AMD Ryzen 3000 Builders Thread

[Fanatical Meat]

It's a good point. This is the first time Intel won't have an alternate enthusiast platform released in the same year. So the motherboard manufacturers are looking at Zen 2 performance and basically hoping not to lose those high margin retail board sales. That's part of the reason they are so gun ho on x570. But the fact is AMD just doesn't sell a decent fraction of the parts that Intel does. Its going to increase a lot this gen. But I doubt the market for AMD based enthusiast priced retail boards is going to be large enough for 50+ different boards.

Pure speculation
I think it’s the typical corporate stuff.
Been in a sales role or sales management role for a long time. Every place I’ve worked when there is a new product or service that is better or does something different it’s always over priced. The marketing people who thought it up overvalue it, engineers do it too but in a deferent way.
Marketing people are typically very passive aggressive, they don’t want to hear anything bad about their projections.
Engineers want budgets to build the perfect thing. Bigger budget means bigger cost on the end product.
When the two work together “good enough” gets thrown to the side. For example how many people really need multiple Ethernet connections, how many people need 10 gigabit connections?
Feels like 2/3rd the 570 boards are targeted at $250+ buyers and 1/3rd at $180-$250 buyers. I cannot imagine there are that many high end buyers, I cannot imagine the low end buyer is looking to spend around $200.
Features will be dropped and we will lower cost 570 boards. I have no doubt about this, just not sure when. Maybe around the holidays.

 

[Markfw]

Pure speculation
I think it’s the typical corporate stuff.
Been in a sales role or sales management role for a long time. Every place I’ve worked when there is a new product or service that is better or does something different it’s always over priced. The marketing people who thought it up overvalue it, engineers do it too but in a deferent way.
Marketing people are typically very passive aggressive, they don’t want to hear anything bad about their projections.
Engineers want budgets to build the perfect thing. Bigger budget means bigger cost on the end product.
When the two work together “good enough” gets thrown to the side. For example how many people really need multiple Ethernet connections, how many people need 10 gigabit connections?
Feels like 2/3rd the 570 boards are targeted at $250+ buyers and 1/3rd at $180-$250 buyers. I cannot imagine there are that many high end buyers, I cannot imagine the low end buyer is looking to spend around $200.
Features will be dropped and we will lower cost 570 boards. I have no doubt about this, just not sure when. Maybe around the holidays.

I have to agree. Now somebody tell me why in ANY situation, you need multiple ethernet connections ? So if one fails ? They will be hooked to the same switch in a data center or a home.
10 gigabit ? only in a data center. And I am sure there are other things we don;t need, but those 2 are good ones.

 

[Fanatical Meat]

I have to agree. Now somebody tell me why in ANY situation, you need multiple ethernet connections ? So if one fails ? They will be hooked to the same switch in a data center or a home.
10 gigabit ? only in a data center. And I am sure there are other things we don;t need, but those 2 are good ones.

Every board having like 20 million color RGB lighting, multiple M2 slots with raid.
I’m sure some people want that stuff but certainly not everyone.
The boards feel like a punch list of stuff people say they want. Reality is people say they want a bunch of stuff until they know it comes with a cost or they use an excuse like I want 20 usb 3 ports when they really mean, “I’m not comfortable buying this Ryzen 1700 because your FX line was so poor”

 

[Markfw]

Every board having like 20 million color RGB lighting, multiple M2 slots with raid.
I’m sure some people want that stuff but certainly not everyone.
The boards feel like a punch list of stuff people say they want. Reality is people say they want a bunch of stuff until they know it comes with a cost or they use an excuse like I want 20 usb 3 ports when they really mean, “I’m not comfortable buying this Ryzen 1700 because your FX line was so poor”

RGB lighting ? the kids love it, but its cheap and you can turn it off. Multiple M.2 slots ? I already have 2 boxes with this, one boots linux, and one boots windows. Raid ? I could even see that for raid 1, but not for speed (raid 0 maybe ?) They are already screamers. Boot windows in 7 seconds ? Everything else almost instantaneous ?

 

[Fanatical Meat]

RGB lighting ? the kids love it, but its cheap and you can turn it off. Multiple M.2 slots ? I already have 2 boxes with this, one boots linux, and one boots windows. Raid ? I could even see that for raid 1, but not for speed (raid 0 maybe ?) They are already screamers. Boot windows in 7 seconds ? Everything else almost instantaneous ?

Not saying it’s not needed, just saying maybe one or two boards with multiple m2 then a couple with one m2.
Can’t imagine that the entire market needs RGB & multiple m2.
Agreed RGB is cheap however it still has cost, even if the cost is sub $175 boards will not have RGB Lighting, buy a $200+ board for RGB. Simply put it is a nudge to get people to spend more.

 

[Topweasel]

I have to agree. Now somebody tell me why in ANY situation, you need multiple ethernet connections ? So if one fails ? They will be hooked to the same switch in a data center or a home.
10 gigabit ? only in a data center. And I am sure there are other things we don;t need, but those 2 are good ones.

I mainly use my wireless for this. But I typically, have a dedicated networking connection for my PC, then I have a bridged connection for my VM's that they all share.

 

[Adul]

I have to agree. Now somebody tell me why in ANY situation, you need multiple ethernet connections ? So if one fails ? They will be hooked to the same switch in a data center or a home.
10 gigabit ? only in a data center. And I am sure there are other things we don;t need, but those 2 are good ones.

Well I use mind in bonded connections for 2 Gbps connection, which is a faster connection to my freenas box which has 4 Gbps connections bonded together. Helpful when you start moving large video files around. Hell, that is enough justification for 10Gbps connections if you are any kind of content creator.

 

[JDG1980]

I have to agree. Now somebody tell me why in ANY situation, you need multiple ethernet connections ? So if one fails ? They will be hooked to the same switch in a data center or a home.
10 gigabit ? only in a data center. And I am sure there are other things we don;t need, but those 2 are good ones.

My NAS can saturate a gigabit connection, so this is one good reason to have faster LAN connectivity. That said, unless the NAS has SSDs, the cheaper 2.5Gb Ethernet offered by Aquantia and Realtek is probably good enough.

As for multiple Ethernet connections... good question. It would make sense on, say, an Atom board which you might want to use as a pfSense router, but on the average desktop, you're going to be using one link.

 

[beginner99]

Their top-end X570 board has a beefy 8-layer PCB so it will be interesting to see how memory clocks on it.

On the otherhand goign over 3776 mhz (o rhwat the exact number is) seems pointless as this accordign to AMD is the limit of the infinity fabric at full speed.

 

[DrMrLordX]

On the otherhand goign over 3776 mhz (o rhwat the exact number is) seems pointless as this accordign to AMD is the limit of the infinity fabric at full speed.

Are they? It seems to me that they're just selecting the 1/2 speed multi when you pick XMP settings over DDR4-3733. I think you can probably get 1:1 IF ratio working if you tune by hand.

 

[Topweasel]

Are they? It seems to me that they're just selecting the 1/2 speed multi when you pick XMP settings over DDR4-3733. I think you can probably get 1:1 IF ratio working if you tune by hand.

Because they selected that as the limit of their IF. I mean its possible that people can generally manually set it at 1:1 higher, but they selected that because they weren't comfortable running IF higher. But point would stand that AMD thinks that the limit of IF somewhere betweetn 3733 and the next general XMP setting.

 

[jpesk2]

I think the ability to go to the 1/2 speed multi for IF/MEM is really going to be useful for the zen2 APUs. I think it will allow the APUs to finally stretch their legs a little bit, and not be so bandwidth bound. Hopefully they can get the APUs out at the end of the year, or the beginning of next year.

 

[DrMrLordX]

Because they selected that as the limit of their IF. I mean its possible that people can generally manually set it at 1:1 higher, but they selected that because they weren't comfortable running IF higher. But point would stand that AMD thinks that the limit of IF somewhere betweetn 3733 and the next general XMP setting.

Exactly. I think AMD did this to improve compatibility with XMP2.0

 

[Sunaiac]

Hey,

Just started buying parts for a while new ryzen tower to replace my x58 main system.

Mainly because I want to change, not because I need.
Still not sure I'll go r9 3000 or threadripper. Most probable choice : x470 taichi, 3900x, full waterblock.
If x570 looks useful for more than gb/s I'd never feel on the NVMe, maybe this then, but I'd be annoyed not having a good waterblock quickly.
I'm also tempted to start with a cheap athlon and wait for the 3950x, but its price is not too good. 25% more cores for 50% more...

Finally, since aiming 12-16 cores, I've basically eliminated the x399 path which *I think* will be more useful for 24+ cores.
I'll only "miss" quad channel. Not sure what impact it gonna have, guess I'll know when the 16 core ripper comes
But I must say I'm kinda tempted to buy a x399 taichi with a 1900x and wait for Zen2 Threadripper too. I'll see the 3900 review to make my final choice.

Parts so far :
- 16GB (two sticks) Gskill 3600 CL15
- 750 seasonic focus gold (started with a 650 focus platinum, but then this one is a bit cheaper and I don't think I'll see any difference, except it might stay in silent mode longer if its based on % of max power consumption)
- 1TB crucial P1 nVME
- Phanteks P600s anthracite

Basically missing one motherboard, one CPU and one graphic card, until 07/07

 

[Christopher Bohling]

Didn't read the whole thread so sorry if I'm beating a dead horse, but has there been any official discussion from AMD regarding VRM requirements to get the most out of the higher-end Zen 2 chips? I want to replace my 1700 (not sure whether I want 8 or 12 cores yet) but have been anxiously waiting for reports of how it will perform on X370. I have an Asus Prime X370 Pro, which is 6 VRM phases for the CPU and 2+2 for the SoC. I'm presuming this will be sufficient for 3700X/3800X but I'm interested to see if the 12-core is even possible. Even stock would be cool.

 

[Fanatical Meat]

Didn't read the whole thread so sorry if I'm beating a dead horse, but has there been any official discussion from AMD regarding VRM requirements to get the most out of the higher-end Zen 2 chips? I want to replace my 1700 (not sure whether I want 8 or 12 cores yet) but have been anxiously waiting for reports of how it will perform on X370. I have an Asus Prime X370 Pro, which is 6 VRM phases for the CPU and 2+2 for the SoC. I'm presuming this will be sufficient for 3700X/3800X but I'm interested to see if the 12-core is even possible. Even stock would be cool.

No its likely wait for reviews.
AMD said they should work, others have pointed out there may not be enough power delivery for 12/24+ cores on some or most of the old stuff.
The 570 boards all appear to have very beefy power delivery.

 

[lightmanek]

Not enough power to OC it to the max. but plenty of power for stock or mild OC.

Guys, even crap 6 phases capable of 40A each gives 240A at 1.2V. This is enough for over 280W drawn by CPU dies alone without SOC!
Obviously you would have to have water cooling with chiller on CPU side to keep it stable and replace MB VRM radiator with something more substantial on X370 Prime to sustain that.

For normal mild operation I doubt you will use more than 160A from VRM even on 16 core Ryzen.

Only thing I would take care of when using this board with 12C or 16C CPUs is active airflow over VRM section.

PS. I do have that exact board in my son's PC and going to upgrade to new Zen with it.

 

[Christopher Bohling]

Not enough power to OC it to the max. but plenty of power for stock or mild OC.

Guys, even crap 6 phases capable of 40A each gives 240A at 1.2V. This is enough for over 280W drawn by CPU dies alone without SOC!
Obviously you would have to have water cooling with chiller on CPU side to keep it stable and replace MB VRM radiator with something more substantial on X370 Prime to sustain that.

For normal mild operation I doubt you will use more than 160A from VRM even on 16 core Ryzen.

Only thing I would take care of when using this board with 12C or 16C CPUs is active airflow over VRM section.

PS. I do have that exact board in my son's PC and going to upgrade to new Zen with it.

I have an NH-D15 cooling the CPU - if I'm right about the board layout, this is exhausting air over the VRM heatsinks as well, right? Should I plan to change this layout? The rest of my layout is just 2x140mm intake fans in the front and 1x120mm exhaust at the back, no top fans, in a Phanteks Enthoo Pro M TG. I could mount a top fan blowing air over the VRMs I guess.

 

[DrMrLordX]

Guys, even crap 6 phases capable of 40A each gives 240A at 1.2V. This is enough for over 280W drawn by CPU dies alone without SOC!

1). "6 phases" often means 4+2 configs. 4 for the CPU, 2 for everything else (SoC functions, RAM controller). You already have to tone down your maximum current limits to 160a. For reference, my R7-1800x @ 4.0 GHz can pull peak current of 152a in an AVX workload. A 2700x @ 4.3 GHz likely pulls much more. If you take 8c Matisse near its clockspeed limits, I expect the current draw to go even higher, not to speak of the 12c and 16c chips. If you're satisfied with not being at the chip's clockspeed limits, then I would say a 2700x @ 4.0 GHz would probably pull less current than my 1800x and an 8c Matisse @ 4.3 GHz would pull less current than a 2700x. So it isn't all bad.

2). Current ratings for VRM components are somewhat fudged. Many components are rated @ 25C which never happens during normal operation. Ratings can go down by as much as 20-25% by the time the VRM temps reach 80-100C. Higher-quality VRM components seem better able to maintain their ratings at higher temps.

3). Duty cycle matters! VRMs can be switched on or off as they're needed. The more phases you have, the longer phases stay switched "off", which lets them run at cooler temperatures without need for additional cooling (fans, etc.).

4). Voltage overshoot is a thing. Quality VRM layouts (twinned vs doubled, or even better "true" VRM layouts that require neither twinning nor doubling, as seen on the high-end Gigabyte x570 boards) can do a lot to mitigate voltage overshoot. In contrast, if you have a sparse VRM config that's being pushed to its limits, vdroop and overshoot are only going to be exaggerated as the VRMs struggle to maintain basic operational parameters.

For normal mild operation I doubt you will use more than 160A from VRM even on 16 core Ryzen.

If you're okay seeing it stay down around 3.5 GHz on anything that uses the majority of the cores (which is why you bought the chip, right?) then . . . have at it!

Only thing I would take care of when using this board with 12C or 16C CPUs is active airflow over VRM section.

The 4+2 boards need active airflow over the VRMs already with 8c chips.

I have an NH-D15 cooling the CPU - if I'm right about the board layout, this is exhausting air over the VRM heatsinks as well, right?

I have an NH-d15 in my current machine, and I had an NH-d14 previously. It, like most "L-shaped" towers, does very little to cool board components. Top-down coolers will help keep VRM sinks saturated with air. My solution to the problem was: case airflow. My old strategy involved taking off the side of the case and pointing a desk fan sort of at parts of the motherboard that I thought needed it. Mostly I targeted the RAM, but it did help the VRMs too. Then I went a little bit less hackjob-ish and just got a case with a huge side panel fan that moves air pretty evenly across the entire board. VRMs on my x370 Taichi stay nice and cool even when my 1800x pulls over 140a (which it easily does). Of course it helps that the board has more phases than it needs.

 

[fleshconsumed]

1). "6 phases" often means 4+2 configs. 4 for the CPU, 2 for everything else (SoC functions, RAM controller). You already have to tone down your maximum current limits to 160a. For reference, my R7-1800x @ 4.0 GHz can pull peak current of 152a in an AVX workload. A 2700x @ 4.3 GHz likely pulls much more. If you take 8c Matisse near its clockspeed limits, I expect the current draw to go even higher, not to speak of the 12c and 16c chips. If you're satisfied with not being at the chip's clockspeed limits, then I would say a 2700x @ 4.0 GHz would probably pull less current than my 1800x and an 8c Matisse @ 4.3 GHz would pull less current than a 2700x. So it isn't all bad.

Overclocking is never guaranteed.

However, if all one wants to do is operate at stock speeds I do not expect any problems from b450/x470 motherboards. I'm running x265 encoding on 1800X paired with Asrock B450M Pro4 board which is a 3x2 design, and it's boosting all the way to it's 3.6GHz limit on all cores without problems. I fully expect that a good X470 board with true 6 phases will have no problem overclocking Zen2 chip.

 

[Christopher Bohling]

I have an NH-d15 in my current machine, and I had an NH-d14 previously. It, like most "L-shaped" towers, does very little to cool board components. Top-down coolers will help keep VRM sinks saturated with air. My solution to the problem was: case airflow. My old strategy involved taking off the side of the case and pointing a desk fan sort of at parts of the motherboard that I thought needed it. Mostly I targeted the RAM, but it did help the VRMs too. Then I went a little bit less hackjob-ish and just got a case with a huge side panel fan that moves air pretty evenly across the entire board. VRMs on my x370 Taichi stay nice and cool even when my 1800x pulls over 140a (which it easily does). Of course it helps that the board has more phases than it needs.

Eh, buying a new case just so I can keep my old motherboard kinda seems like a nonstarter solution, if that's what I had to do I'd rather just buy in and get an X570 board. In any case my board is not 4+2, it's 6 true phases for the CPU + 2 true/2 doubled phases for the SoC. So hopefully I'll be able to push a little further than a 160A limit.

 

[DrMrLordX]

Eh, buying a new case just so I can keep my old motherboard kinda seems like a nonstarter solution, if that's what I had to do I'd rather just buy in and get an X570 board. In any case my board is not 4+2, it's 6 true phases for the CPU + 2 true/2 doubled phases for the SoC. So hopefully I'll be able to push a little further than a 160A limit.

Then you are in slightly better condition than someone trying to run a 4+2 config.

You can rig extra cooling for the VRMs if you like. People do this all the time on B350 and B450 boards. My desk fan solution cost me $25. It was kind of . . . ghetto/trailer park but it worked pretty well.

@fleshconsumed

No, overclocking isn't guaranteed, but the point is that choosing a weaker VRM config puts a very real limit on how high your processor runs before you overload the VRMs and force throttling. Even if you don't reach those limits, increased ripple from poor voltage regulation and vdroop/voltage spikes puts other, less-severe limits on what clockspeeds you can reach, before even thinking about whether you have a golden chip or a dud.

An 1800x "boosting" to 3.6 GHz on all cores is already 100 MHz below it's normal all-core turbo (mine runs 3.7 GHz all day long @ default running Prime95, y-cruncher, you name it). The difference in current draw between 4 GHz and 3.6 or 3.7 GHz is pretty staggering, so it's no wonder that it hums along pretty well at that clockspeed. Regardless, there's almost no reason to have an 1800x in that situation, when a 1700 could do the same thing (might require some P-state overclocking but still). More-recent AMD chips have far more complicated boost options that just won't work if the VRMs don't play ball. If someone tried putting a 3950x on a B450 board, I suspect it'll slum around at or near 3.5 GHz on something like x265 encoding. It might actually go lower than that if the current draw from AVX2 pushes the VRMs too hard. On a board with a healthy VRM config, it should run all-core of 3.8 GHz or higher, especially with good cooling and PBO/XFR/whatever that let it move outside of it's 105w power limit.

 

[Christopher Bohling]

Then you are in slightly better condition than someone trying to run a 4+2 config.

You can rig extra cooling for the VRMs if you like. People do this all the time on B350 and B450 boards. My desk fan solution cost me $25. It was kind of . . . ghetto/trailer park but it worked pretty well.

If I mounted a top fan as intake to point it straight at the VRM heatsinks, would that work? Would it mess up airflow for the rest of the case too much?

 

[lightmanek]

1). "6 phases" often means 4+2 configs. 4 for the CPU, 2 for everything else (SoC functions, RAM controller). You already have to tone down your maximum current limits to 160a. For reference, my R7-1800x @ 4.0 GHz can pull peak current of 152a in an AVX workload. A 2700x @ 4.3 GHz likely pulls much more. If you take 8c Matisse near its clockspeed limits, I expect the current draw to go even higher, not to speak of the 12c and 16c chips. If you're satisfied with not being at the chip's clockspeed limits, then I would say a 2700x @ 4.0 GHz would probably pull less current than my 1800x and an 8c Matisse @ 4.3 GHz would pull less current than a 2700x. So it isn't all bad.

2). Current ratings for VRM components are somewhat fudged. Many components are rated @ 25C which never happens during normal operation. Ratings can go down by as much as 20-25% by the time the VRM temps reach 80-100C. Higher-quality VRM components seem better able to maintain their ratings at higher temps.

3). Duty cycle matters! VRMs can be switched on or off as they're needed. The more phases you have, the longer phases stay switched "off", which lets them run at cooler temperatures without need for additional cooling (fans, etc.).

4). Voltage overshoot is a thing. Quality VRM layouts (twinned vs doubled, or even better "true" VRM layouts that require neither twinning nor doubling, as seen on the high-end Gigabyte x570 boards) can do a lot to mitigate voltage overshoot. In contrast, if you have a sparse VRM config that's being pushed to its limits, vdroop and overshoot are only going to be exaggerated as the VRMs struggle to maintain basic operational parameters.



If you're okay seeing it stay down around 3.5 GHz on anything that uses the majority of the cores (which is why you bought the chip, right?) then . . . have at it!



The 4+2 boards need active airflow over the VRMs already with 8c chips.



I have an NH-d15 in my current machine, and I had an NH-d14 previously. It, like most "L-shaped" towers, does very little to cool board components. Top-down coolers will help keep VRM sinks saturated with air. My solution to the problem was: case airflow. My old strategy involved taking off the side of the case and pointing a desk fan sort of at parts of the motherboard that I thought needed it. Mostly I targeted the RAM, but it did help the VRMs too. Then I went a little bit less hackjob-ish and just got a case with a huge side panel fan that moves air pretty evenly across the entire board. VRMs on my x370 Taichi stay nice and cool even when my 1800x pulls over 140a (which it easily does). Of course it helps that the board has more phases than it needs.

I will only say this to you - I ran Cinebench R15 on this board with my 2700X at 4.25GHz and 1.45V with power draw for CPU well above 250W (390W system power consumption).

No active cooling of VRMs and board handled it fine.
This is 6+2 design and I will repeat it one more time, it will run any certified CPU stock just fine and will have some OC headroom left to spare.

No one here is questioning that newer X470 or X570 boards have better VRM design, but we are talking about VRM being better for serious WC, DI or LN2 overclocking, not for your average +200MHz daily OC.

BTW if you think you need 160A for 3.5GHz on all cores on even 14nm ... my 12C Threadripper takes way less than that!
I have electronic engineering education and some marketing will not sway me from the facts.

Best regards

 

[fleshconsumed]

@fleshconsumed

No, overclocking isn't guaranteed, but the point is that choosing a weaker VRM config puts a very real limit on how high your processor runs before you overload the VRMs and force throttling. Even if you don't reach those limits, increased ripple from poor voltage regulation and vdroop/voltage spikes puts other, less-severe limits on what clockspeeds you can reach, before even thinking about whether you have a golden chip or a dud.

An 1800x "boosting" to 3.6 GHz on all cores is already 100 MHz below it's normal all-core turbo (mine runs 3.7 GHz all day long @ default running Prime95, y-cruncher, you name it). The difference in current draw between 4 GHz and 3.6 or 3.7 GHz is pretty staggering, so it's no wonder that it hums along pretty well at that clockspeed. Regardless, there's almost no reason to have an 1800x in that situation, when a 1700 could do the same thing (might require some P-state overclocking but still). More-recent AMD chips have far more complicated boost options that just won't work if the VRMs don't play ball. If someone tried putting a 3950x on a B450 board, I suspect it'll slum around at or near 3.5 GHz on something like x265 encoding. It might actually go lower than that if the current draw from AVX2 pushes the VRMs too hard. On a board with a healthy VRM config, it should run all-core of 3.8 GHz or higher, especially with good cooling and PBO/XFR/whatever that let it move outside of it's 105w power limit.

3.64GHz if we're splitting nickels, 3.64, 3.7, is 60Hz going to make a difference? For all I know it could be a rounding error, or it could be the fact that I'm running it on Windows Server and not Windows 10, or it could be a myriad of other things. I plan on playing with my hardware this weekend, I'll put same 1800x chip in x370 motherboard with beefy VRMs to see if it'll make a difference and report back. I don't expect to see any but I'll report either way.

I'm not arguing that overclocking does not exponentially increase power draw, it does, it's a well known fact. However, at stock AMD Zen CPUs are incredibly power efficient (unlike their GPUs) and I expect zero problems from B450/X470 motherboards running 3000 series at stock XFR/PBO frequencies.