AMD Previews 7nm Ryzen CPUs at CES

Pieter3dnow

Supreme [H]ardness
Joined
Jul 29, 2009
Messages
6,785
It's juanrga being juanrga.

No because Juanrga would have 5 different videos saying 5 different things and then whenever the dust settles picks and chooses the right ones.
AdoredTV does not care about being right he cares about his sources giving him enough information. He can figure out himself what is in the the realm of possibility rather then marketing propaganda ...

He even posts or makes videos about stuff he got wrong, come on you ever seen Juanrga do that ?
 

Pieter3dnow

Supreme [H]ardness
Joined
Jul 29, 2009
Messages
6,785
AdoredTV said launch. And WCCFTECH said launch. And several sites above the hype train said launch

https://gadgets.ndtv.com/laptops/ne...ces-2019-launch-rumours-16-core-9-cpu-1972085

It clearly stated TBA which is not launched that media have their own agenda in not being careful about material used is obvious.
Specs-pcgh.png


Why juanrga do you state things which are so easy to disprove within minutes ?
 

juanrga

Pro-Intel / Anti-AMD Just FYI
Joined
Feb 22, 2017
Messages
2,733
If you watch his videos he had two thoughts on what it could possibly be, and his hope was for a chiplet based design with an IO die. His original leak/predictions in early December are remarkably similar to exactly how the ryzen 2 die shot looks.

As far as why only one chiplet is being used? Thats actually quite easy to explain. These are 8-core chiplets. It was speculated that the 8-core Ryzen cpu's without integrated graphics would be using 2 4-core chiplets rather than 1 8-core chiplet

It's entirely possible that instead of using 2 dies for a 8 core cpu. Their yeilds on Zen 2 are so good that they can afford to just use a single 8-core chiplet rather than 2 4-core ones and save the 4-core defective chiplets to lower end parts.

There is clearly space on the die for a second CPU chiplet or a GPU chiplet. I would expect for us to see chips in the near future that would use it. Whether those are going to be part of the Ryzen 3000 series is yet to be seen.

His original December 'predictions' and 'leaks' were a literal copy of leaks and predictions made by other ppl as Charlie Demerjian and the Chip Collective, ppl as K. H. Chia draw in early November (before any AdoredTV videos) detailed technical figures about what was coming for Zen2

Dr3OZYcVsAAEcyK.jpg


So I am not talking about videos where AdoredTV copies correct stuff got by other ppl. I am talking about AdoredTV last 'predictions' and 'leaks', when he invented that desktop was different than server. When he said us that desktop would be a single die, not chiplets as in the above diagram.

AdoredTV put out a video specifically saying ANNOUNCED, not launch. He called this out, specifically. I still think it wasn't even much of an announcement, though. More like a teaser. So he's kinda-sorta wrong on that.

It clearly stated TBA which is not launched that media have their own agenda in not being careful about material used is obvious.
View attachment 134138

The problem is that he says one things and the contrary. Let us ignore he said launch and he only said announced. He continues being wrong. No one of the chips in that table has been announced. No one!
 

Pieter3dnow

Supreme [H]ardness
Joined
Jul 29, 2009
Messages
6,785
Indeed it was a blanket statement from AMD of launching mid 2019 :) and no models were announced ...
 

ole-m

Limp Gawd
Joined
Oct 5, 2015
Messages
452
Indeed it was a blanket statement from AMD of launching mid 2019 :) and no models were announced ...

They also said Epyc is launching mid2019.
That means they will have shit tons of chips made to bin for epyc (in high demand too!)
Gamers Nexus made a video about PCI-E 4.0, X570 and said that most of the issues are around the platform and not the chips.

Sounds very plausible when you consider epyc and ryzen launch at same time, Good working chips exists.
 

juanrga

Pro-Intel / Anti-AMD Just FYI
Joined
Feb 22, 2017
Messages
2,733
You know it was a good product teaser when all juanrga can do is nitpick about AdoredTV videos afterward.

LOL. I simply replied to the post #29 where Pieter rewrites history about AdoredTV 'predictions'.

The CES keynote didn't add any new relevant information about 7nm Ryzen that wasn't predicted or leaked before by other people. CB demo performance? Expected. Chiplet design? Expected. Mid-2019 launch? Expected. PCIe4? Expected...

Additional comments in this thread would be needed only if this 7nm Ryzen preview had invalidated what we knew or guessed about 7nm Ryzen since October past year.
 
Last edited:

mikeo

Gawd
Joined
May 17, 2006
Messages
622
I'll probably go with a 16 or 32 core one when they come out assuming the single core performance finally beats out a single core on a 3930k, most likely TR platform for extra pcie lanes.

With all the hate on intel, I think its pretty impressive that they have become so efficient at 14nm revisions that it has essentially matched the performance of a 7nm chip....
 

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211
I am pretty excited by this. It may just wind up being my first CPU upgrade in my workstation in 8 years.


A few thoughts:

- It roughly ties (marginally beats) the 9900K in all core runs of Cinebench. That is promising, but I am more concerned with max turbo clocks when fewer cores are loaded. Can it keep up with the 9900K in loads with fewer cores? Thus far Ryzen has had a significant clock speed gap compared to Intel. I want to know if they have bridged it.

- Will be interesting to see where the final clocks wind up, when the process is allowed to mature for 6 months, and microcode is optimized. I wonder if it will stay pretty much the same as in this benchmark, or if it will increase. I doubt it will fall. AMD wouldn't do themselves that kind of disservice and cause themselves that kind of embarrassment. I bet they went with a clock for this demo they feel pretty confident they can hit at launch

- Backwards compatible with AM4 is nice, but I wonder if there will be any new chipsets. I'd imagine there would be some opportunity involved with the new PCIe gen 4 lanes. In order to keep backwards compatibility they would have had to keep the same number of lanes for the chipset, but now they are gen 4 instead of gen 3, which means the chipset has double the bandwidth. Hopefully we will see more PCIe lanes off of the chipset available, if a new chipset is launched. Relatively few PCIe lanes was my biggest concern with previous Ryzens. I certainly hope this will be the case. If I get enough lanes, I won't have to buy a threadripper. I dont want to spend on extra Threadripper cores I don't need just to get a few more PCIe lanes.

- As others, I'm a little disappointed it is so far off, but I am looking forward to learning more details.
 

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211
I'll probably go with a 16 or 32 core one when they come out assuming the single core performance finally beats out a single core on a 3930k, most likely TR platform for extra pcie lanes.

With all the hate on intel, I think its pretty impressive that they have become so efficient at 14nm revisions that it has essentially matched the performance of a 7nm chip....

What clock do you have your 3930k at? Mines at 4.8Ghz, which gives me a single core score of 163cb in Cinebench R15

This roughly ties a first gen Ryzen 1800x at stock clocks. (presumably it beats my 3930k when overclocked)

Second gen Ryzen 2700x beats my highly overclocked 3930k even at stock clocks.
 

mikeo

Gawd
Joined
May 17, 2006
Messages
622
What clock do you have your 3930k at? Mines at 4.8Ghz, which gives me a single core score of 163cb in Cinebench R15

This roughly ties a first gen Ryzen 1800x at stock clocks. (presumably it beats my 3930k when overclocked)

Second gen Ryzen 2700x beats my highly overclocked 3930k even at stock clocks.

Same as yours 24/7 with CnQ. 2700x beats it in cinebench, but not by enough to make it worth upgrading for me. If you look at other single core benchmarks like superpi 1m, the 3930k still crushes these, probably partially due to memory latency.
 

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211
Same as yours 24/7 with CnQ. 2700x beats it in cinebench, but not by enough to make it worth upgrading for me. If you look at other single core benchmarks like superpi 1m, the 3930k still crushes these, probably partially due to memory latency.

Yeah, the 3930k has lower memory latency and quad channel bandwidth which may contribute.

That said, I don't pay much attention to synthetic benchmarks like SuperPi. I find they generally don't translate into real world performance very well.
 

kamikazi

Gawd
Joined
Jan 19, 2006
Messages
741
I'll probably go with a 16 or 32 core one when they come out assuming the single core performance finally beats out a single core on a 3930k, most likely TR platform for extra pcie lanes.

With all the hate on intel, I think its pretty impressive that they have become so efficient at 14nm revisions that it has essentially matched the performance of a 7nm chip....

AMD 7nm is considered to be roughly equivalent in size to Intel 10nm. One part of the transistor or something like that measures 7nm, the rest is the same as Intel 10nm. The names on process nodes anymore are basically only guidelines and marketing.
 
  • Like
Reactions: mikeo
like this

Mega6

2[H]4U
Joined
Aug 13, 2017
Messages
3,034
AMD 7nm is considered to be roughly equivalent in size to Intel 10nm. One part of the transistor or something like that measures 7nm, the rest is the same as Intel 10nm. The names on process nodes anymore are basically only guidelines and marketing.
It's TSMC's process but yeah.
 

psyclist

Gawd
Joined
Jan 25, 2005
Messages
844
What clock do you have your 3930k at? Mines at 4.8Ghz, which gives me a single core score of 163cb in Cinebench R15

This roughly ties a first gen Ryzen 1800x at stock clocks. (presumably it beats my 3930k when overclocked)

Second gen Ryzen 2700x beats my highly overclocked 3930k even at stock clocks.

My 1660 v2 (4960X) does 175/1316 in CB R15, X79 lives on for another year or so for me. Trying to hold off till DDR5...but the reviews of the 3000 series could sway that.
 

juanrga

Pro-Intel / Anti-AMD Just FYI
Joined
Feb 22, 2017
Messages
2,733
Rumor. The 7nm Ryzen used for the Cinebench demo was a Ryzen 5 qualification sample with the next specs: 8C/16T 65W at 3.7GHz base clock and 4.5GHz boost.
 

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211
AMD 7nm is considered to be roughly equivalent in size to Intel 10nm. One part of the transistor or something like that measures 7nm, the rest is the same as Intel 10nm. The names on process nodes anymore are basically only guidelines and marketing.

It's TSMC's process but yeah.


Yeah,

The ITRS standard actually governs process node sizes and is supåosed to ensure the same name is used for the same size node.

Intel used to actually track the ITRS standard fairöy closely until a few years back. These days no one does. Their marketing people just fabricate numbers based on what they think sounds best. It's sad.

Marketing and sales people should never be allowed to make any decisions.
 

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211

I wonder if dual channel memory bandwidth will start to cause diminishing returns to severely set in and limit the effectiveness of more than 8 cores on AM4.

I'd argue that for more than 8 cores, it's probably wiser to go Threadripper.
 
Last edited:
  • Like
Reactions: mikeo
like this

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211
Rumor. The 7nm Ryzen used for the Cinebench demo was a Ryzen 5 qualification sample with the next specs: 8C/16T 65W at 3.7GHz base clock and 4.5GHz boost.

My guess would have been that this far out they don't have final model names or clocks as of yet and that as the result of this testing they slap names and clocks on them at the very end based on what the process allows them to get away with
 

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211
I feel like a huge thing people are missing that the powers listed are total system powers (180W vs 130W). According to Anandtech the chip power for the i9 is 125W but the AMD CPU is at 75W...

Same performance with numbers of core/threads at about HALF the power means some pretty serious increases in IPC or frequency.

...or much smaller node size. We know the node size numbers are mostly fudged these days, but still TSMC's 7nm is much smaller than Intel's 14nm
 

Nathan_P

[H]ard DCOTM x2
Joined
Mar 2, 2010
Messages
3,433
My guess is that AM4 will top out at 12 cores and anything above that will go to Threadripper.

Also Gamers nexus have posted a video in which they state that the biggest hold up at the moment is the new x570 chipset - not the cpu's.
 

Olle P

Limp Gawd
Joined
Mar 29, 2010
Messages
331
If you extrapolate 2700x performance and use some common sense, ANYONE could have predicted these performance results with cinebench that was shown today.
Prediction is difficult when you don't know how far to push the frequency and IPC gain.

Back in my day we had 1 core, 1 thread and cooled everything with 8 fins of aluminium without these fancy "fans"...
Back in my day the CPU (IBM 486SLC2) didn't even need a heatsink...

These are 8-core chiplets. It was speculated that the 8-core Ryzen cpu's without integrated graphics would be using 2 4-core chiplets rather than 1 8-core chiplet

It's entirely possible that instead of using 2 dies for a 8 core cpu. Their yeilds on Zen 2 are so good that they can afford to just use a single 8-core chiplet rather than 2 4-core ones and save the 4-core defective chiplets to lower end parts.
4-core chiplet should be read as 8-core with (at least) four cores disabled.
Only one type of chiplet is printed, and then it's all a matter of binning. There should be plenty with eight functioning cores but not reaching Epyc demands for voltage/speed/stability.

We've already heard the x570 chipset will be released around may so the 3000 series at that time makes sense.
Steve at Hardware Unboxed claims he's already been approached by a motherboard manufacturer about testing an X570 motherboard. That suggests the release might be earlier than may.
My hope is that Ryzen 3000 will be available the same time of year as the previous generations, by late march to early april.
 
  • Like
Reactions: mikeo
like this

Nobu

Supreme [H]ardness
Joined
Jun 7, 2007
Messages
4,882
I am pretty excited by this. It may just wind up being my first CPU upgrade in my workstation in 8 years.


A few thoughts:

- It roughly ties (marginally beats) the 9900K in all core runs of Cinebench. That is promising, but I am more concerned with max turbo clocks when fewer cores are loaded. Can it keep up with the 9900K in loads with fewer cores? Thus far Ryzen has had a significant clock speed gap compared to Intel. I want to know if they have bridged it.

- Will be interesting to see where the final clocks wind up, when the process is allowed to mature for 6 months, and microcode is optimized. I wonder if it will stay pretty much the same as in this benchmark, or if it will increase. I doubt it will fall. AMD wouldn't do themselves that kind of disservice and cause themselves that kind of embarrassment. I bet they went with a clock for this demo they feel pretty confident they can hit at launch

- Backwards compatible with AM4 is nice, but I wonder if there will be any new chipsets. I'd imagine there would be some opportunity involved with the new PCIe gen 4 lanes. In order to keep backwards compatibility they would have had to keep the same number of lanes for the chipset, but now they are gen 4 instead of gen 3, which means the chipset has double the bandwidth. Hopefully we will see more PCIe lanes off of the chipset available, if a new chipset is launched. Relatively few PCIe lanes was my biggest concern with previous Ryzens. I certainly hope this will be the case. If I get enough lanes, I won't have to buy a threadripper. I dont want to spend on extra Threadripper cores I don't need just to get a few more PCIe lanes.

- As others, I'm a little disappointed it is so far off, but I am looking forward to learning more details.
Someone at AMD confirmed there would need to be new chipsets for pcie gen4, but also that some (current) motherboards would support gen4 speeds in the first pcie x16 slot (the one closest to the socket. It wouldn't work for the other slots because the traces are too long and they need new drivers (or something like that, paraphrasing here).

I'm gunning for a 6c replacement for my 1600, but I may go 8c the 6c processor clocks lower. Also wanna get radeon vii, but I'll probably wait for navi. I wonder if the dual-chiplet processors will have/allow two dual-channel banks (so, four or eight dimms) of ram or if they'll be business as usual...
 

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211
Someone at AMD confirmed there would need to be new chipsets for pcie gen4, but also that some (current) motherboards would support gen4 speeds in the first pcie x16 slot (the one closest to the socket. It wouldn't work for the other slots because the traces are too long and they need new drivers (or something like that, paraphrasing here).

I'm gunning for a 6c replacement for my 1600, but I may go 8c the 6c processor clocks lower. Also wanna get radeon vii, but I'll probably wait for navi. I wonder if the dual-chiplet processors will have/allow two dual-channel banks (so, four or eight dimms) of ram or if they'll be business as usual...

Yeah, I assumed the 16x slot would just work, as the PCIe Gen 4 interconnects seem electrically pretty similar to gen 3, and that slot usually connects straight to the CPU via the socket.

Don't most other slots on AM4 boards go via the chipset though?

If its just a matter of the chipset splitting up the bandwidth it has going to the CPU, then if the chipset is updated to support gen 4 it should suddenly have double the bandwidth.

I - for one - hope that there will at least be an option of a board that utilizes that bandwidth and splits it up providing as many gen 2 slots that will fit in that bandwidth allocation, rather than providing fewer gen 3 or very few gen 4 slots.
 

Nobu

Supreme [H]ardness
Joined
Jun 7, 2007
Messages
4,882
Yeah, I assumed the 16x slot would just work, as the PCIe Gen 4 interconnects seem electrically pretty similar to gen 3, and that slot usually connects straight to the CPU via the socket.

Don't most other slots on AM4 boards go via the chipset though?

If its just a matter of the chipset splitting up the bandwidth it has going to the CPU, then if the chipset is updated to support gen 4 it should suddenly have double the bandwidth.

I - for one - hope that there will at least be an option of a board that utilizes that bandwidth and splits it up providing as many gen 2 slots that will fit in that bandwidth allocation, rather than providing fewer gen 3 or very few gen 4 slots.
I'm sure there will be. Hopefully they're good quality boards and firmware. I don't want another gigabyte x370 gaming k5...
 

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211
I'm sure there will be. Hopefully they're good quality boards and firmware. I don't want another gigabyte x370 gaming k5...


I'm not familiar with that board. Thus far the only AM4 boards I've used have been the MSI B350 Tomahawk for my stepsons 1600x gaming build and the Asrock X470 Gaming-ITX/ac I used for my Fiance's 2400G build.

It had lots of problems?

I'd imagine that motherboard makers will take their AMD boards more seriously now that AMD is gaining market share.
 

Nobu

Supreme [H]ardness
Joined
Jun 7, 2007
Messages
4,882
I'm not familiar with that board. Thus far the only AM4 boards I've used have been the MSI B350 Tomahawk for my stepsons 1600x gaming build and the Asrock X470 Gaming-ITX/ac I used for my Fiance's 2400G build.

It had lots of problems?

I'd imagine that motherboard makers will take their AMD boards more seriously now that AMD is gaining market share.
It doesn't have llc, absolute power control (only offset) and many other overclocking features that would help with bclk overclocking (which is the primary reason to get a 'k' board), but also has a buggy bios and poor memory support. If you corrupt either backup or main bios, it's done (you're down to just one), because there's no way to flash from one to the other (not even a switch). Probably others I'm forgetting.

Edit: I mean, it's a good board if you don't overclock the cpu or ram, but...that's not what I was hoping for when I got it. Unfortunately, I got it when it was fairly new, and kept hoping future updates would resolve my problems (at least some). Memory support improved...a little. But that's about it. They enabled zen common options eventually.
 
Last edited:

DuronBurgerMan

[H]ard|Gawd
Joined
Mar 13, 2017
Messages
1,340
Let us ignore he said launch and he only said announced. He continues being wrong. No one of the chips in that table has been announced. No one!

Yes, on this part, he is wrong with "announced" (I also said so in the very post you quoted - so I don't know what your beef is?).

This wasn't a launch. It wasn't even an announcement. It was a teaser.

However, AdoredTV said announced, specifically (and even issued a video to further clarify this), not launched as you keep claiming.
 

DuronBurgerMan

[H]ard|Gawd
Joined
Mar 13, 2017
Messages
1,340
Heres a better shot of the CPU with the traces clearer, found on RedditView attachment 134155

I still wonder why AMD chose to demo the 8 core product, but still obviously tease us with an almost sure to come higher core count version.

I wonder if they were just have problems with their engineering samples with two core chiplets, or something, and stuck with the single core chiplet version so their demo didn't crash or anything.
 

Zarathustra[H]

Fully [H]
Joined
Oct 29, 2000
Messages
31,211
I still wonder why AMD chose to demo the 8 core product, but still obviously tease us with an almost sure to come higher core count version.

I wonder if they were just have problems with their engineering samples with two core chiplets, or something, and stuck with the single core chiplet version so their demo didn't crash or anything.

They probably wanted to do a like for like comparison with the 9900k
 

mikeo

Gawd
Joined
May 17, 2006
Messages
622
Maybe they want to keep the option of having the 16 core stay on the TR platform. There is no reason for them to go with the rumored pricing when the 9900k is five hundos.
 

dgz

Supreme [H]ardness
Joined
Feb 15, 2010
Messages
5,838
Or, they could use that space for a nice GPU
 
Top