Navi/Ryzen 3000 launch Megathread

[leadeater]

3 minutes ago, LukeSavenije said:

@S w a t s o n @leadeater from what i can see is the 140w lock just for stock clocks, so you can easily go past it if you want to oc it

I know, and GN did OC and still didn't go past 4.3GHz all core.

[cj09beira]

1 minute ago, RobFRaschke said:

4.6 on a single core requires MUCH less power than overclocking 8 or 12 cores to 4.3.

the thing is the voltage is too high, you need more volts to stabilize a certain clock speed, if the cpus need 1.4+ for 4.4ghz that means that there little to no chance of getting much over that even on single cores as you would still need even higher voltages to stabilize the clocks, unless the problem is too much resistance on the power lines in the silicon itself, that could make it so that you could reach those higher clocks on the same voltage due to less voltage drop when using less cores

 

[S w a t s o n]

3 minutes ago, leadeater said:

Well we can apply some simple logic to this. You can, like almost always disable or increase power limits, this is a stock package limit not socket or platform because Zen+ as shown draws more and the 3950X will draw more than current released Zen2 products.

 

But the real kicker is if it were possible on a 240/280 CLC why is there an absence of any reviews showing it? If it were possible it would have been done, GN went through the effort of overclocking the 3900X and specifically stated that is the maximum possible and products are coming out stock closer to the limit than before. If all signs point to what has been shown as what you can achieve on ambient then there is no reason to believe it is not the case. I know GN review was on the 3600 but there was enough information shown and enough explanation in that video to conclude even the best (pre tested by AMD) chips maxing out at 4.3GHz is what you will get.

Sure, I'm not arguing that over 4.3-4.4 is the limit for AM4 right now. I'm saying it's a power delivery limit or at worst a heat density limit. Which is why I want to see TR4. We know the silicon can do 4.7GHz at ambient cooling albeit on a single core. Silicon lottery could be playing into but I'm not so sure with yields apparently being over 85% now.

There's no reason these chips cant do XFR clocks on all cores if the constraints are removed.

[LukeSavenije]

Just now, leadeater said:

I know, and GN did OC and still didn't go past 4.3GHz all core.

yeah... probably a bios error, i've seen some others on x470 and b450 too already

[cj09beira]

1 minute ago, leadeater said:

I know, and GN did OC and still didn't go past 4.3GHz all core.

better still derbauer tested it and said 4.4 all core was the best he could get 

[dfsdfgfkjsefoiqzemnd]

It's been a couple of years since I benched my 5930K.  Time to download R20 and see just how bad my CPU is ...

 

EDIT : Ouch, 2750 at 4.2GHz.  That's little more than a third of what the 3900X does.  (7168 according to Kyle)

[leadeater]

Just now, S w a t s o n said:

Sure, I'm not arguing that over 4.3-4.4 is the limit for AM4 right now. I'm saying it's a power delivery limit or at worst a heat density limit. Which is why I want to see TR4. We know the silicon can do 4.7GHz at ambient cooling albeit on a single core. Silicon lottery could be playing into but I'm not so sure with yields apparently being over 85% now

So what you're saying is that GN overclocked but didn't bother to disable or increase package power limits, or that motherboard with VRMs that are capable of 500W+ don't offer this ability? Yea very unlikely.

[S w a t s o n]

3 minutes ago, leadeater said:

So what you're saying is that GN overclocked but didn't bother to disable or increase package power limits, or that motherboard with VRMs that are capable of 500W+ don't offer this ability? Yea very unlikely.

I would like to see more information, nothing here makes sense. If they did disable the power limit and got an accurate reading of 170W that doesnt mean the VRM's that can do 500W+ make any more sense than at 140W.

[LukeSavenije]

3 minutes ago, leadeater said:

So what you're saying is that GN overclocked but didn't bother to disable or increase package power limits, or that motherboard with VRMs that are capable of 500W+ don't offer this ability? Yea very unlikely.

500w? technically the boards go up to almost 800w, but that would require some serious cooling

[Delicieuxz]

8 minutes ago, RobFRaschke said:

Looking only at the CPU cost itself is the Upgrade metric. The rest of the numbers are based on a full system build at that cost, with the CPU being only a decreasing portion of the cost of the build.

So, what is the 133% "Upgrade" metric 133% of? From the R7 3700X at $329 to the i7 9700K at $399, there's a 21% increase in cost, so the 9700K is 121% the price of the R7 3700X. 

[S w a t s o n]

8 minutes ago, leadeater said:

So what you're saying is that GN overclocked but didn't bother to disable or increase package power limits

Also, Anandtech was unable to?

Quote

Unfortunately we weren’t able to further investigate raising the PPT limit for this article due to time contraints as well as currently non-final firmware version for X570 motherboards from the vendors.

 

[D13H4RD]

Reading through the reviews, definitely looks like 3rd-gen Ryzen is a win.

 

No, it didn't absolutely demolish Intel in everything, but it did make significant gains all while also being quite efficient and also closing the gap in its weakest area.

[cj09beira]

Just now, S w a t s o n said:

Also, Anandtech was unable to?

then the question becomes why did it need so much voltage to reach even 4.3-4, 1.4v on 7nm seems pretty dam high to me, oc3d mentioned 1.5v, then what are the voltages spiking to when its doing single core boosts? 1.6??

[Lathlaer]

@Delicieuxz yeah I can't seem to decipher those charts as well. It seems like poorly chosen medium to give the information. It's obvious that the more your system costs the less % of the price is the CPU upgrade alone but when you post a pie chart with yellow and blue, it seems prudent to show what is yellow and what is blue.

[S w a t s o n]

5 minutes ago, cj09beira said:

then the question becomes why did it need so much voltage to reach even 4.3-4, 1.4v on 7nm seems pretty dam high to me, oc3d mentioned 1.5v, then what are the voltages spiking to when its doing single core boosts? 1.6??

Resistivity goes up as the node goes down. Basically the "wires" in the CPU are so thin that the resistance goes up massively. This lead to TSMC and Intel starting ot use cobalt and other material changes in 7/10nm. So it does make some sense that a high voltage is required

https://fuse.wikichip.org/news/525/iedm-2017-isscc-2018-intels-10nm-switching-to-cobalt-interconnects/2/

Cobalt as a raw material is actually more resistive than copper but the barrier layers and other factors mean it's a better choice when implemented into a CPU. As the nodes shrink further resistivity will continue to go up.

[dfsdfgfkjsefoiqzemnd]

@LukeSavenije Wendell dropped another one focusing on Zen2 + Linux

 

 

[Delicieuxz]

12 minutes ago, Lathlaer said:

@Delicieuxz yeah I can't seem to decipher those charts as well. It seems like the medium chosen to give the information is poorly chosen. It's obvious that the more your system costs the less % of the price is the CPU upgrade alone but when you post a pie chart with yellow and blue, it seems prudent to show what is yellow and what is blue.

I think the % figures for higher-priced systems don't make sense as a % of total system cost, either, because the difference between $329 and $399 is $70, and $70 doesn't represent 15% of the cost of a $1,000 build, 14% of a $1,250 build, 14% of a $1,500 build, and 13% of a $2,000 build. Whatever these graphs are showing, whatever is their base metric that all others are percentages of, I can't tell by looking at the images and calculating any of the figures given within them.

 

The given "weighted overall score" (which stats are being weighted?) for the R7 3700X is 83.21%, and, for the i7 9700K, 76.11% is the given weighted overall score. 83.21 is 109% of 76.11, not 133%. So, I don't see any of the Upgrade or Build percentages as being derived from the weighted overall score, either. I just don't see where the % are coming from, and so have no idea what they mean.

[leadeater]

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

I would like to see more information, nothing here makes sense. If they did disable the power limit and got an accurate reading of 170W that doesnt mean the VRM's that can do 500W+ make any more sense than at 140W.

It doesn't make sense because the boost spec isn't all core, that's the issue. You can go watch the X570 motherboard VRM analysis videos for that information but they are not even close to power limits at 200W, not by a long shot. They are all massively over spec, massively.

 

Just because a single core can hit a higher clock doesn't actually mean it's possible to OC all the cores to that at ambient, that may be the case on Intel or on previous Zen but for Zen those were extremely frequency capped so that's really not even worth looking at in this regard.

 

Either way a current clamp does not lie, software readouts do. If the clamp is telling you 170W (well 126.8A) then 170W is going in to the VRM, of which the VRM would have to be 82% efficient to mean 140W is making to the package and I doubt that it's that bad but who knows.

 

Either you can't remove the package power limit, meaning 5GHz is impossible so that ends that debate, or no one removed the package power limit during OC for the reviews, or they were not allowed to for some reason for the reviews, or the current bios's do not allow it to be removed yet. If find it odd that if the option exists now not a single person did so, even if somehow not allowed under reviewer guides (why would AMD impose this?).

 

In any case I stand by my original post about XOC and the relationship to temperatures and clocks, I'm not saying it's not possible to hit 5GHz on ambient but there is no information that would point to this actually being possible while there is show that it is likely not.

[CommandMan7]

After watching the der8auer and ltt videos and reading the anandtech article;

 

High points: 

Intel's value proposition is completely nonexistent

Gaming perf. is so close as to only be less than 5% difference in many games, with ties in about half

AMD now holds IPC crown, however...

 

Low points:

Intel still holds the single threaded crown due to higher clocks

the maximum observed boosts and overclocking results are pretty disappointing. Der8auer said himself that most chips have a maximum observed boost lower than the advertised spec, and that some chips fail to hit 5Ghz even on LN2.

 

[S w a t s o n]

5 minutes ago, leadeater said:

It doesn't make sense because the boost spec isn't all core, that's the issue. You can go watch the X570 motherboard VRM analysis videos for that information but they are not even close to power limits at 200W, not by a long shot. They are all massively over spec, massively.

 

Just because a single core can hit a higher clock doesn't actually mean it's possible to OC all the cores to that at ambient, that may be the case on Intel or on previous Zen but for Zen those were extremely frequency capped so that's really not even worth looking at in this regard.

 

Either way a current clamp does not lie, software readouts do. If the clamp is telling you 170W (well 126.8A) then 170W is going in to the VRM, of which the VRM would have to be 82% efficient to mean 140W is making to the package and I doubt that it's that bad but who knows.

 

Either you can't remove the package power limit, meaning 5GHz is impossible so that ends that debate, or no one removed the package power limit during OC for the reviews, or they were not allowed to for some reason for the reviews, or the current bios's do not allow it to be removed yet. If find it odd that if the option exists now not a single person did so, even if somehow not allowed under reviewer guides (why would AMD impose this?).

 

In any case I stand by me original post but XOC and the relationship to temperatures and clocks, I'm not saying it's not possible to hit 5GHz on ambient but there is no information that would point to this actually being possible while there is show that it is likely not.

Uh what? You can almost always hit boost on every core when OCing. As you said Zen could, intel can.. who else are we comparing to? Also you mentioned that zen was extremely frequency capped but these only hitting 4.2-4.3GHZ is the same effective cap as Zen1 even though we know it's not actually a frequeuncy wall. In my eyes that supports my opinion that the all-core XFR clock is reachable once constraints are removed.

Current clamps are not 100% accurate, and have variance just in literally how you position it over the wire. 170W into the VRM at say 90% efficiency means the package is at 153W or basically the 140W limit I speak of is basically true.

So we can agree that yes either the socket power limit is the constraint, or maybe heat density of the small AM4 socket? So you agree with me now?

[Drak3]

Just now, S w a t s o n said:

You can almost always hit boost on every core when OCing.

Not on Zen+. It was consistent that XFR was delivering higher single core than was possible to do all core OC, by at least 100MHz.

[cj09beira]

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

Uh what? You can almost always hit boost on every core when OCing. As you said Zen could, intel can.. who else are we comparing to? Also you mentioned that zen was extremely frequency capped but these only hitting 4.2-4.3GHZ is the same effective cap as Zen1 even though we know it's not actually a frequeuncy wall. In my eyes that supports my opinion that the all-core XFR clock is reachable once constraints are removed.

Current clamps are not 100% accurate, and have variance just in literally how you position it over the wire. 170W into the VRM at say 90% efficiency means the package is at 153W or basically the 140W limit I speak of is basically true.

So we can agree that yes either the socket power limit is the constraint, or maybe heat desnity of the small AM$ socket? So you agree with me now?

i think we need someone to make my loved Fmax/Vmin curve, derbauer made one of Fmax/Tmin but its not the same

[S w a t s o n]

Just now, Drak3 said:

Not on Zen+. It was consistent that XFR was delivering higher single core than was possible to do all core OC, by at least 100MHz.

Hence almost always. 100MHZ is a small variance and I have seen Zen+ hit XFR on all core but it's definitely golden sample territory

[cj09beira]

1 minute ago, Drak3 said:

Not on Zen+. It was consistent that XFR was delivering higher single core than was possible to do all core OC, by at least 100MHz.

thats due to xfr pushing more volts into the cpu than one would recommend for all core voltages, would like to see what voltages xfr is pushing into these new cpus because they might be interesting 

[Drak3]

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

Hence almost always.

That’s an entire generation where you’re wrong.

 

And Ryzen 1 was still not an “almost always.”

1 minute ago, cj09beira said:

thats due to xfr pushing more volts into the cpu than one would recommend for all core voltages, would like to see what voltages xfr is pushing into these new cpus because they might be interesting 

Irrelevant to Swatson’s claim. For older Ryzen chips, it’s outright wrong.