16-Core Ryzen 9 3000 Series ES Sample spotted & Zen 2 is a Memory OC Beast, DDR4-5000 Possible! (Updated)

[BiG StroOnZ]

Quote

 

 

Twitter phenomenon Tum Apisak once again threw some interesting information online. First off, he's often right and is respected for what he outs, so we're not questioning the validity of the content. Then again at this stage you need the be careful with information like that.

 

In his tweet, he mentions that he has seen or has hands on an ES (engineering sample) Ryzen 3000 sample. I name it Series 3000 sample as we do not know the ES model he has hands on. He mentions a 16-core/32-thread chip that clocks in at a 3.30 GHz base frequency with 4.20 GHz Precision Boost frequency. If the tweet is valid this unidentified Ryzen 3000-series with 16-cores would be named "Ryzen 9". And if he can back his claim, (you do need to keep in mind it is an engineering sample (ES)) these are often clocked lower and just for testing purposes. TUM_APISAK left a Thai note, which translates he might share a screenshot of the processor soon. We'll update once that happens. 

 

Also, with Zen 2, the company decided to separate the memory controller from the CPU cores into a separate chip, called "IO die". Our resident Ryzen memory guru Yuri "1usmus" Bubliy, author of DRAM Calculator for Ryzen, found technical info that confirms just how much progress AMD has been making.

The third generation Ryzen processors will be able to match their Intel counterparts when it comes to memory overclocking. In the Zen 2 BIOS, the memory frequency options go all the way up to "DDR4-5000", which is a huge increase over the first Ryzens. The DRAM clock is still linked to the Infinity Fabric (IF) clock domain, which means at DDR4-5000, Infinity Fabric would tick at 5000 MHz DDR, too. Since that rate is out of reach for IF, AMD has decided to add a new 1/2 divider mode for their on-chip bus. When enabled, it will run Infinity Fabric at half the DRAM actual clock (eg: 1250 MHz for DDR4-5000).

 

1usmus also discovered that the platform adds a SoC OC mode and VDDG voltage control. We've heard from several sources that AMD invested heavily in improving memory compatibility, especially in the wake of Samsung discontinuing its B-die DRAM chips.

 

Source 1: https://www.guru3d.com/news-story/sixteen-core-ryzen-9-3000-series-socket-am4-es-sample-at-4-2-ghz-spotted.html

Source 2: https://www.techpowerup.com/255386/amd-ryzen-9-3000-is-a-16-core-socket-am4-beast

Source 3: https://www.techpowerup.com/255405/amd-ryzen-3000-zen-2-a-memory-oc-beast-ddr4-5000-possible

 

So a few things, while the clockspeeds appear to be lower than expected according to previous rumors. Please note that ES chips generally are clocked lower, so this doesn't rule out higher clockspeeds as of yet. Although, current speeds aren't too terrible for the amount of cores being offered. At least a 200-300MHz clockspeed boost could be plausible in final samples. As for the memory; this is great and with higher speed kits of ram, combined with the IF - will show some serious gains. Especially considering that we currently are seeing Ryzen top out at around 3000MHz. I cannot even imagine what performance gains with a 5000MHz kit would look like!

SiSoftware Benches: 

Quote

 

[Princess Luna]

3 minutes ago, valdyrgramr said:

!

AdoredTV has liked your comment.

[Firewrath9]

Intel 14nm+++++++ ES 10 Core

Base Clock 6.5ghz

Turbo Boost 8.0ghz

voltage 4.3v

TDP 500W

Mobo Z495X

 

[Jurrunio]

too fast even for ES chips at stock when all you care is if it bugs out or not. Only QS chips run near retail speeds

[jiyeon]

Fiction until confirmed by Lisa Sue's mouth herself.

[justpoet]

Honestly, those clocks seem about right for a 16/32 mid-timeline ES chip (the frequencies happen to be really close to rumored Ryzen 3 frequencies, so simply using the same frequencies across the board makes some sense for testing purposes).  We will see soon though.

 

At that point in core count your doing it for a workstation anyway, which will be helped more by the threads, and packing that many cores on the chip generally makes keeping frequency a challenge.  I would expect the gaming enthusiast crowd, who generally care more about frequency, to be looking at the next chip down in the lineup if that is the case, and then overclocking it further than the 16 core will go in an equal quality silicon lottery…unless they simply produce them the same and bin/disable the worst cores on the top chips to drop down to the next core count, which also makes some sense at a production level.

 

For reference, this was what most of the rumors expected (X chip first for clock speeds):

Ryzen 9 - 16/32 - 4.3/5.1 and 3.9/4.7

Ryzen 7 - 12/24 - 4.2/5.0 and 3.8/4.6

Ryzen 5 - 8/16 - 4.0/4.8 and 3.6/4.4

Ryzen 3 - 6/12 - 3.5/4.3 and 3.2/4.0

[porina]

Having a 5000 ram speed setting is no guarantee it'll get anywhere near that. It is way beyond DDR4 specifications and maybe only something for extreme overclockers to play with.

[yian88]

5 minutes ago, seoz said:

Fiction until confirmed by Lisa Sue's mouth herself.

Fiction until reviewers get a sample and test it out.

If performance per core and IF is not better then AMD chips will still be the underdog.

[Sakkura]

33 minutes ago, BiG StroOnZ said:

As for the memory; this is great and with higher speed kits of ram, combined with the IF - will show some serious gains. Especially considering that we currently are seeing Ryzen top out at around 3000MHz. I cannot even imagine what performance gains with a 5000MHz kit would look like!

Bear in mind there are diminishing returns. Still a nice improvement, especially for ease of memory compatibility, but not necessarily a major performance jump in its own right.

 

6 minutes ago, porina said:

Having a 5000 ram speed setting is no guarantee it'll get anywhere near that. It is way beyond DDR4 specifications and maybe only something for extreme overclockers to play with.

Memory manufacturers routinely exceed JEDEC specs. Nothing to worry about there.

 

DDR4-4800 is already on store shelves.

[porina]

1 minute ago, Sakkura said:

DDR4-4800 is already on store shelves.

Good luck getting it working at that speed. High speed ram is a compatibility minefield.

 

I'm not saying it will be impossible to potentially run ram at 5000, just that it will be nowhere near routine.

[BiG StroOnZ]

7 minutes ago, Sakkura said:

Bear in mind there are diminishing returns. Still a nice improvement, especially for ease of memory compatibility, but not necessarily a major performance jump in its own right.

 

While of course there are diminishing returns, Ryzen has shown a near linear increase in performance with memory speed frequency increase:

 

 

https://www.legitreviews.com/ddr4-memory-scaling-performance-with-ryzen-7-2700x-on-the-amd-x470-platform_205154

 

We have yet to see exactly what those limitations entail though, and we will definitely see them with the Ryzen 3000-series if these rumors of 5000MHz memory compatibly turn out to be true. 

[Sakkura]

1 minute ago, BiG StroOnZ said:

 

While of course there are diminishing returns, Ryzen has shown a near linear increase in performance with memory speed frequency increase:

 

 

https://www.legitreviews.com/ddr4-memory-scaling-performance-with-ryzen-7-2700x-on-the-amd-x470-platform_205154

 

We have yet to see exactly what those limitations entail though, and we will definitely see them with the Ryzen 3000-series if these rumors of 5000MHz compatibly turn out to be true. 

To me that looks like almost zero improvement past 3466.

 

But you're right, we don't have all the info yet. A faster CPU architecture could need faster memory (and IF) to reach its peak performance.

[justpoet]

Interestingly, they say that going faster is to cut the fabric speed in half, since the fabric had to match ram prior.  That would've basically slowed the whole chip communications down due to slower RAM, explaining a lot of the ram speed bottleneck.  If they're now thinking to cut that in half, there very well may be a tipping point where faster ram is actually slower for the base level of it that needs to cut the fabric speed in half.  If they're less reliant on that fabric speed though, because chiplets and double the job from the same fabric clock, then it may still work out.  Nobody really knows except the chip designers at this point, since we don't know the actual implementation details.  Let's just wait for some actual tests to be shown.

[BiG StroOnZ]

23 minutes ago, Sakkura said:

To me that looks like almost zero improvement past 3466.

 

But you're right, we don't have all the info yet. A faster CPU architecture could need faster memory (and IF) to reach its peak performance.

 

Yeah, there's definitely more to it that has not yet been revealed. Since on average, people usually plateau at around 3200MHz currently with the Ryzen 2000-series. Also, note that in that previous graph, no Cas Latency timings were changed; all were ran at Cas 16. Meaning, there's even more improvement to eventually be seen.

 

Here's some examples when Latency is taken into account as well:

 

 

https://www.techpowerup.com/reviews/AMD/Ryzen_Memory_Analysis/12.html

[ravenshrike]

Do remember that Zen2 has a setting to clock the IF at half memory speed. I imagine this will reduce compatibility problems for memory above 3000-3400 quite a bit. That doesn't mean you'll see the IF running at that speed.

[GoldenLag]

2 hours ago, BiG StroOnZ said:

I cannot even imagine what performance gains with a 5000MHz kit would look like

Like pretty much none except in memmory sensetive applications. 

 

And the only way to get memmory to run the memmory that fast would be to put a 0.5x modifier on IF to remove that OC limiter. 

 

While IF isnt supposed to be fully locked to memmory speeds anymore, you apperantly are only able to change the ratio between IF and memmory speed. 

 

Edit: unless you are able to finetune the more than just running at 1,5 or 0,5 modifier. The gain from actually managing to run at 5000mhz would be highly situational as you woul d only be able to have eother great Ramspeed and low IF speed or ok Ramspeed an Great IF speed. 

 

Great for different workloads, but only for one workload at the time. 

 

Then again being able to select 5k speed doesnt equate to ever running at such speeds. 

[BiG StroOnZ]

14 minutes ago, GoldenLag said:

Like pretty much none except in memmory sensetive applications. 

 

Did you rush to reply, without first examining other posts to see if this surmising comment was accurate?

[GoldenLag]

16 minutes ago, BiG StroOnZ said:

 

Did you rush to reply, without first examining other posts to see if this surmising comment was accurate?

Yeah i forgot to elaborate on the downside of OC' ing to the extent on the 0,5x modifier. Will do an edit. It was something o intended to include

[TetraSky]

Taking this news with a large pinch of salt.

 

Would be great if it was real.... but somehow I highly doubt it.

[sazrocks]

2 hours ago, BiG StroOnZ said:

the memory frequency options go all the way up to "DDR4-5000",

Which means literally nothing. I can type 99 into the multiplier of my CPU; that doesn't mean that it will actually do it.

[BiG StroOnZ]

14 minutes ago, sazrocks said:

Which means literally nothing. I can type 99 into the multiplier of my CPU; that doesn't mean that it will actually do it.

 

Yes, but this isn't quite the same. For instance, the ASRock X470 Taichi Ultimate tops out at 3466MHz (as do most X470 motherboards); which is the top end of what a Ryzen chip could achieve. Of course not all, if not most, won't even be able to achieve that. However, there are instances where it was possible, as seen in the legitreviews testing posted. Legitreviews was even able to go way beyond that. Of course that is an overclock, it still nonetheless is impressive. Even if slightly true, the 5000MHz option gives more hope to acceptable frequencies like 3600MHz+ etc.

[Dabombinable]

Quote

Since that rate is out of reach for IF, AMD has decided to add a new 1/2 divider mode for their on-chip bus. When enabled, it will run Infinity Fabric at half the DRAM actual clock (eg: 1250 MHz for DDR4-5000).

I'm getting Pentium II flashbacks.

[S w a t s o n]

Apisak is a well known leaker and decoder, anyone doubting his leak is joking themselves.

 

Here here, to 5GHZ on all 16cores whether boost or OC

 

Edit: To be clear I mean at least 1 core turbo, i dont ever expect a factory 5ghz on all cores unless amd goes 9590 level

[PlayStation 2]

I'll wait for more information, but that being said, there's a genuine possibility that AMD could target Intel right at the heart and start making more enthusiast CPUs for the standard consumer sockets and only leaving the very ridiculous core counts for Threadripper if they still want to do that.

To me, them taking Threadripper off the road map and then making a 16-core Ryzen 9 does make a bit of sense.

[PlayStation 2]

4 minutes ago, S w a t s o n said:

Apisak is a well known leaker and decoder, anyone doubting his leak is joking themselves.

 

Here here, to 5GHZ on all 16cores whether boost or OC

I would say that considering how many leaks have come out that have been disproven in general (not really by this source but just in general), it's always good to hold a decent amount of skepticism. It will be very interesting, though, if AMD drops this bomb in a few months. I'd just be interested in how well it holds up.