(16core added)AMD 3000 specs! 4.7 GHZ, R9 3950x, R7 3700x, 3800x.

[Stefan Payne]

2 minutes ago, leadeater said:

One of the videos I watched mentioned that there are MSI motherboards that won't support Zen2, think that list is rather small though. Don't remember who's video it was.

all A320 probably won't support it, for various reasons...

But I can understand it and won't ever demand awesome support from a 50€ Board. There's a reason why it's 50€. 

 

In general, I'd expect Support on higher end Boards while I wouldn't have a Problem with no Support on lower End Boards.

Though I haven't found what I meant from MSI but an ASROCK AB350M-HDV R3.0 is one of those things (and I wouldn't want anything more than 65W on those Boards either)

There are similar Boards from other Manufacturers as well.

But its also more than a month until then, so much can still change.

 

Even if a manufacturer doesn't support a processor, that doesn't mean they don't allow it!

Meaning that it still might work, you just don't get any (official) Support from the manufacturer.

[GhostTownParty]

So i'm running actually a FX8320 with a GTX 1060 6GB, wich Ryzen should i get for that?

[The Benjamins]

3 minutes ago, Screaming Orca said:

So i'm running actually a FX8320 with a GTX 1060 6GB, wich Ryzen should i get for that?

What do you do, and how much do you want to spend?

 

honestly the 3600/3600x is a great entry, if you are ok with getting up there in price the 3700x/3800x will be amazing, and if you do a lot of CPU heavy work or want to go balls to the wall the 3900x will last you forever.

 

You are going to need new ram and motherboard so there is a lot to think about for a switch for you.

[BiG StroOnZ]

Some interesting "Zen 2" information (AMD Ryzen 3000-series CPUs Overclocking Numbers, DDR4 speeds, and 5GHz Anniversary chips):

Quote

Spoiler

 

• 4.8GHz is achievable on all cores
• ~4.4GHz performs similar to a 5GHz 9900k – in Cinebench
• 5.0GHz is doable, but it’s a challenge
• Overclock for overclocking, Ryzen 3000 is still faster
• 5GHz boost isn’t infeasible
• 5Ghz all core is pretty much a no-go.
• 1.35V for all core 4.5Ghz
• Memory is being run very loose and slow to assure the stability for testing

It is also said that the chip is faster than its competitor, the Core i9-9900K when overclocked to 4.4 GHz in Cinebench benchmark and that 5 GHz boost frequencies (single-core) are quite feasible and silicon lottery variants will be able to hit that pretty much with ease. At the same time, a 4.5 GHz OC can be achieved on the part with just 1.35V which is something that the community should be very excited for and that’s the all core frequency that the leaker is talking about.

 

In addition to the CPU overclock, we also have a first look at the overclocking capabilities of AMD Ryzen 3900X 12 core processor in AIDA64. The CPU can be seen running at a core frequency of 4.6 GHz across all cores while the memory speed is configured at 4000 MHz (CL18 timing). You could see some beefy read/write/copy numbers being pumped out by the chip but the main thing to focus here is the latency which is simply amazing.

 

 

Spoiler

 

Quote

In addition to the overclocking details, it looks like BitsAndChips has details that AMD might offer a 5 GHz SKU in the Ryzen 3000 lineup in 2020. The 5 GHz variant will launch as a special anniversary edition to mark the 20th Anniversary of their Athlon CPU which was the first chip to achieve 1 GHz. The core count for this chip is not mentioned but it seems very possible if AMD uses heavily binned dies since this is a special variant and not a mass-produced consumer variant.

 

 

[Fnige]

1 hour ago, BiG StroOnZ said:

Some interesting "Zen 2" information (AMD Ryzen 3000-series CPUs Overclocking Numbers, DDR4 speeds, and 5GHz Anniversary chips):

 

*Snippity*

Yeah... I'd have a hard time believing this

[BiG StroOnZ]

On 5/30/2019 at 9:29 PM, SlimyPython said:

Yeah... I'd have a hard time believing this

 

What exactly do you have a hard time believing? As there is plenty of information in there that is definitely plausible. To be fair, most of it seems pretty feasible. I mean, we can go over some of it:

 

1.) "4.8GHz all-core overclock." We know there is a SKU in the lineup already that can do 4.6GHz out of the box. A 4% overclock is not really all that much (going from 4.6 to 4.8GHz).

 

2.) "4.4GHz Zen 2 = Intel Core i9-9900k @ 5GHz in CB." There were many people in this thread that were pretty certain, that if what AMD exclaimed in their slides and presentation were correct. That essentially, AMD would have surpassed Intel in IPC. Which would in turn mean they would be able to match Intel at lower clocks. Now even if this is incorrect for the 8c16t part, since the "leak" doesn't explain specifically what part this is that beats the 9900k @ 5GHz, it could simply be a case of it being the 12c24t part that is pulling this off.  

 

3.) "5GHz is doable, but a challenge." Okay, so they admit 5GHz is probably not likely but maybe Golden Samples can. I would agree.

 

4.) "OC for OC, Ryzen is faster." (See 2).

 

5.) "5GHz boost isn’t infeasible." (See 3)

 

6.) "5GHz all core is pretty much a no-go." (Something we already established from 3 & 5; perhaps a couple of cores boosting to 5GHz though).

 

7.) "1.35V for all core 4.5GHz." Don't see what's so crazy about this one. They already have chips that do 4.4-4.6GHz boosts. Additionally, 1.35-1.4v was average voltage for all-core overclocks in previous iterations of Ryzen. 

 

8.) "AMD Ryzen 3900X 12-core processor overclocked to 4.6GHz across all cores with memory @ 4000MHz (CL18 timings)." We know the memory controller has been vastly improved on Zen 2, and this has been documented extensively. Also, @ CAS18, these timings are pretty loose, so nothing out of the ordinary in all reality. As for the 4.6GHz overclock on all 12 cores, the 16-core ES sample did 4.3GHz on all cores. Ergo, around a 7% higher all core overclock for the 12-core than the 16-core but the 12-core with 25% less cores than the 16-core. Seems easy to extrapolate. 

 

7.) "AMD will commercialize Ryzen 3000 @ 5GHz clocks in March 2020." This last one here, again, while relies on some previous comparisons for increased belief, is not too far off from being possible. By March 2020, yields for Zen 2 (3000 series) will be greatly improved. Which means, higher binned dies are much easier to procure. Therefore, translating to a possibility of such a chip being able to be released. Also, again, the actual core count of this processor is not yet revealed. Thus, dismissing its validity doesn't seem fair. Finally, if you saw the trend of Polaris. AMD definitely knows how to improve upon previous products with refreshes. 480 to 580 to 590 saw nice gains in clockspeeds spread out in a small amount of time: around 22-25% increase in clockspeeds from the RX 480 to RX 590. If Intel can do it with their "+++++" silliness, I'm sure AMD can as well. 

 

Of course it's wise to not run into any rumor with blind faith, which is why I understand your skepticism, but nonetheless there are definitely plenty of glimpses of hope in most of these rumors. Besides, we are not too faraway from much of this being confirmed (as early as 10 days from today).

[porina]

Quote

Clarification on 13% or 15%: Prior to presenting onstage @ Computex , we planned to use a Cinebench 1T-derived IPC figure. That is 13%. Slides were made and shared. In the end, we decided it would be best to use a more rigorous SPECint-derived figure. That is 15%.

According to linkedin writer is AMD senior technical marketing manager. Source:

 

 

[CarlBar]

7 hours ago, BiG StroOnZ said:

 

What exactly do you have a hard time believing? As there is plenty of information in there that is definitely plausible. To be fair, most of it seems pretty feasible. I mean, we can go over some of it:

 

1.) "4.8GHz all-core overclock." We know there is a SKU in the lineup already that can do 4.6GHz out of the box. A 4% overclock is not really all that much (going from 4.6 to 4.8GHz).

 

2.) "4.4GHz Zen 2 = Intel Core i9-9900k @ 5GHz in CB." There were many people in this thread that were pretty certain, that if what AMD exclaimed in their slides and presentation was correct. That essentially, AMD would have surpassed Intel in IPC. Which would in turn mean they would be able to match Intel at lower clocks. Now even if this is incorrect for the 8c16t part, since the "leak" doesn't explain specifically what part this is that beats the 9900k @ 5GHz, it could simply be a case of it being the 12c24t part that is pulling this off.  

 

3.) "5GHz is doable, but a challenge." Okay, so they admit 5GHz is probably not likely but maybe Golden Samples can. I would agree.

 

4.) "OC for OC, Ryzen is faster." (See 2).

 

5.) "5GHz boost isn’t infeasible." (See 3)

 

6.) "5GHz all core is pretty much a no-go." (Something we already established from 3 & 5; perhaps a couple of cores boosting to 5GHz though).

 

7.) "1.35V for all core 4.5GHz." Don't see what's so crazy about this one. They already have chips that do 4.4-4.6GHz boosts. Additionally, 1.35-1.4v was average voltage for all-core overclocks in previous iterations of Ryzen. 

 

8.) "AMD Ryzen 3900X 12-core processor overclocked to 4.6GHz across all cores with memory @ 4000MHz (CL18 timings)." We know the memory controller has been vastly improved on Zen 2, and this has been documented extensively. Also, @ CAS18, these timings are pretty loose, so nothing out of the ordinary in all reality. As for the 4.6GHz overclock on all 12 cores, the 16-core ES sample did 4.3GHz on all cores. Ergo, around a 7% higher all core overclock for the 12-core than the 16-core but the 12-core with 25% less cores than the 16-core. Seems easy to extrapolate. 

 

7.) "AMD will commercialize Ryzen 3000 @ 5GHz clocks in March 2020." This last one here, again, while relies on some previous comparisons for increased belief, is not too far off from being possible. By March 2020, yields for Zen 2 (3000 series) will be greatly improved. Which means, higher binned dies are much easier to procure. Therefore, translating to a possibility of such a chip being able to be released. Also, again, the actual core count of this processor is not yet revealed. Thus, dismissing its validity doesn't seem fair. Finally, if you saw the trend of Polaris. AMD definitely knows how to improve upon previous products with refreshes. 480 to 580 to 590 saw nice gains in clockspeeds spread out in a small amount of time: around 22-25% increase in clockspeeds from the RX 480 to RX 590. If Intel can do it with their "+++++" silliness, I'm sure AMD can as well. 

 

Of course it's wise to not run into any rumor with blind faith, which is why I understand your skepticism, but nonetheless there are definitely plenty of glimpses of hope in most of these rumors. Besides, we are not too faraway from much of this being confirmed (as early as 10 days from today).

 

I pretty much agree. GN reported that they'd heard rom sources that where playing with 16 core samples that where getting 300w on a high spec ambient water loop. Given we can predict the power draw at 4.4ghz from the 8 core, (130w), it was obvious there was going to be significant headroom as even a truly terrible voltage curve was going to demand some really serious clock uptick. The question was allways exactly how high they can get things.

 

I also don;t see an issue with a 5ghz chip in future if the rest of this is accurate. Epyc gets the most voltage efficient silicon normally. Use those in a desktop chip and you can get higher clocks out of the same power consumption, and ultimately it';s the thermal effects of high power consumption that limit peak clocks.

 

In fact that may be why we haven't seen the 16 core ethier, if they want to keep it as easy as possible for non-hardcore enthusiasts to cool it may demand higher quality silicon than they're willing to put in desktop chips right now. I'm sure segmentation is a factor too, but i wouldn't be surprised if that was an extra factor.

 

 

 

Also the latency is very nice as it strongly suggests the 12 and 16 core will scale very well compared to the 8 core. Thats probably another reason we don't have the 16 core. If the 12 core is achieving nearly perfect multi-core scaling the 12 core may be performing better than they initially hoped. In comparison to the 8 core making it more of an upgrade and lessening the need for a higher still skew to hit the performance target, (which is ultimately a big part of why segmentation exists).

 

Looking forward to when the review embargo lifts on the 7th.

[Stefan Payne]

Interesting.

L1I Cache got halved. Is now 32kiB as well!

L1I Cache on Zen1 was 64kiB.

 

Anyway, benchmarks of a 3600 was spotted - its about on par with the 2700X on Multi Core...

 

And someone also mentioned that its a VIPT Cache.

Maybe @leadeater might say more about that...

[leadeater]

5 hours ago, Stefan Payne said:

And someone also mentioned that its a VIPT Cache.

Maybe @leadeater might say more about that...

Likely it's to address performance. Smaller caches are actually faster and VIPT is also lower latency compared to PIPT, I suspect this is in some way required because of the doubled FP throughput.

 

However I suspect I know where you saw this information and it was taken from an OC 16 core, could be incorrect.

[Robin_Code]

Guys, 

 

I have a €1000 budget for 3 components (CPU, Motherboard, Graphics Card) now that the AMD CPU's / Graphics card will get a huge boost, I was wondering what would be the perfect €1000 combination. I have a RTX 2070 but it's probably not the best bet since I already spend €530 on it. Does anyone have a better trio for this?

[MeatFeastMan]

6 minutes ago, Robin_Code said:

Guys, 

 

I have a €1000 budget for 3 components (CPU, Motherboard, Graphics Card) now that the AMD CPU's / Graphics card will get a huge boost, I was wondering what would be the perfect €1000 combination. I have a RTX 2070 but it's probably not the best bet since I already spend €530 on it. Does anyone have a better trio for this?

Don't replace your 2070. The only upgrade for GPU that would make sense is a 2080ti, which...doesn't make sense because it costs an arm and a leg. Just keep your 2070, save some money and just get a cpu and motherboard.

 

Or do you mean that you are planning on buying the 2070? If you haven't actually bought it yet (Hard to tell what you meant) then go for the new Navi if it's a good price. Otherwise, vega 56 or rtx 2060 are good options. Don't buy the 2070 if you haven't got one yet. It's a horrible value card, especially at 530 euros.

 

If you do stuff other than gaming then maybe the 12 core would make sense. If not, then go for the 3700x and a good x570 motherboard. Thankfully you won't see any bad x570 boards because they've gone overkill on even the lesser boards from what we've seen.

 

[RejZoR]

33 minutes ago, MeatFeastMan said:

Don't replace your 2070. The only upgrade for GPU that would make sense is a 2080ti, which...doesn't make sense because it costs an arm and a leg. Just keep your 2070, save some money and just get a cpu and motherboard.

 

Or do you mean that you are planning on buying the 2070? If you haven't actually bought it yet (Hard to tell what you meant) then go for the new Navi if it's a good price. Otherwise, vega 56 or rtx 2060 are good options. Don't buy the 2070 if you haven't got one yet. It's a horrible value card, especially at 530 euros.

 

If you do stuff other than gaming then maybe the 12 core would make sense. If not, then go for the 3700x and a good x570 motherboard. Thankfully you won't see any bad x570 boards because they've gone overkill on even the lesser boards from what we've seen.

 

I kinda follow the rule of doubling the performance. When next gen graphic card is 100% faster than my existing one, I replace it. Only exceptions I make is when cards die and you have to get one from scratch. But that's very rare. In all the years it happened only once.

[Robin_Code]

6 minutes ago, RejZoR said:

I kinda follow the rule of doubling the performance. When next gen graphic card is 100% faster than my existing one, I replace it. Only exceptions I make is when cards die and you have to get one from scratch. But that's very rare. In all the years it happened only once.

 

So like upping from a GTX1660 to a RTX 2080 or what do you mean?

[RejZoR]

5 minutes ago, Robin_Code said:

 

So like upping from a GTX1660 to a RTX 2080 or what do you mean?

Yes. Or waiting for so long that mid tier reaches that level of performance.

[XenosTech]

Only thing I got disappointed about was they kept the same product stack  Was looking for r3's to be 4c/8t with 2 cpu's the x and non x variants and then 6c/12t x and non x and then r7's with 8c/16t x and non x. Only expected 4 cpu's to be under the stack when the ipgu (navi) versions rolled out

[Trixanity]

1 hour ago, XenosTech said:

Only thing I got disappointed about was they kept the same product stack  Was looking for r3's to be 4c/8t with 2 cpu's the x and non x variants and then 6c/12t x and non x and then r7's with 8c/16t x and non x. Only expected 4 cpu's to be under the stack when the ipgu (navi) versions rolled out

I'm not sure I follow. The stack you want to see appears to be exactly the one you got except there's now an R9 at the top. APUs are quite removed from the CPU stack and it'll be at least 6 months before we can expect to hear about Navi APUs so the stack would be quite barebones if you only expected 4 products considering there's 4 tiers.

[kuddlesworth9419]

I'll wait for benchmarks before I get excited for obvious reasons. 

[justpoet]

On 5/30/2019 at 11:11 AM, VegetableStu said:

X570 block diagram's out

 

https://www.hardwareluxx.de/index.php/news/hardware/mainboards/49743-amd-erlaeutert-die-am4-plattform-fuer-3rd-gen-ryzen-prozessoren.html

This is interesting.  Not only does it show that X570 is rather capable, but it also has another interesting possibility.  Low end small boards can basically skip the chipset to have 1 graphics card slot, 4x usb, 2 sata.  Then with the 4x "chipset downlink" they can still do things like more sata and/or wifi/ethernet on a small and cheap chipset if they want to.

 

I don't think anybody will, due to it only being cheep in scale production, which I don't think it'll hit.  However, that could make this much more interesting as NUC competitors, especially if they just used the chipset downlink to run thunderbolt/ethernet/wifi.  I wonder how different the APU variants will be off the chip itself in this regard, or if it will simply take the 16x graphics portion and be done with it.

[Stefan Payne]

5 hours ago, leadeater said:

However I suspect I know where you saw this information and it was taken from an OC 16 core, could be incorrect.

No, it was a 6 Core Ryzen 3600 in Geekbench, not a 16 Cores.

 

Was identified as "AuthenticAMD Family 23 Model 113 Stepping 0"

[leadeater]

1 hour ago, Stefan Payne said:

No, it was a 6 Core Ryzen 3600 in Geekbench, not a 16 Cores.

 

Was identified as "AuthenticAMD Family 23 Model 113 Stepping 0"

Regardless it does make sense to move from PIPT to VIPT as when it's working well it's much better and you need less of it, could even be one of those edge areas that makes Intel faster in gaming but I wouldn't get my hopes on on that. VIPT and TLB need to be working well together otherwise there's potential for performance to regress.

 

 

[Guest]

On 5/26/2019 at 10:14 PM, Origami Cactus said:

-snip-

ponystyle faster then light .... don't rip him because his predictions where off... be human ... lol

[XenosTech]

5 hours ago, Trixanity said:

I'm not sure I follow. The stack you want to see appears to be exactly the one you got except there's now an R9 at the top. APUs are quite removed from the CPU stack and it'll be at least 6 months before we can expect to hear about Navi APUs so the stack would be quite barebones if you only expected 4 products considering there's 4 tiers.

Not quite, r3's are 4/4, r5's are either 4/8 OR 6/12. With the r7's it makes no sense having 4 exact cpus with marginally higher clocks (unless they were doing what intel did with the "T" variants of their processors.) Come to think of it there wasn't even a 2800(x) which was good move on the refresh. They could just bump up athlons to 4c/4t, make r3's 4c/8t, r5's then only have 6c\12t instead of a mix and then there's no change to r7's are they were always just 8c\16t. R9's being 12c/24t and 16c/23t aren't that huge of a deal since those would be halo products/budget hedt for people who really need that kind of workhorse for cheap. I just hoped they would have simplified the stack.

[KaitouX]

15 minutes ago, XenosTech said:

Not quite, r3's are 4/4, r5's are either 4/8 OR 6/12. With the r7's it makes no sense having 4 exact cpus with marginally higher clocks (unless they were doing what intel did with the "T" variants of their processors.) Come to think of it there wasn't even a 2800(x) which was good move on the refresh. They could just bump up athlons to 4c/4t, make r3's 4c/8t, r5's then only have 6c\12t instead of a mix and then there's no change to r7's are they were always just 8c\16t. R9's being 12c/24t and 16c/23t aren't that huge of a deal since those would be halo products/budget hedt for people who really need that kind of workhorse for cheap. I just hoped they would have simplified the stack.

Ryzen 3 have no official information but most likely it is going to be 4 cores 8 threads, both Ryzen 5(3600 and 3600x) are 6 cores 12 threads, Ryzen 7 is only the 3700x and 3800x, from what we know there are no other variants of Ryzen 7, 3700 and 3800 aren't a thing, at least not yet.

[Stefan Payne]

Ryzen 3 are probably Picasso, the Zen2 based Renoir is announced for next year...