AMD Readying a 10-core AM4 Processor to Thwart Core i9-9900K?

[dj-electric]

There comes a point where adding more cores isn't the solution to the problem, because the problem wasn't "not enough cores".

 

[Valantar]

Don't remember where I saw it, but this has been debunked. It's fake. Why?

• CineBench scores can be manipulated just by editing a text file
• The description of the CPU doesn't match AMD naming conventions in Cinebench (among other things, it says "10-core", while AMD CPUs are described with words, not numbers ("six-core", "eight-core")
• And last but certainly not least: There's no way to get a 10-core AM4 chip without new silicon, which would mean this is 7nm. There's no way they'd launch their first 7nm CPUs as an afterthought like this.

Please stop treating this like it's real.

 

[nemesis.ie]

I don't put much stock in the validity of this "leak".

However, just for fun, imagine for a moment:

1. 7nm samples are being tested for Epyc right now.
2. They have run through "quite a few" wafers testing 7nm.
3. The yields are pretty terrible, but they have managed to get some working cores on some dies - the CCX is now 8 cores, meaning 16 cores on a fully working die.
4. They put out "2800X" as a stepping stone to the full Ryzen 3/Zen 2 release using up these not 100% working dies giving anything from 2 to 14 cores (assuming ones with 16 would go to Epyc or an early TR release).

Pure speculation/wishful thinking here of course, but is it possible that there are usable, fully tested 7nm dice with 8 or more cores working available in "reasonable quantities" that could be packaged and sold?

Obviously if they did have such a product and wanted to sell it at a 9900k competitive price they would probably need to have a lot of them available as they would likely go out of stock very quickly.

They could of course make it a "halo product" for this generation and charge extra.

/wishful thinking

 

[jpvalverde85]

Too much r&d for a 10 core. Dont think Ms. Su would aprove it, and if zen2 AMD will not release it as 2000 series.

 

[Space Lynx]

Hmm I did not realize that Apple, nvidia, etc lots of big names are in line for 7nm production before AMD... that has to mean Nvidia will have 7nm gpu's next year, I am sure they are throwing their big wallet around the TSMC building to make sure AMD is behind them on the production schedule... this does not bode well for AMD, people are already sick of waiting so many years for competitive GPU's, heh.

 

[HTC]

This is highly unlikely @ this time. Perhaps after gen 2 is released (as in not the 2000 series) but not in this generation.

Cinebench scores can be faked rather easily, apparently:



Seen the above pic in a similar topic @ AnandTech, a few days ago ...

 

[Caring1]

This is highly unlikely @ this time. Perhaps after gen 2 is released (as in not the 2000 series) but not in this generation.

Cinebench scores can be faked rather easily, apparently:

View attachment 107036

Seen the above pic in a similar topic @ AnandTech, a few days ago ...

If a soggy loaf of bread can do that, I would love to know how much better a potato is.

 

[HTC]

If a soggy loaf of bread can do that, I would love to know how much better a potato is.

That makes 3 os us: you, me ... and the potato ...

 

[Valantar]

Yeah, that wasn't difficult.

 

[Gasaraki]

That leak was so fake 2 weeks ago. AMD can't make a 10 core Ryzen right now due to the way their architecture works.

Hmm I did not realize that Apple, nvidia, etc lots of big names are in line for 7nm production before AMD... that has to mean Nvidia will have 7nm gpu's next year, I am sure they are throwing their big wallet around the TSMC building to make sure AMD is behind them on the production schedule... this does not bode well for AMD, people are already sick of waiting so many years for competitive GPU's, heh.

He's just throwing shit out there. He doesn't really know that Apple, nVidia, are in line for 7nm before AMD. Think about it, RTX20x0 is not 7nm, why would nVidia be in line now? More like end of next year in preparation for RTX21x0 or Titan RTX.

 

[Vya Domus]

that has to mean Nvidia will have 7nm gpu's next year, I am sure they are throwing their big wallet around the TSMC building to make sure AMD is behind them on the production schedule

Thankfully TSMC isn't run by morons, they don't give a shit about any of that. Their goal is to sell as many wafers as possible.

He doesn't really know that Apple, nVidia, are in line for 7nm before AMD.

Apple already is already about to ship 7nm chips inside the new phones. No one is in line before anyone, whoever has the designs ready and their yields are met gets the chips. Given those chips are in the region of 100 mm^2 or less no wonder they were the first to get them.

 

[newtekie1]

There's no way to get a 10-core AM4 chip without new silicon, which would mean this is 7nm. There's no way they'd launch their first 7nm CPUs as an afterthought like this.

The only thing I disagree with is that new silicon would require 7nm. There is no reason, that I see, that they couldn't make changes to the CCX and leave it on 12nm. I mean, they already have the 12nm design down, so adding a few more cores to it shouldn't be a big task.

That said, I still don't think this is real.

 

[GorbazTheDragon]

Yep, called bullshit on this as soon as I saw 10... Going to have to give me something more convincing.

I would be much more believable if it was 12 cores too..

 

[Valantar]

The only thing I disagree with is that new silicon would require 7nm. There is no reason, that I see, that they couldn't make changes to the CCX and leave it on 12nm. I mean, they already have the 12nm design down, so adding a few more cores to it shouldn't be a big task.

That said, I still don't think this is real.

I didn't say it would require 7nm. If it wasn't clear, I simply meant that launching a new piece of silicon at this time on 12nm makes no sense whatsoever. Why? Because what constitutes a "big task" is mighty relative.

First, let's ignore how exactly they're getting to 10 cores, and focus on die production. Taping out and ramping a production line for a brand-new large-size silicon design, even on a well-known process, is a multi-million-dollar investment at the very least. Also, the process from initial tape-out to volume production is at least 6 months. Which means they'd need to sell a lot of these to cover those costs alone. Is it possible? Absolutely. But considering their 7nm Zen2 design has been sampling to server customers for several months now, a die like this would be obsolete before it was packaged, and if it existed it's rather likely that AMD would have hinted to it. Even without that, there's no way they'd recoup the R&D costs.

Then, of course, there's changing the CCX design, which in and of itself would not be a small undertaking. Either they'd need to fundamentally redesign the core building block of their entire Zen lineup until now, or they'd need to figure out how to connect three CCXes on a single die. Both of these would be a "big task", even with the inherent modularity of AMD's designs.

 

[newtekie1]

I don't agree that it would be nearly as big of an undertaking as you suggest. They are already working on a CCX design with more cores, we know that, it is slated for 7nm. However, adapting it to 12nm for an accelerated release would not be that difficult. In fact, it has been done many times in the past when processing nodes were not up to the task of meeting production goals. I mean, when 14nm wasn't able to produce enough A9 processors for Apple, they reworked the processor in a matter of weeks to get it into production on 16nm.

So, if they already have the reworked CCX, which we are pretty sure they do, adapting it to 12nm shouldn't be a major undertaking. Plus, now that the only 7nm producer will likely be TSMC, there is a question on if they can keep up with demand. Having the option to fall back on Globalfoundries 12nm for the desktop chips if needed wouldn't necessarily be a bad thing.

 

[mohammed2006]

2019 we will see
16 core AM4
64 core TR4

 

[HTC]

2019
16 core AM4
64 core TR4

I can totally picture it in Zen 2 arch but not in Zen + arch, which is why no 2800X should be released.

 

[dwade]

No thanks. We’re all interested in an Intel 8 cores instead. Fast 8 cores > slower 10 cores. Zero compromises for the target audience which is gamers.

 

[Valantar]

I don't agree that it would be nearly as big of an undertaking as you suggest. They are already working on a CCX design with more cores, we know that, it is slated for 7nm. However, adapting it to 12nm for an accelerated release would not be that difficult. In fact, it has been done many times in the past when processing nodes were not up to the task of meeting production goals. I mean, when 14nm wasn't able to produce enough A9 processors for Apple, they reworked the processor in a matter of weeks to get it into production on 16nm.

So, if they already have the reworked CCX, which we are pretty sure they do, adapting it to 12nm shouldn't be a major undertaking. Plus, now that the only 7nm producer will likely be TSMC, there is a question on if they can keep up with demand. Having the option to fall back on Globalfoundries 12nm for the desktop chips if needed wouldn't necessarily be a bad thing.

Apple sells tens of millions of each model of the iPhone, and has the largest cash reserves of any company on the planet. They could likely afford to hire every engineer available from TSMC to make that work. Also, TSMC 16nm and Samsung 14nm are nowhere near as different as Samsung/GloFo 12nm (refined 14nm) and TSMC 7nm. While the redesign of the A9 was no doubt a reasonably-sized undertaking, this would be quite a lot larger. Also, do you have any sources documenting how they did this "in a matter of weeks"? I'd love to read about how they managed to pull that off. Considering the lead times on phone designs (SoCs enter volume production around half a year before launch), it's likely they had more than a few weeks for this.

There is absolutely a question of whether TSMC can produce enough Zen2 dice for AMD, at least in the short term. This is not an argument against the major undertaking back-porting the design to 12nm would be, though. With the dramatically different density and power attributes of these process nodes, it'd likely require quite a lot of tuning to get right.

 

[HimymCZe]

WTF? We already established that 9900k is no more then 8% stronger then 2700X.
AND you don't even need a decent water cooler to OC 2700X way beyond 9900k performace.
(SOURCE).
THIS is just another nail in Intel coffin.

 

[Slizzo]

Hmm I did not realize that Apple, nvidia, etc lots of big names are in line for 7nm production before AMD... that has to mean Nvidia will have 7nm gpu's next year, I am sure they are throwing their big wallet around the TSMC building to make sure AMD is behind them on the production schedule... this does not bode well for AMD, people are already sick of waiting so many years for competitive GPU's, heh.

He's just throwing shit out there. He doesn't really know that Apple, nVidia, are in line for 7nm before AMD.

As pointed out above, Apple's new iPhones are using TSMC built 7nm A12 chips. They're being delivered to customer hands in a couple days here now, so they likely shipped last month.

 

[nemesis.ie]

@dwade, no, not all of us are interested in an Intel 8 core instead.

 

[dwade]

@dwade, no, not all of us are interested in an Intel 8 core instead.

Most of us then. The world’s first gaming 8 core CPU would look glorious. Intel is smart to release the beast next to Turing. New GPU means more CPU bottleneck for AMD. Smart move is smart.

 

[TheLaughingMan]

This is easily possible. The highest tier Threadripper CCX have a max of 8 cores. So you just need two chips with 3 disabled/defective cores. put two on the die like you would for Threadripper 2950X but with only two chips instead of 4. There you have 10 to 16 cores. There would be some heat concerns mind you with two dies that close together. Plus what would be the point if you are just trying to maintain your multi-threaded crown. You have an entire product skew for that already.

 

[notb]

This is easily possible. The highest tier Threadripper CCX have a max of 8 cores. So you just need two chips with 3 disabled/defective cores. put two on the die like you would for Threadripper 2950X but with only two chips instead of 4. There you have 10 to 16 cores.

No, you don't. Have you ever read anything about how your Ryzen works and looks inside? Aside from benchmarks, obviously. ;-)

All currently available Ryzen and EPYC CPUs are made using the same 4-core CCX.

Were you aware of AM3 processors that supported both DDR2 and DDR3?

Not really a huge feat, let's be honest.

However, probably not all boards will do this. Support varied from board to board; it was up to the board makers to provide BIOS updates for support. I've taken advantage of this feature myself a number of times in the past.

Probably. That makes the whole "upgrade path" argument making even less sense.

It's the same reason we had motherboards like the Asrock 775Dual-VSTA. That was a weird (but useful) board for sure. It had DDR and DDR2 slots, and it had an AGP slot as well as a PCI-E slot.

We seen motherboards with dual-DIMM support fairly recently (Skylake). BTW: anything similar in the AMD camp?

Some of us only want to (or can only afford to) upgrade one component at a time once in a while, so these things really come in handy in such situations.

But, as you said, future AM4 CPU support will depend on mobo. You don't know which one would be updated. Doesn't this make the "upgrade path" argument a bit... poor?

How nice would it be if I could put one of the upcoming Whiskey Lake chips in my old, but still functioning socket 1155 board? I wouldn't need to buy a new motherboard and RAM just to upgrade the CPU.

But it would be a worse CPU, so maybe you wouldn't be tempted to upgrade at all?
Intel's strategy is based around building very precise products. They make only what addresses current demand - hence, sells well. They control waste, they minimize costs. That's how you make money in this business.

WTF? We already established that 9900k is no more then 8% stronger then 2700X.

Weeks before the launch? Man, you should try the lottery.

AND you don't even need a decent water cooler to OC 2700X way beyond 9900k performace.
(SOURCE).

Well... I don't care that much about OC in general, but something tells me you'll be able to OC that 9900K as well.