in an R5 last year. This one is stable approximately 75 mv lower at a given clock vs the 3900x chiplet . Craziness.

It's also doing 1900 FLCK stable.

Lovely!

I wonder if 3700x and 3900x's made from this latest batch have the same insane behavior.

 

[DrMrLordX]

Are those the CCDs that will go into the refresh products?

 

[.vodka]

With these results, they're most likely to be going into the refresh products right now.

We'll see if the XT parts do manual overclocks as good as these 3600s or better as proof. I'm more interested in PBO right now as a valuable tool with this CCD quality.

Maybe we'll finally see this infamous video become true.

 

[jpiniero]

With these results, they're most likely to be going into the refresh products right now.

We'll see if the XT parts do manual overclocks as good as these 3600s or better as proof. I'm more interested in PBO right now as a valuable tool with this CCD quality.

I was thinking that the good dies will end up in the XT models, so quality of 3600s will go back down.

 

[Gideon]

Though from twitter leaks it seems FCLK is closer to 1900 than rumored 2000. Which is still nice, but not quite as big a feal as it seemed

 

[VirtualLarry]

This is 3950x tier silicon... at the 3600 quality level.

I thought I had read that someone suggested that AMD took all of their dual-CPU-die packages, that were tested out to 3950X specifications, and re-sold onto the market as 3600 CPUs, hence what we're seeing. Because the top-tier 3950X wasn't selling as well as they had hoped, and their binning and silicon process were just that good! (*)

I was thinking that the good dies will end up in the XT models, so quality of 3600s will go back down.

I was thinking that too.

(*) That happened with the Pentium II. The SL2W8 S-Spec was rumored to be a batch of 450Mhz PII CPUs, that were down-labeled as 300Mhz, and sold onto the market, because Intel made a change to the heat-plate at the last minute, to add more contact between the L2 cache dies and the heat-plate (updated heat-plate had some "dimples" there).

I believe that to be true, I had one of those, and it clocked up to 450Mhz lickety-split, with a BX6 board. Just change FSB from 66Mhz to 100Mhz, bingo!

 

[Shivansps]

I believe that to be true, I had one of those, and it clocked up to 450Mhz lickety-split, with a BX6 board. Just change FSB from 66Mhz to 100Mhz, bingo!

HA! When overcloking was done by manually moving bridges on the motherboard... good times.

 

[JoeRambo]

HA! When overcloking was done by manually moving bridges on the motherboard... good times.

Yeah, but ABIT BX6 was legendary for being able to do those adjustments from BIOS. Those were the days, many were rocking 300A Cellys, with budget 128KB of L2 clocked at 450Mhz or faster were better than off chip full monty P2 with 512kb of L2 that was 1/2 of CPU clock.

 

[VirtualLarry]

Those were the days, many were rocking 300A Cellys, with budget 128KB of L2 clocked at 450Mhz or faster were better than off chip full monty P2 with 512kb of L2 that was 1/2 of CPU clock.

... for games. For actual multi-tasking and productivity work, the 512KB L2 cache of the PII actually was faster than the Celly 300A chips. Sorry to burst your bubble. If I still had that kit, I'd bust it out for some benchmarks.

 

[JoeRambo]

... for games. For actual multi-tasking and productivity work, the 512KB L2 cache of the PII actually was faster than the Celly 300A chips. Sorry to burst your bubble. If I still had that kit, I'd bust it out for some benchmarks.

Sorry to burst your overspend bubble, but Anand back in the day found this:

https://images.anandtech.com/old/cpu/celerona/index41.jpg

It was quite a performer, that little Celly, it might not had as much longevity as P2 450, but both were soon obsoleted by P3 and Athlon clocking near Ghz.

 

[Rigg]

Lovely!

I wonder if 3700x and 3900x's made from this latest batch have the same insane behavior.

I would assume so. I can't think of any logical reason why they wouldn't be putting these into everything.

I was thinking that the good dies will end up in the XT models, so quality of 3600s will go back down.

The fact that dies of this quality are making their way down to budget tier 3600's mean's they must have a plethora of silicon that's as good or better than this.Clearly they want to seize the opportunity to sell these with better stock settings than launch SKU's at a premium. They don't want to ditch their current best selling SKU's so they'll just keep making them with better silicon. It makes them attractive for manual overclock (and people/OEM's who need the lower TDP stock settings) while everyone else can pay extra for new SKU's with better stock clocks. I'm of the opinion this is probably a pretty standard bin as the process has matured. Time will tell.

I thought I had read that someone suggested that AMD took all of their dual-CPU-die packages, that were tested out to 3950X specifications, and re-sold onto the market as 3600 CPUs, hence what we're seeing. Because the top-tier 3950X wasn't selling as well as they had hoped, and their binning and silicon process were just that good! (*)

The first part doesn't makes sense to me. If this is really top grade silicon reserved fo 3950x why would they dump it on low cost 3600's and not spread it around the 8 core and 12 core parts with higher margin. This must be near the bottom of barrel bins they can churn out right now. I don't see any other reason it would be showing up in low-mid tier SKU's.

 

[Topweasel]

The first part doesn't makes sense to me. If this is really top grade silicon reserved fo 3950x why would they dump it on low cost 3600's and not spread it around the 8 core and 12 core parts with higher margin. This must be near the bottom of barrel bins they can churn out right now. I don't see any other reason it would be showing up in low-mid tier SKU's.

If you are binning a bunch of dies with a certain rating for a certain application, not just the 3950x but maybe Threadripper as well and you have the option of just installing them on a low clocked part or boosting the value of the lower cost parts by offering them as higher clocked versions, option 2 wins.

Now keep in mind the 3600 is the best selling chip, a X is going to cost more, a XT more than that. What this does is probably cannibalize 3600x sales for higher margin XT sales. So its not like we are going to see 3600 sales with the XT. Higher than the 3800xt and 3900xt, but still.

So this means that yes AMD either has a stockpile of higher clocked chips or process improvements they can offer better versions. Or my guess is a little of both. But down offering chips to fill volume requirements when the higher product isn't selling has always happened. There is a reluctance to lower "value" of higher end chips by lowering costs to up demand. Intel won't do it all. AMD does, but again if the stock is high enough, prices would have to drop dramatically and something like the 3950x is worth more as a Halo product with its high price to sell more 3900x's and 3700/3800s. For all we know they would need to lower the 3950 to $500 to keep up with stock. Better to keep margins high on that, and raise margins on 3600, 3800, and rebound the 3900 margins, by using excess stock.

 

[Kenmitch]

The first part doesn't makes sense to me. If this is really top grade silicon reserved fo 3950x why would they dump it on low cost 3600's and not spread it around the 8 core and 12 core parts with higher margin. This must be near the bottom of barrel bins they can churn out right now. I don't see any other reason it would be showing up in low-mid tier SKU's.

If you can't sell the overpriced offerings then you don't have a choice. You sell it where you can or you sit on it and stare at it. /end sarcasm

I'd imagine the new improved silicon is going where it's needed currently. The current stock along the anticipated sales would play a huge role in what's currently being produced in the pipeline.

 

[Indus]

Still have an R5 1600 w. Ddr2400.

go for 3600 or xt?

 

[VirtualLarry]

I gambled and ordered another "Fresh" Ryzen R5 3600, for $167. My reasoning is thus: If current silicon and/or binning, is allowing people to buy "current model" 3600 CPUs, and get 4.4-4.5Ghz all-core OCs on them, at reasonable volts, and it's likely that the "XT" models will attempt to "reset" the pricing schedule, thus more than likely $200 or possibly more due to demand for the 3600XT, and given that the explicit single-core boost is higher on the 3600XT, but my workloads are mostly all-core, I chose the "current model" 3600.

When I get it, it remains to be seen how it will OC, and what the temps under all-core (DC) load will be (on my 240mm AIO LC kit).

Basically, I was figuring that, if ALL of the silicon currently being produced, is "better", and current pricing for the 3600 is cheaper, then might as well buy now. Of course, possibly, if I had waited, it's possible that most people would go for the XT, for the explicitly higher single-core boost clock (better for gaming, with a "master thread" at highest clocks), and thus, prices on the "vanilla" 3600 might drop further. So maybe I jumped the gun. Likewise, though, possibly with more binning for the 3600XT, the "vanilla" 3600 CPUs might actually clock worse, but I doubt that they would clock even as bad as the initial-release 3600 CPUs. At least, in all-core OCs.

I guess life is a trade-off. I made my (relatively easy) choice, I figured, heck, I was waiting for the 3300X @ MSRP ($130), and the (current) price of the 3600 (with two more REAL cores) was only $167, might as well spring for the (current) 3600 NOW.

Edit: My (really early, initial-release) Ryzen R5 3600 CPU, won't even give me a stable all-core OC of 4.00Ghz, even with or without pumping the voltage, with a DC workload that taxes the CPU 100% including AVX2 opcodes.

 

[Mk pt]

Still have an R5 1600 w. Ddr2400.

go for 3600 or xt?

Wait to see 3600XT performance and price.
Then choose what fits better in performance/price.

3600 will be cheaper, so if you wait at least you'll save money.

And buy ddr4 3600 kit.
There are cheap kits (3600cl16) and worth the price.
Ddr4 2400 will be a big bottleneck for any Ryzen 3000.

 

[VirtualLarry]

Still have an R5 1600 w. Ddr2400.

go for 3600 or xt?

So to answer your question:

If your goal is primarily gaming, then I would wait for the XT models to surface, just because they explicitly have a higher single-core boost-clock. (I think 4.7 was reported?)

If you are budget-minding, and mostly do an all-core OC, I don't feel that if the silicon is similarly better between a "current model" 3600 and an upcoming 3600XT (binning notwithstanding), that the all-core manual OCs will be all that much different between the two. In that case, save a few bucks, IMHO, and get a "current model" 3600 CPU. At least, that's what I did. It's a gamble, and basically we don't know if AMD is planning on "binning out" the "good" 3600 CPUs into XTs, but I imagine that they have enough "good" silicon to go around, and that both of those SKUs will have decent all-core OCs. Of course, there may be still some (relative) "duds" out there, but again, I doubt that they will even be as bad as initial-release silicon was.