AMD 2800x with 10 Cores spotted

[Joseph Combs]

A new benchmark has leaked and it shows that the AMD Ryzen 2800X might well have 10 cores and 20 threads.

https://segmentnext.com/2018/09/10/amd-ryzen-2800x-10-cores/

 

[The Stilt]

Nope.

 

[Joseph Combs]

Based on ...... what?

 

[epsilon84]

Nope.

Fake I assume?

 

[Joseph Combs]

Time will tell .... just found it interesting as I have thought it was strange that AMD didn't announce a 2800x with the original launch. Figured there was a reason.

 

[Hitman928]

Based on ...... what?

There's no way to get to 10 cores with the current CCX configuration of Ryzen.

 

[Joseph Combs]

So its either fake, or AMD has new tricks up their sleeve.

 

[Hitman928]

So its either fake, or AMD has new tricks up their sleeve.

Yeah. . . it's fake.

 

[dacostafilipe]

I vote for fake.

 

[NostaSeronx]

Unless, it is a 7nm die harvest of two 8-core CCXs with 3-cores of each being disabled. That is the only way to get that.

7nm FinFET with Rome was taped out 1H2018 and AMD is sampling in servers in 2H2018. Releasing Zen2 CPUs in a die harvested Ryzen would be immediate profit while yields grow for EPYC, etc.

 

[Joseph Combs]

Agreed.

 

[psychobriggsy]

It's definitely fake. Below is purely conjecture IF if turned out to be true. Which is won't.

10-core obviously doesn't fit into any current Zen die.
AM4 doesn't support dual-dies, but if it did, you would have 2+2+2+4 cores enabled, and AMD has always kept the enabled-cores/CCX the same (6C = 3+3, for example).

So it would have to be a new piece of silicon. Out of nowhere, no rumours, no nothing.
It would be dual-CCX, as 10 divides by 2 but not 3 or 4.
And each CCX would have 6 or 8 dies, with 1 or 3 cores disabled on each.
But it's not a Zen 2 design, because AMD would definitely call that 3800X - and that's months off anyway (likely March/April next year IMO).

So is it a new 12nm processor that's never appeared on a roapmap, that offers 6+6 cores on 12nm, optimised for desktop loads?

Logic here: Zen 2 is Epyc-targeted at first. Maybe all the Zen 2 7nm dies will go to Epyc. AMD need a temporary (8-10 month) SKU to compete against Intel's high end and give them flexibility.

So this would be Zen 1 core, with a Zen 2 CCX crossbar, 12 cores on a die, 250mm^2 (maybe native 12nm this time, so could be a little smaller).

Also gives AMD room for an AM4 2820X with 12C.

Problem with this entire concept is that it doesn't sort out the single-thread performance issue against a 5GHz Intel in certain games. Now if it was a cunning plan to use 'sampling' or 'early ramp' Zen 2 dies, not worry about the branding not aligning for Zen 2, or losing a lot of the future hype for Zen 2 3000-series Ryzens, and then use the Zen 2 IPC gains ... (I note they could disable some of the Zen 2 only features in hardware, e.g., AVX-512 if present, etc).


But why would this core design exist but not be mentioned publicly or on a roadmap? Simple - insurance. 7nm was a risk, so it is likely that AMD had a back up design on the old process ready to go just in case 7nm was delayed a year.

 

[PeterScott]

Time will tell .... just found it interesting as I have thought it was strange that AMD didn't announce a 2800x with the original launch. Figured there was a reason.

Simple. Too many SKUs and 1800x was essentially a flop that had to be sold off at 1700 prices.

 

[NostaSeronx]

Simple. Too many SKUs and 1800x was essentially a flop that had to be sold off at 1700 prices.

The 2800X was apparently saved for a later day via the echo chamber:
https://digiworthy.com/2018/04/16/amd-ryzen-7-2800x-intel-i7-9700k/
https://digiworthy.com/2018/04/22/amd-ryzen-7-2800x-release/
https://www.techpowerup.com/243521/amd-is-saving-the-ryzen-7-2800x-for-a-rainy-day
https://ocaholic.ch/modules/news/article.php?storyid=17765
https://hothardware.com/news/amd-ryzen-7-2800x-8-core-intel-coffee-lake-s
https://www.pcgamesn.com/amd-ryzen-7-2800X-release-date-performance
https://www.tweaktown.com/news/61629/amd-teases-ryzen-7-2800x-10c-20t-beast-4ghz/index.html
https://www.theinquirer.net/inquire...pu-to-take-on-intels-8-core-coffee-lake-chips

https://www.techpowerup.com/img/acTToiQiQ6rH5MoO.jpg
Pepperidge Farm always remembers.

 

[Joseph Combs]

Another story that agrees it doesn't make sense given what we know .....
https://www.guru3d.com/news-story/amd-10-core-ryzen-2800x-in-response-to-intel-core-i9-9900k.html

 

[Joseph Combs]

And another ....
https://www.overclock3d.net/news/cp..._2800x_10-processor_cinebench_score_appears/1

 

[PeterScott]

And another ....
https://www.overclock3d.net/news/cp..._2800x_10-processor_cinebench_score_appears/1

Other "news" sites passing along click bait rumors isn't confirmation.

Click bait regurgitation has unfortunately become the substitute for any kind of real reporting/analysis.

Any kind of thoughtful analysis would highlight this rumor to be about as real as a 3 dollar bill.

 

[LTC8K6]

Other "news" sites passing along click bait rumors isn't confirmation.

Click bait regurgitation has unfortunately become the substitute for any kind of real reporting/analysis.

Any kind of thoughtful analysis would highlight this rumor to be about as real as a 3 dollar bill.

The other links are saying it's unlikely to be real.

 

[LTC8K6]

Should say "Ryzen 7 2800X" but it says "Ryzen R7 2800X".

 

[psychobriggsy]

Should say "Ryzen 7 2800X" but it says "Ryzen R7 2800X".

Yup. An obvious mistake ... too obvious

Also "10-core" instead of "Ten-core" as 2700X is "Eight-core".

 

[PeterScott]

The other links are saying it's unlikely to be real.

Fair point, I wasn't going to pay much attention to sites regurgitating this baseless rumor. I just glanced and saw the same image.

 

[Topweasel]

Nope to 10 core 2800x.

The 1800x was not a flop it served it's purpose. When Zen launched there was 2 other CPU's on the market that 8 or more cores that was as fast or faster. One was $1k and the other $1700 I believe. The 1800x existed to A.) Increase margins selling to people who wanted the fastest non- $1k CPU. Against the 6900k the 1800x was a major value. B.) To show that Ryzen was a premium performance chip allowing them to sell more 1700s and 1700x as major values.

With Zen+/Ryzen 2k the need for an expensive flagship product wasn't needed. They have Threadripper for HEDT. Intel has released more 6 and 8c parts. The room for a CPU that cost more than the 2700x and still a value compared to Intel products was small. Not worth dirtying the waters. The 2800x doesn't exist because the spot it needed to fill is gone. It can be a future replacement for the 2700x with more cherry picked dies later if AMD feels they need to lower their prices across the board and still want to try to keep something at the $330-$350 point.

But none of that gets away from a major fact. AMD doesn't have a 10c Die. Hell Intel barely has this 8c die for the coffeelake and only has it because they knew years ago (~2015) that they were going to be really late with 10nm. They also around the same time got word the Zen might actually be competitive and was an 8c chip. This is how long CPU designs take. Intel is barely getting out for 2018, basically the only thing they will have to sell for most of 2019 a chip that they would have started in 2015 after they figured out a 6c coffee lake wasn't going to be enough. Zen+ is a water treading release its sole existence is have something to release that was a little faster to hold over till Zen 2. No way did AMD decide in 2015/2016 they the were going to have two Zen+ non-APU dies.

 

[NostaSeronx]

I speculate that it is indeed possible for a 10-core solution.

7nm EPYC 2 samples will probably strive for 48-core/56-core/64-core.

https://www.anandtech.com/show/1058...-naples-32cores-dual-socket-platforms-q2-2017
AMD showed off Zen EPYC in 2016.
https://www.anandtech.com/show/12912/amd-zen-2-update-7nm-epyc-in-labs-now-launching-in-2019
AMD says EPYC2 is labs now, and sampling then in 2H2018. This matches up with EPYC 1.0 sampling in 2016.

If there is a 10-core die in the 2000 series, it is based on 7nm FinFET. Four dies to achieve 64C/56C/48C, leaves open for four die salvages of 10-cores. Leaving 10-core for a special Ryzen SKU in the Ryzen 2000 series.

 

[Topweasel]

If it's 7nm it's not Ryzen 2k. There would be a lot more different than just a die shrink.

 

[psychobriggsy]

NB: I still think this is fake.

No way did AMD decide in 2015/2016 they the were going to have two Zen+ non-APU dies.

7nm insurance design.

You have your 7nm design, you go all out on this - 8 core CCX, 16 core dies, 64 core Epycs.

But ... in 2016 and 2017 AMD didn't know that TSMC would pull through on their 7nm promise in H2 2018. But they needed Epyc 2 by Q1 2019.

So you design a 12nm backup. You have power and die size issues here, you cannot go all out like with the 7nm design. So you go for a 6 core CCX, 12 core dies, 48 core Epyc 2. This 12nm was likely taped-out and probably had some silicon spun.

This fits with some rumours that it would be 48C initially.

But then GlobalFoundries tell you they won't be doing 7nm (initially this was probably a strong hint, or saying it will be late, or similar). Or TSMC say that Apple bought all the space. Result: your booked Zen 2 runs at TSMC cannot cover all of your product need for consumer space.

7nm consumer Zen 2 is put back, H2 2019 say. Not a disaster to wait that long really, but right now, Intel is hurting on their 10nm process, and don't you just want to stick the knife in?

And you have a 12C Zen or Zen 2 based 12nm die, and you also need to make stuff on GlobalFoundries because of wafer agreement. And you also want to have something to compete in some metric (i.e., multi-threaded performance) against Intel's 8C consumer chips coming in Q4.

So you leave a space in your line-up just in case you need a 2800X later in the year. You get a thousand wafers of your new design fabbed, you fuse off some of the Zen 2 features for next year's products. You package up the samples and send them out to OEMs and partners.

Given that timeline, I'd expect to see initial leaks happening right about now.

NB: I still think this is fake.