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

[MyBrainz]

Just gona put this here..

The Ryzen 3000 series will debut a new product tier for AMD: Ryzen 9. In this case, the Ryzen 9 3900X will be AMD’s first mainstream desktop 12-core processor. The processor is the only one of the group that uses two chiplets, in a 6+6 configuration.

https://www.anandtech.com/show/1440...-cores-for-499-up-to-46-ghz-pcie-40-coming-77

Not sure what the meaning of 'the group' they are talking about, just the R9 or R7 too?

 

[triss]

Just gona put this here..



https://www.anandtech.com/show/1440...-cores-for-499-up-to-46-ghz-pcie-40-coming-77

Not sure what the meaning of 'the group' they are talking about, just the R9 or R7 too?

I'd assume its the ones mentioned in article, so Ryzen 5,7,9 .

 

[decto]

I think the reason why the 3800x seems to have such a high tdp compared to the 3700x and 3900x is because the 3700x has 2 4 core chiplets and it shuts down 1 chiplet to get maximum 4 core boost while keeping power use down and the 3900x does the same but with 6 core boost.

TDP is purely boost clock management for new processors. 2700 and 2700x are same silicon with different boost profiles, 3700x and 3800x with different boost profiles.

At 105W there is only a 200Mhz base/150Mhz boost clock difference between 2700X and 3800X. That's a pretty modest improvement, just 5% clocks for new node at same power. The 2700X already has an agressive boost profile so I expect the 3800X to mirror it + 200Mhz.

The 3900x will have a much more conservative boost profile at 105W.

Most of the processors have XFR so they can boost well above TDP if you have the right cooling.

The question for me is which coolers can shift 200W+ while keeping die temperature <60C to maximise boost.

Maximising boost will be much easier on the 3800X with 8 cores, so is likely to need high end air with good fans or a beefy AIO / CLC. Whereas maximising XFR on the 3900X is going to need a quality high flow AIO or full water-cooling.

2xxx threadripper is a good analogue as with the XFR equivalent these will boost well over TDP given the right cooling.

 

[FoxEye]

Can they do that? You mean during work loads? What's does the OS think?

I very much doubt it.

Why would you "shut down" half of your CPU cores under load?

The OS can't (I wouldn't have thought) decide between using 8 cores or 4 cores, how would it even make that call? And when would it make the call to turn the other 4 cores back on?

To me that doesn't make any sense. The only time I can see CPU cores being turned off is in a low-power idle state, not under load.

 

[Murphy]

Why are people associating TDP with with things like expect boost clocks and power usage when it's just the maximum amount of heat generated by the chip and without knowing how many chiplet are under the IHS it's almost impossible to know how that relates to boost clocks and power usage, there seems to be way to many variables to make any sort of educated guess.

I know it's fun to speculate but at least base the speculation on a theory and not just guesses, assumptions, and wishes, a two chiplet package could have higher heat dissipation requirements but then so could a single chiplet package that's clocked higher, two chiplets could also have lower heat dissipation requirements because they're spread out, i mean it's hard enough to workout how AMD and Intel both calculate TDP differently let alone draw any assumptions on how the silicon behaves, and in AMD's case is configured.

 

[Journey]

At 105W there is only a 200Mhz base/150Mhz boost clock difference between 2700X and 3800X.

Are you assuming that it is the same number of cores that will boost? What about the 3600X which is 95w 6c/12t, 10w less than the 3800X, 8c/16t.

You are making a lot of assumptions with information based on older node, and no first hand experience of the new products, when in reality it might be nothing at all like the 2xxx generation

 

[Troezar]

Are you assuming that it is the same number of cores that will boost? What about the 3600X which is 95w 6c/12t, 10w less than the 3800X, 8c/16t.

You are making a lot of assumptions with information based on older node, and no first hand experience of the new products, when in reality it might be nothing at all like the 2xxx generation

Nail on the head. There are so many changes its just guesswork until we have more details. That's what keeps everyone coming back here. If we knew it all there'd be nothing to discuss

 

[keyser]

Are reviews are out 7th July or sooner?

3800X 4.5Ghz 105W
3700X 4.4Ghz 65W

These are interesting but what I hope they aren't suggesting that you aren't going to get much overclock out of these. Still impressive power. Anandtech/Gamers Nexus reviews will put em through their paces

 

[Perfect_Chaos]

Waiting for Digital Foundry's benchmarks, they always seem to be quite late though. Are the chips already given to reviewers etc so they can test well before the 7th?

 

[decto]

Why are people associating TDP with with things like expect boost clocks and power usage when it's just the maximum amount of heat generated by the chip and without knowing how many chiplet are under the IHS it's almost impossible to know how that relates to boost clocks and power usage, there seems to be way to many variables to make any sort of educated guess.

I know it's fun to speculate but at least base the speculation on a theory and not just guesses, assumptions, and wishes, a two chiplet package could have higher heat dissipation requirements but then so could a single chiplet package that's clocked higher, two chiplets could also have lower heat dissipation requirements because they're spread out, i mean it's hard enough to workout how AMD and Intel both calculate TDP differently let alone draw any assumptions on how the silicon behaves, and in AMD's case is configured.

In AMD's case, TDP is the wattage the cooling must dissipate at a given difference in core and air temperature to get the design performance of the CPU. In the case of the 3800X that's 4.5 GHz on 1 / possibly 2 cores or 3.9 GHz on all cores. If it could do 4.0Ghz at 105W that would be the base spec.

They calculate is somewhat backwards in that they qualify the boost clocks with a design TDP rather the pick clocks and the go with the output TDP.

With XFR the CPU can exceed the specified clocks if there is cooling headroom. Any boost above these clocks exceeds TDP, that's how Ryzan 1xxx, 2xxx and theadripper operate and how AMD describes XFR. Fairly safe assumption that 3xxxX series with XFR will behave similarly, though the exact core boost vs. number threads will not be known until we have full analysis.

TDP is absolutely not the maximum power usage of the chip, nothing like in fact. When over clocked a 2700x can consume 200W+ on EPS12V and high core theadripper near 500W.

By enabling XFR you allow CPU to exceed TDP.

TDP is relevant as 3900X will need a lot more cooling to run XFR clocks as high as the 3800X as it has 50% more cores.

There is no magic formulae for power consumption of a set number of cores, power usage is managed in real time by precision boost and XFR.

The only certainty is that if you want the best boost clocks then you need good cooling and the more cores you want to boost high, the better cooling you're going to need.

 

[Murphy]

In everyone's case it's the wattage the cooling must dissipate, how they calculate that number though is very different, I'm not sure what you mean when you say "at a given difference in core and air temperature to get the design performance of the CPU" though as it sound like you're spouting techno-babble in an attempt to cover up a lack of understanding.

The rest of what you've said is pretty much Gish Gallop as like i said you cant make any assumption of how TDP relates to anything without knowing the physical makeup of the chip, throw in knowing very little about how efficient 7nm is and what type of 14nm they're using for the I/O die you may as well be pulling theories out of thin air.

 

[StriderX]

Does it even matter... just use water and it wont matter how hot it gets

 

[decto]

In everyone's case it's the wattage the cooling must dissipate, how they calculate that number though is very different, I'm not sure what you mean when you say "at a given difference in core and air temperature to get the design performance of the CPU" though as it sound like you're spouting techno-babble in an attempt to cover up a lack of understanding.

The rest of what you've said is pretty much Gish Gallop as like i said you cant make any assumption of how TDP relates to anything without knowing the physical makeup of the chip, throw in knowing very little about how efficient 7nm is and what type of 14nm they're using for the I/O die you may as well be pulling theories out of thin air.

This might help.

https://www.anandtech.com/show/13124/the-amd-threadripper-2990wx-and-2950x-review/12

There a multiple ways to measure TDP, I did quote AMD incorrectly , link shows that AMD actually use idle and load temp differential however the result is the same in that it is standardised by the OEM in this case. At equal TDP of 105W measured by the same method by the same OEM the increase in clocks is 5%. Just wow, that tells me all I need to know about 7mm frequency scaling.

As for the 14nm die, that seems to work just fine on the 65W parts leaving power budget for 6 to 8 cores. The 14nm die can run at a much lower frequency than the CCX chiplet(s). If IO die impacted performance or power budget significantly then it would be on 7/12nm.

So the 'obvious' assumption that the 14nm chiplet has a modest and relatively fixed power budget can only lead to the conclusion that in AMD's standardised TDP testing we don't see a significant gain in clock frequency from the new node.

There seems to be a collective delusion that 7nm is some miracle node, at low frequency it is far more power efficient but with this process maturity and such high frequencies the main benefit is die shrink and cost.

65nm Pentium extreme edition had a base clock of 3.7Ghz and could be OC'd to 5Ghz back in 2006 and Intel has barely made any progress from 32nm Sandy max clocks to 14nm++++++++ max clocks despite the significantly lower nodes.

Lower nodes work well in GPU where massive parallel at high efficiency and moderate clocks <2Ghz allows many more 'cores' in the same silicon space and within a fixed power budget.

Above 4Ghz silicon is exponentially power hungry and so CPU needs to go parallel.

So I'll stick to my original point. Zen 2 needs big coolers to get the best out of it or you'll be wasting your cash on a 12 core chip that will spend all its time throttled by power budget.

 

[Murphy]

You seem to be missing the point, by quiet someway, you can't use previous TDP, power draw, or clock speeds to extrapolate anything because just hiving off the I/O onto a separate die would have a dramatic effect on the thermal output of the package, and that's before we throw the one or two compute dies into the mix, 7nm compute dies, percentages of dark silicon changing and all the other physical changes that may have happened.

And no, you won't need big coolers to get the best out of it because it will be spending all its time throttled by power budget, you'll need a cooler capable of dissipating the TDP to achieve the advertised boost clocks, if you have a cooler that's capable of dissipating higher wattage it may boost higher than the advertised speed but it would be wrong to say if you're cooler doesn't exceed the TDP that it won't reach the advertised boost clocks.

 

[robfosters]

Sorry for the real general question, but so many things are floating round my head.

I know it’s difficult to say precisely, but do you think a £1500 budget will be enough for:

Ryzen 3800X,
X570 mobo
3400Mhz RAM (16GB)
M2 drive
New Navi card
Nice RGB case

 

[easyrider]

Sorry for the real general question, but so many things are floating round my head.

I know it’s difficult to say precisely, but do you think a £1500 budget will be enough for:

Ryzen 3800X,
X570 mobo
3400Mhz RAM (16GB)
M2 drive
New Navi card
Nice RGB case

3800x £399
X570 mobo £200
Ram £120 ish
M2 depending on size ?
Navi...400-600?

Don’t see why not

 

[pmc25]

TDP is purely boost clock management for new processors. 2700 and 2700x are same silicon with different boost profiles, 3700x and 3800x with different boost profiles.

At 105W there is only a 200Mhz base/150Mhz boost clock difference between 2700X and 3800X. That's a pretty modest improvement, just 5% clocks for new node at same power. The 2700X already has an agressive boost profile so I expect the 3800X to mirror it + 200Mhz.

The 3900x will have a much more conservative boost profile at 105W.

Most of the processors have XFR so they can boost well above TDP if you have the right cooling.

The question for me is which coolers can shift 200W+ while keeping die temperature <60C to maximise boost.

Maximising boost will be much easier on the 3800X with 8 cores, so is likely to need high end air with good fans or a beefy AIO / CLC. Whereas maximising XFR on the 3900X is going to need a quality high flow AIO or full water-cooling.

2xxx threadripper is a good analogue as with the XFR equivalent these will boost well over TDP given the right cooling.

Boost profile is managed according to mobo BIOS and heat. It doesn't work the way you seem to think it does. Boost will go way higher than 105W on performance oriented boards at reasonable temperatures, the 3900X would only be more 'conservative' within certain strictures. So it doesn't really matter what the spec for boost is on 3900X vs 3800X. If the 3900X is binned higher and you have adequate cooling, the 3900X will almost certainly clock higher (boost or OC).

But of course there's no guarantee there's any difference in bin at all.

Higher bins may all be going to EPYC, TR4 and potentially 16 core Ryzen if it actually launches.

 

[robfosters]

3800x £399
X570 mobo £200
Ram £120 ish
M2 depending on size ?
Navi...400-600?

Don’t see why not

M2 will probably be the 500GB Samsung Evo 970. It’s £140 atm.

 

[CAT-THE-FIFTH]

Sorry for the real general question, but so many things are floating round my head.

I know it’s difficult to say precisely, but do you think a £1500 budget will be enough for:

Ryzen 3800X,
X570 mobo
3400Mhz RAM (16GB)
M2 drive
New Navi card
Nice RGB case

You should be fine IMHO.

M2 will probably be the 500GB Samsung Evo 970. It’s £140 atm.

That is way too much,as competing 500GB M2 drives are £100ish already. My WD Black 500GB was £100 last year. Aim for 1TB at that price or even for a bit more.

 

[robfosters]

You should be fine IMHO.



That is way too much,as competing 500GB M2 drives are £100ish already. My WD Black 500GB was £100 last year. Aim for 1TB at that price or even for a bit more.

Are you talking ordinary SATA or NVMe though?