Navi/Ryzen 3000 launch Megathread

[ravenshrike]

9 hours ago, Chett_Manly said:

If it is the chiplet architecture, why is Ryzen 3000 able to pull so far ahead of Ryzen 2000, which is also chiplet, but 12nm not 7nm. Do you really think they could have put out a 16 core Ryzen at 12nm? 

It is unlikely the clock for clock IPC improvements on Zen 2 are an artifact of the node. Rather, it was the Zen team taking the lessons learned from themselves and the Zen+ team to improve the architecture and underlying mechanisms like the IF. What 7nm did was allow the higher boost clocks(but seems to be too dense to allow really high multi-core clocks) and just plain give them more room on the chip to shove in more cores.

[leadeater]

11 hours ago, Chett_Manly said:

If it is the chiplet architecture, why is Ryzen 3000 able to pull so far ahead of Ryzen 2000, which is also chiplet, but 12nm not 7nm. Do you really think they could have put out a 16 core Ryzen at 12nm? 

Ryzen 2000 wasn't chiplet, Threadripper and EPYC were multi-die package which isn't the same thing. Chiplet still comes under multi-die package however a single die is not an entire functional X86 CPU.

 

Ryzen 3000 is able to pull so far ahead of 2000 because there are major improvements like double the load/store bit width for Floating Point and the FMAs were also increased from 128bit to 256bit each. There a lot of both major and minor improvements in Zen2 compared to Zen+.

 

1 hour ago, ravenshrike said:

What 7nm did was allow the higher boost clocks(but seems to be too dense to allow really high multi-core clocks) and just plain give them more room on the chip to shove in more cores.

Actually 7nm was supposed to lower clocks and was a significant concern for the engineering team but a solution to the expected degradation was found (don't ask me what, no idea).

[MMKing]

16 hours ago, Chett_Manly said:

People counting Intel out are being silly. AMD's biggest advantage right now is the 7nm process. A process Intel could use if they wanted to swallow their pride. Intel can give up on their 10nm and pay Samsung or TSMC for 7nm and start pumping out chips. They have the engineers, money and market power to do it.

 

Just think where Intel would be today if 2 years ago they gave up on 10nm and got onboard TSMC's 7nm customer list. How well would Ryzen 3000 stack up to a 7nm instead of 14nm+++ Intel chip?

Implying it's only about pride is a bit narrow minded. There are allot of reasons why Intel would want to stay on their own fabrication other than pride. There are among other things, 24 extremely good reasons for Intel to buy from Intel. Intel operates 14 fab sites in USA, Israel, China and Ireland, as well as 10 test/assembly sites in USA, China, Malaysia, Israel, Philippines, Israel and Vietnam.

 

Intel is a 71 billion dollar company, with 107 thousand employees. You don't put the breaks on a company this size just because competition is getting a little stiff.

But is Intel even hurting? Not really, at least not yet. Revenue from 2017, when Zen and the 1800x released to 2018 rose by 12.89% from 62.7 to 70.8 billion. Only just recently has there been signs of stagnation over at Intel, with a quarter to quarter revenue loss by 0.03% comparing 2019Q1 to 2018Q1.

 

But even if there was significant loss in revenue. You would be a lunatic to run off to TSMC, after a single bad year, which there have yet to be.

[RejZoR]

Can someone tell me something, CPU's have 105W TDP for example, but everyone is freaking out about power delivery. What has changed so dramatically that at same or lower wattage that we had for years, VRM is such an issue now? Given how much I know electricity, this can only mean something changed dramatically on the voltage side since W, A and V are connected between each other tightly as output values always depend on these 3. Amperage wouldn't change at same wattage if voltage is the same as well at 105W TDP. So voltage changed then.

 

So, what changed here exactly that we need such high amperage on the VRM side now?

[leadeater]

48 minutes ago, RejZoR said:

Can someone tell me something, CPU's have 105W TDP for example, but everyone is freaking out about power delivery. What has changed so dramatically that at same or lower wattage that we had for years, VRM is such an issue now? Given how much I know electricity, this can only mean something changed dramatically on the voltage side since W, A and V are connected between each other tightly as output values always depend on these 3. Amperage wouldn't change at same wattage if voltage is the same as well at 105W TDP. So voltage changed then.

 

So, what changed here exactly that we need such high amperage on the VRM side now?

Nothing changed other than board vendors willingness to actually spend money on high end designs and components for AMD, even if it's not actually required. Not that it isn't any different for Intel either on the standard desktop side. There is no reason to make 300A to 500A motherboards, unless it's a single SKU targeted specifically for XOC, Intel and AMD alike.

 

Only the higher end of Skylake-X actually needs that as standard.

[cj09beira]

16 hours ago, Chett_Manly said:

If it is the chiplet architecture, why is Ryzen 3000 able to pull so far ahead of Ryzen 2000, which is also chiplet, but 12nm not 7nm. Do you really think they could have put out a 16 core Ryzen at 12nm? 

its amd's biggest advantage but its certainly not the only one right now, even if intel had 10nm cpus out right now amd would still be able to sell more cores for cheaper because of the chiplet design, chiplet vs multidie doesn't have many differences at the 8 core mark but at 16+ it makes a lot of difference, in server cpus its huge, and overall it allows for their effective yields to be very close to 100%.

3 hours ago, RejZoR said:

Can someone tell me something, CPU's have 105W TDP for example, but everyone is freaking out about power delivery. What has changed so dramatically that at same or lower wattage that we had for years, VRM is such an issue now? Given how much I know electricity, this can only mean something changed dramatically on the voltage side since W, A and V are connected between each other tightly as output values always depend on these 3. Amperage wouldn't change at same wattage if voltage is the same as well at 105W TDP. So voltage changed then.

 

So, what changed here exactly that we need such high amperage on the VRM side now?

each time we move to smaller nodes the voltage needed to turn the transistors on is also reducing, so for the same wattage their are more amps

though it seems voltage didn't actually reduce much, and because clocks aren't scaling a lot the power usage is still fairly low considering the amount of cores,

so it could have very well been a problem but its not, at least if you dont count the 3950x

[RejZoR]

Just now, cj09beira said:

its amd's biggest advantage but its certainly not the only one right now, even if intel had 10nm cpus out right now amd would still be able to sell more cores for cheaper because of the chiplet design, chiplet vs multidie doesn't have many differences at the 8 core mark but at 16+ it makes a lot of difference, in server cpus its huge, and overall it allows for their effective yields to be very close to 100%.

each time we move to smaller nodes the voltage needed to turn the transistors on is also reducing, so for the same wattage their are more amps

though it seems voltage didn't actually reduce much, and because clocks aren't scaling a lot the power usage is still fairly low considering the amount of cores,

so it could have very well been a problem but its not, at least if you dont count the 3950x

Well, that's what I was thinking. Voltages haven't really changed that much to make massive amps differences yet everyone is freaking out around it like voltages changed 500% compared to what we had just 2-3 years ago...

[mr moose]

8 hours ago, comander said:

Kodak had serious management/bureaucracy issues. 

Think people having meetings before the real meeting so that they wouldn't say anything stupid. Honesty and transparency were in the gutter and the business was not nimble. 

Their business has managed to remain a float now 20 years after digital photography displaced (Literally destroyed) their core revenue product. 

[mr moose]

1 minute ago, comander said:

They declared bankruptcy in 2012 and while they still exist they've been losing money. 

 

The notion of Kodak continuing to exist is very questionable. 

They are here today, still trading when by all measures they should have gone the way of steam tractors and analog TV.

[BiG StroOnZ]

 

[RejZoR]

Has anyone done a review of 3800X and 3900X using different memory speeds? I'm really interested to see whether I'd have to invest into new memory too or I could keep my existing kit if there is not much of a difference. Coz with memory I'm looking at 1000€ investment for 3900X. I frankly think I'd have more benefit sticking a 1000€ graphic card next year and use it with 5820K and 32GB RAM that I have now. Damn. And I wanted to go Ryzen so much...

 

EDIT:

No matter how tempted I am about Ryzen, I'll be sticking with 5820K for a while. Found this test that sort of confirmed my suspicions:

 

Sure, it depends on conditions, but 5820K is apparently just 10-15fps behind Ryzen 3900X. I'm assuming that's stock clock. Given that mine is running at 4.5GHz (stock it only goes up to 3.6GHz and in heavy multithreading stays around 3.4GHz afaik). So, when I normalize the framerate with 4.5GHz in mind, it's really not worth investing 1000€ into all new AMD platform as my gains for mostly gaming will be so small it's just not worth it. Granted, if you're buying a new system or upgrading from something older and weaker than my Intel HEDT, then hell yeah, AMD is totally worth it. But for people who bought high end not long ago, it's better to invest same money into graphic card. The gains will be much bigger.

[illegalwater]

It turns out that PBO + AutoOC barely does anything.

[ravenshrike]

40 minutes ago, System32.exe said:

It turns out that PBO + AutoOC barely does anything.

It would be interesting to take 5 samples of each processor, undervolt them to 1.1 with PBO and AOC on, overclock the IF to 1800 while paired with 3600 memory, and then do a bunch of benches.

[RejZoR]

So, bascally all of this literally does nothing at all. I don't get it why is it so hard to have 3900X spec'd at 105W TDP and just give user ability to unlock the headroom. This alone should allow clocks to stay higher and closer to boost clock, if temperature allows it of course. It's how boost OC should be. Not this mess of hundred settings that in the end seem to do nothing.

[porina]

I don't think we're that different from recent past AMD CPUs when it comes to overclocking. For example, my 2600 stock would run all cores active around 3.9, but with a manual overclock I could run it Cinebench stable at 4.25. The 3700X... isn't much different. Clocks are more variable with load, typically 3.8 to 4.0 all cores loaded, and in fixed voltage/clock overclocking 4.3 is about manageable with my current cooling. I'm climbing the voltage wall though, and still have voltage headroom but not thermal headroom with a 240mm AIO. With better cooling I can get a bit more, but doubt I'll get much past the single core boost of 4.4 GHz without extreme cooling.

 

These CPUs as sold are running close to their limit, so overclockers have extremely limited potential to extract further performance. On the plus side, normal users can get most of the performance without worrying about tinkering. I still think the coolers are inadequate for sustained higher loads so and upgrade there will prevent thermal limiting.

[cj09beira]

3 hours ago, porina said:

I don't think we're that different from recent past AMD CPUs when it comes to overclocking. For example, my 2600 stock would run all cores active around 3.9, but with a manual overclock I could run it Cinebench stable at 4.25. The 3700X... isn't much different. Clocks are more variable with load, typically 3.8 to 4.0 all cores loaded, and in fixed voltage/clock overclocking 4.3 is about manageable with my current cooling. I'm climbing the voltage wall though, and still have voltage headroom but not thermal headroom with a 240mm AIO. With better cooling I can get a bit more, but doubt I'll get much past the single core boost of 4.4 GHz without extreme cooling.

 

These CPUs as sold are running close to their limit, so overclockers have extremely limited potential to extract further performance. On the plus side, normal users can get most of the performance without worrying about tinkering. I still think the coolers are inadequate for sustained higher loads so and upgrade there will prevent thermal limiting.

i would like to see how a big water loop would do, as i suspect heat density is the bigger issue 

[thorhammerz]

7 minutes ago, cj09beira said:

i would like to see how a big water loop would do, as i suspect heat density is the bigger issue 

If it's a heat density issue like the i7-7700k, then the bottleneck is more on transferring the thermal energy efficiently out of the die(s) to the IHS / cooling-solution, rather than the thermal capacity of the cooling solution itself.

[leadeater]

49 minutes ago, cj09beira said:

i would like to see how a big water loop would do, as i suspect heat density is the bigger issue 

Just watch the GN LN2 overclocking video, he does above ambient gradient temperature drops and shows how it effects clocks all the way down to around 0C.

[Spotty]

1 hour ago, leadeater said:

Just watch the GN LN2 overclocking video, he does above ambient gradient temperature drops and shows how it effects clocks all the way down to around 0C.

I haven't seen GNs video, but Der8auer showed a graph plotting clock speeds against CPU temp in his video for the R9 3900x as well. Goes from 50°C down to -140°C
 

 

 

[mr moose]

7 minutes ago, Spotty said:

I haven't seen GNs video, but Der8auer showed a graph plotting clock speeds against CPU temp in his video for the R9 3900x as well. Goes from 50°C down to -140°C
 

 

 

 

That graph has to be wrong, the TDP is only 105W and it says over 170W, DID AMD LIE!!!!!!1!!!    

[BiG StroOnZ]

 

[thorhammerz]

9 hours ago, mr moose said:

That graph has to be wrong, the TDP is only 105W and it says over 170W, DID AMD LIE!!!!!!1!!!    

If you squint hard enough after a beer (or three), the line does look to flatten out at 105W .

[TechyBen]

Halp! Do I go Intel or Ryzen... I'm now confused! 

 

I will have to check out price/performance for the few games left I play... anyone know any good Kerbal Space Program benchmark comparisons? 

[CiBi]

1 hour ago, TechyBen said:

Halp! Do I go Intel or Ryzen... I'm now confused! 

 

I will have to check out price/performance for the few games left I play... anyone know any good Kerbal Space Program benchmark comparisons? 

I think for KSP you'd be better with an 9700K, but I wouldn't recommend buying into the Intel ecosystem right now...
Tbh if price/performance is all you're after, go with the 3600 non X and enjoy that superb value

[RejZoR]

Why is X570 AORUS Pro WiFi cheaper than non-WiFi version?! WTF? I've compared both and there is just difference in WiFi and some USB ports iirc. And 40€ price difference in favor of WiFi version...