New Zen 2 Leak

Status
Not open for further replies.
TaintedSquirrel said:
Don't most chips generally do all-core OC that matches their stock single-core boost? At minimum. So the boost clocks here would represent a theoretical max all-core OC.
Click to expand...

Yea 2700x has single core boost of 4.3ghz I think and even though I got mine running stock. My OC does max out around 4.2ghz all core. So I would suspect around 4.7ghz minimum from the 16 all core OC. Given those numbers are legit.
 
TaintedSquirrel said:
Don't most chips generally do all-core OC that matches their stock single-core boost? At minimum. So the boost clocks here would represent a theoretical max all-core OC.
Click to expand...

Generally, yes, the big question here is efficiency. Assuming AMD maintains Zen efficiency going to Zen2, it should be fine. Better, as moving to a smaller node would generally suggest, would be ideal, but as we've found process jumps have increasingly gotten difficult. And this is AMD's first 7nm-at-TSMC CPU project.
 
So we'll know what's up at CES in Jan?

I can def wait to find out if Zen2 is gonna change up the per-core pricing as this video claims....
 
IdiotInCharge said:
Even bigger question: what is the utility of sixteen cores without significantly increased memory bandwidth?

At some point, which is perhaps testable, the workloads that will put sixteen cores to good use will be limited by bandwidth.
Click to expand...

The effect is attrition, we have already seen Intel start to cannibalize the HEDT platform and even eat into entry Xeon levels which is a massive monetary implication given the platforms cost a lot more. If AMD keeps proliferating core count it forces Intel to do the same and basically demolish the higher cost platforms which more expensive = more profit.

To get a high core count mainstream CPU for 400 bucks US that can do high performance video editing on sony vegas pro, count me in.
 
OrangeKhrush said:
The effect is attrition, we have already seen Intel start to cannibalize the HEDT platform and even eat into entry Xeon levels which is a massive monetary implication given the platforms cost a lot more. If AMD keeps proliferating core count it forces Intel to do the same and basically demolish the higher cost platforms which more expensive = more profit.

To get a high core count mainstream CPU for 400 bucks US that can do high performance video editing on sony vegas pro, count me in.
Click to expand...

Why did you quote my post about memory bandwidth and higher core-counts?
 
TaintedSquirrel said:
So AMD is launching 7 CPU's in 4 weeks and the only leak we have is this one? Nothing on SiS or any other CPU bench software? No mobo/chipset leaks?
And despite the other X570 leak showing Computex?

"Take it with a grain of salt" = I'm talking out of my ass for easy clickbait.

I will say if this is accurate my Intel parts are going on Ebay day one. Won't even bother waiting 6+ months for Ice Lake.
Click to expand...

I find it suspicious as well.

AMD's roadmap has Picasso launching in Q1, followed by Matisse in Q2.

AMD don't typically name and prices its products until close to release.
 
TaintedSquirrel said:
Don't most chips generally do all-core OC that matches their stock single-core boost? At minimum. So the boost clocks here would represent a theoretical max all-core OC.
Click to expand...

For non x chips, yeah.
X chips are usually a little below their max boost for all core, at least on reasonable cooling. For example, most 2700x can hit 4.2 but they single core boost to 4.35 (4.3 boost + 50mhz xfr). 2700 non x can usually hit 4.1, which is also their max boost. That's why most recommendations are to use pbo instead of traditional oc, because at lower loads they can hit higher freq than what you can usually do with an all core oc.
 
I got 4.3 out of my 2700X, but it required extreme micromanagement to do. I wound up going back to PBO after - which normally isn't enabled on X370, but is on my specific board for some reason.
 
BoiseTech said:
We all want 16 cores 32 threads with a (even a single core) speed of 5ghz. But it'll never happen. I'd upgrade at least 3 of my machines if this was true.
Click to expand...

No one has made a 7nm part before so let's see what these can do before you sing doom and gloom.
 
DuronBurgerMan said:
I got 4.3 out of my 2700X, but it required extreme micromanagement to do. I wound up going back to PBO after - which normally isn't enabled on X370, but is on my specific board for some reason.
Click to expand...

A good amount of them have it, the higher end asus x370 do at least.
 
BoiseTech said:
We all want 16 cores 32 threads with a (even a single core) speed of 5ghz. But it'll never happen. I'd upgrade at least 3 of my machines if this was true.
Click to expand...

The possibility is exists. Intel got 5 GHz single core boost out of the 9900k. This would be essentially AMD doing the same thing - remember, we're talking two 8-core chiplets, not one 16 core monolithic die. As a result, only one core on one chiplet needs to be able to sustain a 5GHz clock. AMD's PB2 already shows an ability to identify the best performing cores and boost them higher than the others.

TSMC's 7nm process is somewhat better than Intel's 14++ nm process (though not by as much as the numbers might imply - 14++ is REALLY good). So I don't see why this is particularly implausible. If the leak suggested 5GHz BASE clock, yeah, I'd be calling bullshit. But single-core boost? Totally plausible.
 
DuronBurgerMan said:
The possibility is exists. Intel got 5 GHz single core boost out of the 9900k. This would be essentially AMD doing the same thing - remember, we're talking two 8-core chiplets, not one 16 core monolithic die. As a result, only one core on one chiplet needs to be able to sustain a 5GHz clock. AMD's PB2 already shows an ability to identify the best performing cores and boost them higher than the others.

TSMC's 7nm process is somewhat better than Intel's 14++ nm process (though not by as much as the numbers might imply - 14++ is REALLY good). So I don't see why this is particularly implausible. If the leak suggested 5GHz BASE clock, yeah, I'd be calling bullshit. But single-core boost? Totally plausible.
Click to expand...

I'm with you, it could be possible, but it encroaches too much on the Threadripper platform. I think 8/16 will the Ryzen 7's limit, MAYBE 12/24. But 16/32+ is the reason Threadripper exists.

Since Ryzen 3/5/7 all support ECC, the only difference between the two platforms would be the extra PCIe lanes. And thats not enough of a difference to justify a whole product line (or bottom string of that line.)
 
BoiseTech said:
I'm with you, it could be possible, but it encroaches too much on the Threadripper platform. I think 8/16 will the Ryzen 7's limit, MAYBE 12/24. But 16/32+ is the reason Threadripper exists.

Since Ryzen 3/5/7 all support ECC, the only difference between the two platforms would be the extra PCIe lanes. And thats not enough of a difference to justify a whole product line (or bottom string of that line.)
Click to expand...

It's more than that. The supposed 3850X, the 16c/32t AM4 chip marked as a Ryzen 9, will likely require a new X570 motherboard because of 135 watt TDP, has less PCIe lanes, AND only has dual channel memory. It will lose any fair fight against an equivalent Zen 2 16c/32t Threadripper as a result of all this, especially since an AM4 16 core CPU will be memory-starved with only the dual channel memory controller.

Also, I expect TR to advance to a more stable 32c/64t solution than the current 2990X, which has the strange 'good cores/bad cores' paradigm going on now. It will stay quad channel, but will have more equal access than the weird 2990X setup. So next gen TRs would probably focus more on the 24 and 32 core models, with the 16 core model being a base solution with (probably) higher base clocks because of higher 180 watt TDP, similar turbo clocks, and the other platform benefits (quad channel RAM, more lanes) of TR.
 
DuronBurgerMan said:
It's more than that. The supposed 3850X, the 16c/32t AM4 chip marked as a Ryzen 9, will likely require a new X570 motherboard because of 135 watt TDP, has less PCIe lanes, AND only has dual channel memory. It will lose any fair fight against an equivalent Zen 2 16c/32t Threadripper as a result of all this, especially since an AM4 16 core CPU will be memory-starved with only the dual channel memory controller.

Also, I expect TR to advance to a more stable 32c/64t solution than the current 2990X, which has the strange 'good cores/bad cores' paradigm going on now. It will stay quad channel, but will have more equal access than the weird 2990X setup. So next gen TRs would probably focus more on the 24 and 32 core models, with the 16 core model being a base solution with (probably) higher base clocks because of higher 180 watt TDP, similar turbo clocks, and the other platform benefits (quad channel RAM, more lanes) of TR.
Click to expand...

All these points are valid. Lets just say I'm skeptic, and don't get me wrong, I WANT a 16/32 core Ryzen consumer processor. I guess I'll just have to wait and see, if it happens you bet I'll apologize and tell you that you were right. :)
 
I am all about it. Sad though that my threadripper 2950x will be surpassed by a desktop version but its all good. I will just have two 16 core processors in my house. The threadripper is probably still going to perform better in video compression etc.. due to the large cache and quad channel memory performance but that quad channel has Zero impact in anything else and even questionable in video performance.
 
BoiseTech said:
I'm with you, it could be possible, but it encroaches too much on the Threadripper platform. I think 8/16 will the Ryzen 7's limit, MAYBE 12/24. But 16/32+ is the reason Threadripper exists.

Since Ryzen 3/5/7 all support ECC, the only difference between the two platforms would be the extra PCIe lanes. And thats not enough of a difference to justify a whole product line (or bottom string of that line.)
Click to expand...
given the move to the io chip whats stopping them from doing a 64/128 TR chip? it alleviates the issue they ran into with the 2990wx so who knows with the chiplet design there's a ton of possibilities and core layouts available to them for TR.
 
Don't be surprised if we don't see consumer 7nm AMD Ryzen 3000 CPU's and X570 motherboards until Q3 2019. AMD has already made it clear that the first 7nm batch is not for consumers. CES 2019 would be too soon for such a release. Computex in June 2019 at the earliest.

AMD has been quite vague about a specific release date for consumer release, and based on AMD's release history I'm remembering the Vega GPU that was announced at CES 2017 but didn't get released until Q3 2017 - like 9 months late.

AMD has been clear that the first round of 7nm AMD CPU's will *NOT* be for consumers. CES seems to early for such a release - I hope I'm wrong but again, based on AMD's release history it could be announced at CES but still not be released for 9 months.

And we won't see PCIe 4.0 on consumer boards until 2020. Btw, PCIe 4.0 will be for consumers while PCIe 5.0 will not be for consumers. PCIe 5.0 will only be for AI and science, medical research etc.

; )
 
sirmonkey1985 said:
given the move to the io chip whats stopping them from doing a 64/128 TR chip? it alleviates the issue they ran into with the 2990wx so who knows with the chiplet design there's a ton of possibilities and core layouts available to them for TR.
Click to expand...

I think physical CPU size is the first issue :) That 32 core TR 2990wx was pretty full.
 
JoseJones said:
Don't be surprised if we don't see consumer 7nm AMD Ryzen 3000 CPU's and X570 motherboards until Q3 2019. AMD has already made it clear that the first 7nm batch is not for consumers. CES 2019 would be too soon for such a release. Computex in June 2019 at the earliest.

AMD has been quite vague about a specific release date for consumer release, and based on AMD's release history I'm remembering the Vega GPU that was announced at CES 2017 but didn't get released until Q3 2017 - like 9 months late.

AMD has been clear that the first round of 7nm AMD CPU's will *NOT* be for consumers. CES seems to early for such a release - I hope I'm wrong but again, based on AMD's release history it could be announced at CES but still not be released for 9 months.

And we won't see PCIe 4.0 on consumer boards until 2020. Btw, PCIe 4.0 will be for consumers while PCIe 5.0 will not be for consumers. PCIe 5.0 will only be for AI and science, medical research etc.

; )
Click to expand...

Yea for vega which was led by RTG and Raja. AMD has never let down on CPU launches. You can't compare the two. They have announced and released on time. This is a chiplet design. You can not be more wrong about ryzen launches. I don't see anywhere where they announced ryzen at CES and with availability 9 months later. Over exaggerating much? lol. Ryzen has always been announced and launched at the time they promised. Not sure where you are getting those 9 months. Their CPU side has been spot on for every ryzen launch. So why be so surprised if they announce zen 2 at ces even for consumer they did it for zen+ no? Wich availability like in March and april. If 7nm is going as we are hearing and they have had multiple products ahead of schedule I won't be surprised a bit if they announce 3000 series at CES or at least some parts.

On the flip side they might not, who knows. But all I am saying is you can't count it out because they have done it before.
 
BoiseTech said:
All these points are valid. Lets just say I'm skeptic, and don't get me wrong, I WANT a 16/32 core Ryzen consumer processor. I guess I'll just have to wait and see, if it happens you bet I'll apologize and tell you that you were right. :)
Click to expand...

Oh I don't think I'm RIGHT so much as I think it's plausible. Might sound like a CYA, but bear in mind, this is a leak. A rumor. Yes, yes, one that comes from a less-than-disreputable source. Yes, one that has an air of plausibility. But still a rumor/leak.

It's a crapshoot. It's great discussion, but none of us are going to know shit until AMD either issues a statement on it, or releases the damned thing.
 
BoiseTech said:
I think physical CPU size is the first issue :) That 32 core TR 2990wx was pretty full.
Click to expand...

The Naples socket is more or less the same as TR4, size-wise, and Rome will be a drop-in replacement with BIOS update. If it fits on Epyc, it can PHYSICALLY fit on TR4. Now, TR4 only supports quad channel memory. And there are, presumably, other limitations when you run 64 cores like that at higher TR-like clockspeeds, instead of Epyc-like clocks. So I don't think a 64 core TR is terribly likely. And even if it did happen, it would suffer from memory bandwidth problems, I suspect. So not sure AMD would bother. But it will fit in terms of physical dimensions, at least.
 
tangoseal said:
I am all about it. Sad though that my threadripper 2950x will be surpassed by a desktop version but its all good. I will just have two 16 core processors in my house. The threadripper is probably still going to perform better in video compression etc.. due to the large cache and quad channel memory performance but that quad channel has Zero impact in anything else and even questionable in video performance.
Click to expand...

You may be able to drop in a 24 or 32 core TR that isn't hobbled like the 2990x is. Might.
 
IdiotInCharge said:
Even bigger question: what is the utility of sixteen cores without significantly increased memory bandwidth?

At some point, which is perhaps testable, the workloads that will put sixteen cores to good use will be limited by bandwidth.
Click to expand...
That's what the main segmentation between HEDT and Enthusiast or Desktop use will be going forward, that said I'm not counting DDR5. As it is, if you want 16 well fed cores, get Threadripper. If you want 32+ well fed cores, get Epyc.
 
With TSMC getting 7nm orders reduced from the phone manufacturers, perhaps there will be some additional foundry capacity to move CPU products forward? Or is die manufacturing so long lead that you can't drop in another product?
 
Randall Stephens said:
With TSMC getting 7nm orders reduced from the phone manufacturers, perhaps there will be some additional foundry capacity to move CPU products forward? Or is die manufacturing so long lead that you can't drop in another product?
Click to expand...

They can probably pull it off, if they're in production. At certain points it would be economical to do so; in others, not so much, I'd imagine. They do want to avoid the situation they and Nvidia are facing with GPU stock if they can.
 
JoseJones said:
And we won't see PCIe 4.0 on consumer boards until 2020. Btw, PCIe 4.0 will be for consumers while PCIe 5.0 will not be for consumers. PCIe 5.0 will only be for AI and science, medical research etc.

; )
Click to expand...
PCI-E 3.0 will be saturated within the next 2-3 years. In fact it will probably happen before 5 GHz 12~16C gets saturated in games. I will not be going anywhere near any new CPU/Mobo that don't have 4.0. I even hesitated buying my current board last year because of it.
 
TaintedSquirrel said:
PCI-E 3.0 will be saturated within the next 2-3 years. In fact it will probably happen before 5 GHz 12~16C gets saturated in games. I will not be going anywhere near any new CPU/Mobo that don't have 4.0. I even hesitated buying my current board last year because of it.
Click to expand...
I bet you're at least 6 years away from seeing any tangible FPS difference between pcie 3 and 4. Hell, you barely see a difference between 8x and 16x pci-e 3.0 right now.
 
OrangeKhrush said:
Ryzen 9 3850X, shurup and take my money.
Click to expand...

I for one will probably settle for the 12 cores R7 this time around, and even this is a massive overkill for my needs (gaming and possibly some light video editing and programming). Better for me to invest more in a video card, so let's hope AMD will make video cards great again, too.
 
wyqtor said:
I for one will probably settle for the 12 cores R7 this time around, and even this is a massive overkill for my needs (gaming and possibly some light video editing and programming). Better for me to invest more in a video card, so let's hope AMD will make video cards great again, too.
Click to expand...

No kidding. I'm wondering if I need more than 8 cores 16 threads.
 
kirbyrj said:
No kidding. I'm wondering if I need more than 8 cores 16 threads.
Click to expand...
I think you're overlooking two key considerations:
1. Upgrade interval -- how long will it be more than you want/need? Maybe not for as long as you think
2. Use case. Everyone on PC forums seems to consider games their #1 driver, but outside of games more cores is incredibly useful, even for multitasking if you don't use hardcore multithreaded applications.
 
compgeek89 said:
I think you're overlooking two key considerations:
1. Upgrade interval -- how long will it be more than you want/need? Maybe not for as long as you think
2. Use case. Everyone on PC forums seems to consider games their #1 driver, but outside of games more cores is incredibly useful, even for multitasking if you don't use hardcore multithreaded applications.
Click to expand...

1. I upgrade whenever I need something new to play with. I think in the last two years I've had a X5660, 3930k, 5930k, 5960x, G4560, 6700k, 7700k, 8400, 8700k, 7820x, 8320e, 8350, 1700, 1600x, 2200g, 2700x, 2600, and 2700, so I don't mind swapping in something different when the mood strikes me ;).

2. I agree. I'm just wondering about diminishing returns over 8C/16T which is probably why I haven't tried a Threadripper (yet). But I am interested to see eventual benchmarks (as we all are).
 
kirbyrj said:
1. I upgrade whenever I need something new to play with. I think in the last two years I've had a X5660, 3930k, G4560, 6700k, 7700k, 8400, 8700k, 1700, 1600x, 2200g, 2700x, 2600, and 2700, so I don't mind swapping in something different when the mood strikes me ;).

2. I agree. I'm just wondering about diminishing returns over 8C/16T.
Click to expand...
Considering my last PC lasted 8 full years, I'm iin a different boat these days :D
 
"AMD's next-gen Zen CPU due in 2016"
pcgamer.com/amds-next-gen-zen-cpu-due-in-2016/

"AMD Zen Launching In January At CES 2017 – High-End X370 AM4 Motherboards Shipping In December"
wccftech.com/amd-zen-launching-ces-january-2017/

"AMD's high-end X370 chipset teased, arrives in Feb 2017
tweaktown.com/news/53862/amds-high-end-x370-chipset-teased-arrives-feb-2017/index.html

"AMD gave us a sneak peek at its new Vega graphics card architecture way back at CES 2017. What we didn’t know then, though, is that it wouldn’t end up in the hands of consumers until August 2017"
https://www.techradar.com/news/amd-vega-release-date-news-and-features-everything-you-need-to-know

Never happened did it. CES is in January, August is 7 months later, the wait for Vega was an embarrassment for AMD

Zen CPU release date: March 2, 2017
Vega release date: August 14, 2017
Ryzen+ release date: April 19, 2018
Zen2 ...

I hope all the AMD 7nm CPU's & GPU's are released at CES but, based on AMD's terrible history of releases I won't hold my breathe. Nowhere has AMD actually claimed anything would actually be released at CES next month. In fact, AMD has been quite vague so, until AMD makes an official statement I'm betting they'll be late as usual - I realize AMD fans don't want to hear that.




NKD said:
Yea for vega which was led by RTG and Raja. AMD has never let down on CPU launches. You can't compare the two. They have announced and released on time. This is a chiplet design. You can not be more wrong about ryzen launches. I don't see anywhere where they announced ryzen at CES and with availability 9 months later. Over exaggerating much? lol. Ryzen has always been announced and launched at the time they promised. Not sure where you are getting those 9 months. Their CPU side has been spot on for every ryzen launch. So why be so surprised if they announce zen 2 at ces even for consumer they did it for zen+ no? Wich availability like in March and april. If 7nm is going as we are hearing and they have had multiple products ahead of schedule I won't be surprised a bit if they announce 3000 series at CES or at least some parts.

On the flip side they might not, who knows. But all I am saying is you can't count it out because they have done it before.
Click to expand...
 
Status
Not open for further replies.
Back
Top