AMD Zen 2 (Ryzen 3000) - *** NO COMPETITOR HINTING ***

[MaXxBoulton]

They said under normal used the fan will be off. And the temps only get high when running m.2's in raid and or benching said raid set up.

 

[Murphy]

The only get hot when running M.2 in RAID thing was basically speculation (iirc from Buildzoid) based on limited information, it was the only thing that seemed to explain why they needed active cooling some of the time, however with the new information from GN about the chipset not being able to down clock it sounds like the active cooling is more needed depending on how many devices are connected to it rather than the amount of data being pushed through it.

Having said that i still can't see many people having a need to even touch the chipset as the CPU has enough connectivity options for what's probably the vast majority of users.

 

[decto]

The other odd thing about your analysis is that you state that the node shrink will have directly benefitted the per core power consumption, yet we know that the 2nd gen weren't consuming anywhere near the 12.5w per core that you're attributing to the 3950X.

https://www.anandtech.com/show/12625/amd-second-generation-ryzen-7-2700x-2700-ryzen-5-2600x-2600/5

2700X at 10.5W per core with all core load at stock frequency, no overclock.

A single core on full boost is 23W at stock settings (possibly with PBO), which is the same as an overclock and shows 2700X cores can use way more than 12.5W even with limited 'PBO' overclock.

CPU stays within power budget by limiting clock frequency and so the voltage required by each core. We can bust this power limit but then we deal with the heat.

The 3800x will use a comparable 10.5W per core at stock all core load as AMD have specified the TDP. What we gain is approx 200Mhz and IPC uplift. Overclock or PBO will increase this power usage significantly.

The article clearly shows AMD complies with TDP at full load and how that power is shared between core and uncore.

The similar threadripper article also shows exactly the same TDP compliance and core power scaling at stock.

Why would AMD calculate TDP any different from 1xxx 2xxx and TR?

AMD is only claiming 20% power reduction from Vega 12nm to Navi 7nm. They say 1.5x performance per watt but 1.25x improvement is architecture changes so remaining 1.2x improvement is node efficiency at half the frequency of CPU's. These figures are from E3 so I don't see how we expect the CPU to have significantly higher efficiency gains from 7nm node than we can get with GPU.

 

[Journey]

Well regarding power usage from the I/O and IF connections, we already know that a 2950X uses 40-45w when using 8+ threads, and the I/O die for Ryzen 3xxx is still 12nm so little to no power saving there, other than using 2-channels of RAM vs. 4. Easier to work out per core numbers based on having the I/O power consumption.

 

[Ross Thomson]

True, but it's also at >80C so either they are over clocking on the box cooler, or it's pushing some power

245W of heat it is pushing out. My 1700 pusing 141W (1.4V @ 4.0GHz) using a 240AIO with the fans @ 100% hits 63C using cinebench R15. at my normal 3.9ghz @ 1.325v it pushes 123W of heat and it hits 57C (fans about 70%). I've bought the bykski AM4/TR4 block for the 3950x

 

[Cooper]

I wouldn’t get a 6 unless you’re happy to replace in 1 or 2 years when it starts to get hammered in games

Why would it? next gen Xbox & PS5?

They maybe 8c16t but not all of those cores & threads will be available to games, in fact they may actually turn out to only have 6c12t for games and at a much lower clock speed than a desktop 3600.

6c 12t will be more than enough for gaming through the next generation of consoles, which is where the vast majority of PC games will come from for the next 5yrs.

 

[Potatowithearsontheside]

Their actual data shows all-core at 4.0GHz and 10.5w per core, and the gains from 7nm itself may only be 20%...and directly on the per core level. However, what you've failed to address is that we're comparing a GF node with a TSMC node; one of which was operating significantly outside of it's optimum frequency curve.
My whole point is that I don't think we should be using 2nd gen Ryzen as a reference point here; there's too many differences to be relying upon it.

 

[4K8KW10]

Very, very interesting if true: http://tieba.baidu.com/p/6164223711

 

[Destination]

Did people get the 9900k to manage 5.4?

 

[Jerry64]

Very, very interesting if true: http://tieba.baid

Very taxing bench Cpuz lol

此图为3款Ryzen 3000系列cpu在cpuz下的单核跑分成绩，及一颗2700x锁定4.35ghz的跑分，对比其他Intel 11款处理器的成绩

translates to

This picture shows the scores of the three Ryzen 3000 series cpu in cpuz single core running score, and a 2700x locked 4.35ghz running score, compared with other Intel 11 processor scores.

 

[decto]

Their actual data shows all-core at 4.0GHz and 10.5w per core, and the gains from 7nm itself may only be 20%...and directly on the per core level. However, what you've failed to address is that we're comparing a GF node with a TSMC node; one of which was operating significantly outside of it's optimum frequency curve.
My whole point is that I don't think we should be using 2nd gen Ryzen as a reference point here; there's too many differences to be relying upon it.

There is an AMD stock slide that's shows 4Ghz, however I didn't see that confirmed in the text that it's what they used. With OK cooling many will boost to this.

I'll have to disagree on the process changes based on the TDP Vs clocks AMD has shared. The clocks they shared for the 3800X are the best they will guarantee at 105W and only a small increment (5%) from 2700X so I think the 2xxx is sufficient for indicative comparisons.

My original comment was that 4.5Ghz on all 16 cores on air was unlikely as air coolers start to run into limits around 200W and even at 200W you need to be moving a lot of air so it's not an ideal solution.

Based on the data we have available, including 3950X @ 5.2ghz on LN2 where 2700x hit 6Ghz on LN2 suggests in incremental per core improvement in frequency/power and that despite the 105W TDP on the 3900X and 3950X we will need high end cooling to get the best from them.

 

[Potatowithearsontheside]

Very taxing bench Cpuz lol

此图为3款Ryzen 3000系列cpu在cpuz下的单核跑分成绩，及一颗2700x锁定4.35ghz的跑分，对比其他Intel 11款处理器的成绩

translates to

This picture shows the scores of the three Ryzen 3000 series cpu in cpuz single core running score, and a 2700x locked 4.35ghz running score, compared with other Intel 11 processor scores.

The picfure shows Sunnycove at 3.6GHz pretty much matching current top-tier single thread performance regardless of whether we're looking at current AMD or Intel.
Sometimes what you miss is more telling than what you see.

 

[4K8KW10]

Very taxing bench Cpuz lol

此图为3款Ryzen 3000系列cpu在cpuz下的单核跑分成绩，及一颗2700x锁定4.35ghz的跑分，对比其他Intel 11款处理器的成绩

translates to

This picture shows the scores of the three Ryzen 3000 series cpu in cpuz single core running score, and a 2700x locked 4.35ghz running score, compared with other Intel 11 processor scores.

Well, the question is whether the SunnyCove @3.7GHz will be equal to Mattise @4.7GHz in all applications which would equal to 25-30% IPC lead or the particular CPU-Z bench utilises something in SunnyCove which shows way above average performance gains over the Mattise?

With regards to Mattise to Pinnacle Ridge, 30% higher performance with 8% higher frequency.

 

[GAC]

shows or claims

 

[Ad_Augendae]

I love this little comment......

AMD: 3 weeks till ryzen 3000 Linus: "Time to upgrade to a 9900k"

 

[Potatowithearsontheside]

There is an AMD stock slide that's shows 4Ghz, however I didn't see that confirmed in the text that it's what they used. With OK cooling many will boost to this.

I'll have to disagree on the process changes based on the TDP Vs clocks AMD has shared. The clocks they shared for the 3800X are the best they will guarantee at 105W and only a small increment (5%) from 2700X so I think the 2xxx is sufficient for indicative comparisons.

My original comment was that 4.5Ghz on all 16 cores on air was unlikely as air coolers start to run into limits around 200W and even at 200W you need to be moving a lot of air so it's not an ideal solution.

Based on the data we have available, including 3950X @ 5.2ghz on LN2 where 2700x hit 6Ghz on LN2 suggests in incremental per core improvement in frequency/power and that despite the 105W TDP on the 3900X and 3950X we will need high end cooling to get the best from them.

The Anandtech article says something along the lines of "we initially turned Precision Boost off in our early tests, but that was because we didn't understand what it did." The implication being that it was on for the tests in that particular article. One of the first graphs on that page of the adticle shows that they were getting 4.0GHz across all cores. The later table confirms the power draw through the cores when all cores were loaded.
I think it is interesting that the 3800X gets used as a comparator with the 3950X for making assumptions about power consumption. Is there a reason for not also looking at the 3600X versus the 3900X? Both run the same base, and have fairly similar listed boost clocks, yet one is 6c at 95w, the other 12c at 105w. What is so fundamentally different about them that you are able to say that a lower core count must be drawing close to its rated TDP through the cores alone? I don't get how that could be; we've seen that the IF power draw plateaus with each additional core.
My instinct tells me that the 3600X and 3800X have a whole bunch of headroom.
I'm also fairly confident that the 7nm TSMC node's optimal power curve is 2-300MHz above that of GF's 12nm. IMO, that is completely separate to the efficiencies gained by the node shrink itself.
Maybe I'm wrong, and the beefier VRMs potentially suggest this, but I'm actually not so pessimistic about the gains that have been made here.

 

[EsaT]

Well if you want to be a pedant, I could have typed 'power usage scales exponentially with frequency because in 30 years of overclocking I've yet to find a CPU where you don't need to ramp up the voltage with any significant overclock and this has a drastic effect on power usage.

Then you still have problems in telling the difference between AC and DC.
Overclocking makes power consumption rise linearly with clocks. (actually less because affecting only dynamic consumption and only cores)
It's overvolting which makes power consumption bloat exponentially and even without any clock increase.

I had 1.25VID Q9550 which I undervolted to 1.2V and overclocked fair 30% to 3.7Ghz+change.
Now that was real overclocking.
Time of that that overclocking loose in CPUs is well past.
Now all this hyped over"clocking" is really mostly just brute force making of e-penis space heaters with small performance gains for high power cost.
Making no sense for average user.

And in gaming depending mostly on execution speed of few most heavily loaded threads manual all core clocking is even challenged by automatic boost clocks.
Which basically automatically utilizes that reasonable clocking room.

 

[4K8KW10]

I love this little comment......

AMD: 3 weeks till ryzen 3000 Linus: "Time to upgrade to a 9900k"

I think Linus is only being or trying to be funny. People should avoid the power house i9-9900K. For God's sake, the only wise solution today is to wait.

 

[decto]

Then you still have problems in telling the difference between AC and DC.
Overclocking makes power consumption rise linearly with clocks. (actually less because affecting only dynamic consumption and only cores)
It's overvolting which makes power consumption bloat exponentially and even without any clock increase.

I had 1.25VID Q9550 which I undervolted to 1.2V and overclocked fair 30% to 3.7Ghz+change.
Now that was real overclocking.
Time of that that overclocking loose in CPUs is well past.
Now all this hyped over"clocking" is really mostly just brute force making of e-penis space heaters with small performance gains for high power cost.
Making no sense for average user.

And in gaming depending mostly on execution speed of few most heavily loaded threads manual all core clocking is even challenged by automatic boost clocks.
Which basically automatically utilizes that reasonable clocking room.

I'd agree there, I'm looking for the cooling I need to maximise PBO and XFR and will leave it on full auto except for increasing the TDP limit if it's accessable in the BIOS.

My current 2600k with a phantechs dual tower 140mm fan cooler holds 60C at around 100W consumption (4.5Ghz). Speeding the fans up helps but at the cost of noise.

New build will be full WC so heat won't be an issue and I can use lots of slow and quiet fans on a much bigger cooling area. No RGB or other bling though as no window in the case

 

[andy_mk3]

I love this little comment......

Hah it was sponsored by NVIDIA and gigabyte though. Its hardly a bad choice as much as AMD should be very competitive against it.