• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD Readies 3rd Gen Ryzen "Matisse Refresh" Ryzen 7 3850X and 3750X Processors

Fixed the word "be" with "need".
I agree with you that these are absolutely pointless and shooting in the feet for AMD.
Yeah thanks, typing on the phone is a PITA.
 
3 months before the release of Ryzen 4000, it is absolutely pointless to bring salvaged parts.
Better bring Renoir with higher performance.
Well, for some reason AMD thinks differently. For me that is the bedrock to which I'm trying to dig to, finding out why AMD is releasing it even though Renoir is just around the corner. Saying "pointless" gives me no answers really.
 
Do you think this will include process optimisations (so 7nm+) or is it just going to be slightly different configuration/clocks of existing Zen2?

I mean, adding a plus to each process seems like the status quo now - I'll presume TSMC have been working on improvements to their 7nm node for a while now....
 
This could be weapons-grade enriched bovine defecation that puts out over 9000 Rontgen, but what the hell.

A few weeks ago, a fairly reliable source sent me a "slide" he sketched up on mspaint looking at another slide. It talked about 8-core dual-CCD chips (two 4+0 CCX chiplets). The idea being reduced thermal density and double the I/O bandwidth per core to the cIOD vs current single-CCD packages; and freeing up clock-speed headroom per CCD to drive boost clocks around the 5 GHz mark. Something excruciatingly hard for Zen 2 to begin with.

I have serious doubts about this. For one, such an approach could affect performance of games that scale beyond 8 threads. It will also drive up costs for AMD, unless it's sitting on a pile of 4+0 CCDs capable of reaching 5 GHz.

Now the theory of 3850X being a 4+4 single CCD chip that's clocked to hell with increased power limits, seems more credible.

I'll put my money on single CCD 4+4.
I had mentioned in a Threadripper post on the 3980X that it would be intriguing and interesting if AMD were retrofit a SKU with a 50/50 chiplet split APU. The more I think about AMD's entire chiplet approach I can't help, but think it offers them a very immense amount of overall flexibility across the spectrum of things they could do. Just as one example beyond a Threadripper split APU they could take 2 of the chiplets for CPU CCX's and the other 6 and use them for GPU CCX's. Now that's for Threadripper taking things a step further they could similar with Epyc and have even more GPU CCX's for a dual socket configuration with more memory channels for bandwidth that could in turn feed the APU's. Now think about that from the power efficiency aspect. Depending entirely on how AMD went about it they could cram a ton of APU horsepower into a dual socket Epyc server while still offering Threadripper or AM4 CPU core counts. It all depends how much GPU chiplet horsepower they want to combine together. They could go more CPU heavy more GPU heavy or balanced or semi balanced it's entirely flexible.

I wouldn't rule out the possibility of twin 4+0 chiplets even if it seems a bit unlikely at this stage. If anything given what I just said I believe it's likely we'll see more SKU variety from AMD in the future because they've got the flexibility to leverage it much like Nvidia for years on the GPU side has had a array of slightly differing segmented SKU's AMD could mimic that same approach rather trivially with it's chiplets to it's advantage. They've got a clear design win they've come up with right now that Intel doesn't have a immediate answer to quite yet though something is inevitably in the works.

I still think another very interesting and novel thing AMD could do is discrete PCIe add in cards that are APU based or even CPU based. I think with the APU route it would be make sense is if they did the affinity fabric approach like they are doing on the new Radeon Pro cards. What would be rather neat is if AMD bought out Romex Software and integrated it at the hardware and software level as discrete PCIe add in cards that matted a AMD chiplet, M.2, and LPDDR slot together and did read caching of the M.2 with it. Now best yet all of it could be done with no CPU or memory overhead to your main CPU and system memory and they could even have it bridge to Radeon GPU cards over infinity fabric. That would be a great acquisition play on AMD's part. That could all easily fit in a slim single slot PCIe card slot.
 
I had mentioned in a Threadripper post on the 3980X that it would be intriguing and interesting if AMD were retrofit a SKU with a 50/50 chiplet split APU. The more I think about AMD's entire chiplet approach I can't help, but think it offers them a very immense amount of overall flexibility across the spectrum of things they could do. Just as one example beyond a Threadripper split APU they could take 2 of the chiplets for CPU CCX's and the other 6 and use them for GPU CCX's. Now that's for Threadripper taking things a step further they could similar with Epyc and have even more GPU CCX's for a dual socket configuration with more memory channels for bandwidth that could in turn feed the APU's. Now think about that from the power efficiency aspect. Depending entirely on how AMD went about it they could cram a ton of APU horsepower into a dual socket Epyc server while still offering Threadripper or AM4 CPU core counts. It all depends how much GPU chiplet horsepower they want to combine together. They could go more CPU heavy more GPU heavy or balanced or semi balanced it's entirely flexible.

I wouldn't rule out the possibility of twin 4+0 chiplets even if it seems a bit unlikely at this stage. If anything given what I just said I believe it's likely we'll see more SKU variety from AMD in the future because they've got the flexibility to leverage it much like Nvidia for years on the GPU side has had a array of slightly differing segmented SKU's AMD could mimic that same approach rather trivially with it's chiplets to it's advantage. They've got a clear design win they've come up with right now that Intel doesn't have a immediate answer to quite yet though something is inevitably in the works.

I still think another very interesting and novel thing AMD could do is discrete PCIe add in cards that are APU based or even CPU based. I think with the APU route it would be make sense is if they did the affinity fabric approach like they are doing on the new Radeon Pro cards. What would be rather neat is if AMD bought out Romex Software and integrated it at the hardware and software level as discrete PCIe add in cards that matted a AMD chiplet, M.2, and LPDDR slot together and did read caching of the M.2 with it. Now best yet all of it could be done with no CPU or memory overhead to your main CPU and system memory and they could even have it bridge to Radeon GPU cards over infinity fabric. That would be a great acquisition play on AMD's part. That could all easily fit in a slim single slot PCIe card slot.

On a loosely related note, with EPYC F SKUs, AMD was gunning for frequency by spreading fewer cores across more CCDs. They didn't have gaming in mind with the resulting product (something that spreads time-critical workload across multiple cores). For games, you need a large number of (4~8) cores clustered into as closely-knit space as possible. 4+4 is as close as it gets with Zen so far. Which is why I think Zen 3 chiplet (which does away with CCX and puts all cores of the CCD to share common resources), would make huge gains (+10-19%) in gaming performance, even with lower clocks (boosting around 4.50 GHz). <----this is why I believe Intel is desperate to restore IPC leadership at any cost, even if it meant backporting future uArch to an older fab process.
 
Since (I assume) these parts are going to have the same core/thread counts as the non-50 current models, releasing them now doesn't really make much sense for many reasons (as many others have pointed out). 3100 and 3300X do, as they are filling a pretty big hole in the Ryzen lineup, in particular at the low end where there's a lot of volume.

Thus, speculation time: does the release of these Zen 2 parts, so close to the supposed Zen 3 launch date - imply that Zen 3 is going to be delayed?

Sorry, don't know, I only knew it was coming.

No prob, I assume they are both 8c/16t as their predecessors, just wanted to confirm.

huge gains (+10-19%) in gaming performance

That's a very specific upper bound, why not just go with 20%?
 
I dont get it... what will these bring to the table? 100mhz more being binned?

I'll pass unless these are all 4.5ghz+ cpus.
 
I dont get it... what will these bring to the table? 100mhz more being binned?

Even if they're the same core config and only a 100MHz bump, it gives AMD the chance to re-price their offerings with a brand-new SKU.

100MHz bump and a corresponding price reduction won't change too much given that the market prices for 3700X and 3800X are discounted almost everywhere but for reviewers and streamers who often talk about MSRPs that has a fairly big impact on the perceived value for money.

Almost all of these existing product re-brand excercises from Intel, AMD, Nvidia are more about marketing and market-posturing then they are about the actual substance of the product. Even if the product is underwhelming it's good for the consumer as the "old" SKUs get put on clearance deals to get rid of inventory and the prices/performance metrics usually move in the consumers' favour of "more performance, same price" or "same performance, lower price". It's a win-win for everyone except the competing manufacturer, usually.
 
Last edited:
Since (I assume) these parts are going to have the same core/thread counts as the non-50 current models, releasing them now doesn't really make much sense for many reasons (as many others have pointed out). 3100 and 3300X do, as they are filling a pretty big hole in the Ryzen lineup, in particular at the low end where there's a lot of volume.

Thus, speculation time: does the release of these Zen 2 parts, so close to the supposed Zen 3 launch date - imply that Zen 3 is going to be delayed?



No prob, I assume they are both 8c/16t as their predecessors, just wanted to confirm.



That's a very specific upper bound, why not just go with 20%?
The way I see it it's not going to be easy for any tech company to hit it's release dates at the moment, so many obstacles.
I see them holding onto limited zen3 production for the server space while using the maturity gained on zen2 to provide a meaningful challenge to Intel's releases.
Sure I wouldn't upgrade nor would most.
But the same applies to 9900k typical owner's.
However it could be a choice for older system owner's.

Plus Intel are clearly sandbagging rocket lake , trying to get some competitive bins built up.

So AMD needs an answer to comet lake yet also needs to reserve an answer for rocket lake.
Hopefully I got my lakes right but you know what I'm saying.

Delays yes hopefully not long one's.
 
Since (I assume) these parts are going to have the same core/thread counts as the non-50 current models, releasing them now doesn't really make much sense for many reasons (as many others have pointed out). 3100 and 3300X do, as they are filling a pretty big hole in the Ryzen lineup, in particular at the low end where there's a lot of volume.
I can understand releasing a higher binned chip and toying around with ccx config, but what I don't understand is, why release both the 3750 and 3850. Non-50 versions are basically the same chip. One refreshed SKU should cover all the needs, no?
 
Unnecessary, I do think this is another frivolous stunts from AMD to keep gaming crowd pleased.
We all already knew what these xx50 will bring, and overclocked version of xx00, need more juice, excessive heat all for the sake of 5 fps.
 
I can understand releasing a higher binned chip and toying around with ccx config, but what I don't understand is, why release both the 3750 and 3850. Non-50 versions are basically the same chip. One refreshed SKU should cover all the needs, no?

Maybe it's 3850 and 3920 (ie increasing clocks for different core counts).

Anyway, I don't see why all the complaints, if AMD gives you 100-200MHz more for the same price because the process has matured, what's not to like? If you don't need a processor right now, don't buy, but if you need one, having more choice can only be better.
 
3 months before the release of Ryzen 4000, it is absolutely pointless to bring salvaged parts.
Better bring Renoir with higher performance.
For you, not for a company that makes them and probably has piles of them sitting somewhere.
It makes perfect sense from economical standpoint as long as there is demand.
 
Maybe it's 3850 and 3920 (ie increasing clocks for different core counts).

Anyway, I don't see why all the complaints, if AMD gives you 100-200MHz more for the same price because the process has matured, what's not to like? If you don't need a processor right now, don't buy, but if you need one, having more choice can only be better.


They are moving shortly to a new N7 process and a new architecture with Zen 3. If it was back in January, I would somehow understand, but this so late it is unnecessary.

Don't buy but wait for the new generation!
 
4+0 + 4+0 is gonna be slower than 4+4 + 0+0 bc performance depends on latency. 3300x is faster than 3100 not bc of some magic scheme but bc all the cores are much closer to each other electrically.
 
They are moving shortly to a new N7 process and a new architecture with Zen 3. If it was back in January, I would somehow understand, but this so late it is unnecessary.

Don't buy but wait for the new generation!
You know right at this moment there are lots of buyers grabbing those Ryzen 1600AF's as fast as they can, right? Horses for courses, not everybody is interested in the same segment of the market, but large companies like AMD are trying to cater to all of them.

For me personally it's not enough of an upgrade, but for somebody with a b350 r1300 combo, maybe a 3850x would sound like the good option, no need to dump the mobo, and sufficient performance for the next generation of games.
 
It will be all the same except speed increased a bit. 3750x would push 4.2 all cores and 3850x would push 4.3 all cores without sweat.so just fine tuned/ cherry picked.
 
I'm just waiting for Renoir to launch so 3000 series drop in price just like the 2000 series did for a bit. I won't be able to run the 4000 parts without a new mobo anyways..
 
Wouldn't it be crazy if these are the chips in the Xbox1 and PS5?
 
You know right at this moment there are lots of buyers grabbing those Ryzen 1600AF's as fast as they can, right? Horses for courses, not everybody is interested in the same segment of the market, but large companies like AMD are trying to cater to all of them.

For me personally it's not enough of an upgrade, but for somebody with a b350 r1300 combo, maybe a 3850x would sound like the good option, no need to dump the mobo, and sufficient performance for the next generation of games.

This. Sometimes people just like to assume what a company is doing is for them without thinking about the other people out there like you mentioned who are looking for the upgrade from say a Ryzen 1300 or even the first Ryzen 1600 and have b350 or x370. If its not for you then its not for you but don't assume its pointless because it doesn't work for you. What works for people who thinks is pointless will come this fall in Zen 3. :)

Honestly it could even work for me depending on what I feel like doing. I upgraded from my Core i7 2700k and HD7950 a week ago with Ryzen 1600AF and RX 5500xt and Asrock b450m Pro because they were so cheap and something was giving issues on old setup. Now I can either get one of these July chips or Just go balls deep with Zen 3. I have a good amount of choices.
 
Can we all agree competition is good. Even if some of us see this pointless.
 
My eyes fixed on 3080Ti at this point, at 4K gaming , CPU in the backseat from here on out for me.
 
It has some possibility (though that low) that amd refresh the current line 3XXX to 6nm. I just remembered this:

"TSMC in its quarterly earnings call expressed confidence in that most of its 7 nm (N7) process production node customers would be looking to make the transition to their 6 nm (N6) process. In fact, the company expects that node to become the biggest target for volume ordering (and thus production) amongst its customers, since the new N6 fabrication technology will bring about a sort of "backwards compatibility" with design tools and semiconductor designs that manufacturers have already invested in for its N7 node, thus allowing for cost savings for its clients. " https://www.techpowerup.com/255097/tsmc-expects-most-7nm-customers-to-move-to-6nm-density
 
Honestly it could even work for me depending on what I feel like doing. I upgraded from my Core i7 2700k and HD7950 a week ago with Ryzen 1600AF and RX 5500xt and Asrock b450m Pro because they were so cheap and something was giving issues on old setup. Now I can either get one of these July chips or Just go balls deep with Zen 3. I have a good amount of choices.

Nah, I'd say you're pretty well set for now, if you don't go higher than 1440p, better to wait at least for the next gen, in both CPU and GPU.

But otherwise, yeah, choice is great, as long as the price-performance ratio is improving, we should all be happy, even if we're not buying right away. It just means the market is going in the right direction for when we will.

And there's another reason AMD might need this refresh, the new 10600K from Intel seems like a good buy for gamers, so they need to respond. competition is great for users.
 
Back
Top