PSU for a new build

[kiriakost]

Can you ?

This question makes me wonder if my profile information this appears only to me.

https://www.techpowerup.com/forums/members/kiriakost.63752/#about

 

[Vya Domus]

This question makes me wonder if my profile information this appears only to me.

https://www.techpowerup.com/forums/members/kiriakost.63752/#about

Thought so.

 

[Zach_01]

The CPU will peak at around 150W (if AMD are to be believed about the exact same power draw as Zen2) and we'll need to wait on reviews to see what the power consumption of either the 3070Ti or 6900XT are. If the 3070Ti is a cut down 3080 then you may be out of luck with your 550W. You should add your peak CPU and GPU power draw together - AS TESTED, not as claimed by TDP figures - and make sure that the PSU can handle that. As others have said it's extremely unlikely that you'll ever be able to pull peak power out of both at once, but running an older PSU beyond 100% of its rating is just asking for component failure in the PSU, even in a high-quality model like your Seasonic.

Thats a 142W if exactly the same as ZEN2. As I said the system will be around 400W avg and 450W peak with a 250W GPU. Its is within PSU specs but still its a 75~80% load.
I personally wouldnt run that for long... As a matter of fact... at all. Especially if that PSU has a few hundread (or more) hours on its back.

 

[AusWolf]

I'd have to say upsize it a little or by however much you want really.

The one you have now should be able to run it but like another has said you're close to the edge. Also by upsizing the next build could probrably use it without worry, this way you'd not have to buy another one the next time or two you upgrade. I don't know how often you do but that's an eventuality, many do it every two to three years and a good PSU will last way longer than that.

Just make sure if you do replace, to get one with alot of cables available for different things to accomidate what you may have later without worrying if you have the right cable or enough of them for your planned build. Modular PSU's are good for this, just add what you want and leave out what you don't.

Any good quality 750W PSU should hold you over for this and the next build too at least, plus you won't be right at the edge of what it needs to run. I've noted myself over time it seems setups are needing more and more power overall to run, the newest Nvidia card is an example of this and no telling what future cards will require regardless of who makes it.

That's my take on it.

Personally I've NO worries about that since I'm running a modular 1600W in this one, no way it's gonna have an issue about being too small for a very long time to come.

To be honest, my way of thinking has changed a lot recently from "mid-tier will always be fine for 1080p gaming" all the way to "I want the best I can afford to make sure I won't have to upgrade so often". But even if I don't upgrade for a good 3-4 years, future compatibility will always be a thing, and the PSU isn't something you'd want to think about just for an upgrade - you'd want it to serve you as long as its lifespan allows.

Now that I've read a bunch of opinions and suggestions, I think I have a plan:
1. I'll wait for reviews (as I intended anyway), and see if actual power consumption figures necessitate a PSU swap.
2. I'll try using the current PSU if I can, and save up for a replacement whether it works or not.

Speaking of replacement, what do you guys recommend? I love my Prime Ultra Platinum, but it's a bit on the expensive side in the 750 W version. Would I be alright with a Focus GX? As far as I can tell, the only difference between that, and the Prime line is the fan size, which is a bit confusing.

 

[Zach_01]

Speaking of replacement, what do you guys recommend? I love my Prime Ultra Platinum, but it's a bit on the expensive side in the 750 W version. Would I be alright with a Focus GX? As far as I can tell, the only difference between that, and the Prime line is the fan size, which is a bit confusing.

I cannot recomend you a PSU (Brand) as others know better, but I'm certain that you dont need a platinum one. A Gold and 750W will be plenty.

 

[AusWolf]

I cannot recomend you a PSU (Brand) as others know better, but I'm certain that you dont need a platinum one. A Gold and 750W will be plenty.

I've looked at the Prime GX 750 W as well, which is a good £50 more expensive than the Focus GX equivalent. Sure, it's got a bigger fan, but I really don't know if it's worth that much of an extra.

Edit: Or maybe an EVGA Supernova 750 GT? All of these are 80+ Gold models.

 

[Chrispy_]

Thats a 142W if exactly the same as ZEN2. As I said the system will be around 400W avg and 450W peak with a 250W GPU. Its is within PSU specs but still its a 75~80% load.
I personally wouldnt run that for long... As a matter of fact... at all. Especially if that PSU has a few hundread (or more) hours on its back.

Yes and no. I'd round it up to 150W for safety but I really am spitballing numbers here without experience of the B550M TUF mentioned. 150W is a ballpark figure for a typical board on "auto" settings.

142W is the MINIMUM package power, along with 90A TDC and 140A EDC limits for a 105W-rated chip, as per AMD's specs for boosting. Most motherboards provide more headroom than this for PBO and many motherboards (Asus are one of the main offenders) apply aggressive (or lazy, depends how you look at it) voltages using "auto" settings even when PBO isn't enabled - all of which serves to increase the peak power draw. On MSI B450 boards I actually had to tweak quite a lot of settings to get 3900X chips running at "stock" and not overclocked in any way.

The only real way to know is to find a review of that exact board that includes the same class of chip and has compreshensive power consumption figures, then deduct the (hopefully known) Idle GPU power draw to get your answer. Failing that, get a kill-a-watt meter and test yourself once you have the hardware in hand.

Normally I wouldn't care too much about +/-25W because most builds have plenty of headroom in the PSU, so it's the difference between the PSU operating at 60% load or 65% load there but given that the OP wants to run a potentially 300W+ graphics card with a PSU that's 200W less than the recommended Wattage, I guess every little bit matters. Certainly a 3900X with a RTX3080 would fail to boot on his 550W PSU - quite a few mainstream reviewers commented on having to upgrade their test-bench PSU in order to get their 3080 and 3090 reviews done because of the insane peak draw of those cards.

 

[Rei]

Spare your expenses if you are sticking with your currently planned build. Your current PSU is still good for your upcoming rig. Best to upgrade the PSU on your next upgrade cycle when changing to a more power-hungry component and have lots of USB and other peripherals plugged in. No need to blow over the budget for something that is not yet necessary.

 

[Zach_01]

Yes and no. I'd round it up to 150W for safety but I really am spitballing numbers here without experience of the B550M TUF mentioned. 150W is a ballpark figure for a typical board on "auto" settings.

142W is the MINIMUM package power, along with 90A TDC and 140A EDC limits for a 105W-rated chip, as per AMD's specs for boosting. Most motherboards provide more headroom than this for PBO and many motherboards (Asus are one of the main offenders) apply aggressive (or lazy, depends how you look at it) voltages using "auto" settings even when PBO isn't enabled - all of which serves to increase the peak power draw. On MSI B450 boards I actually had to tweak quite a lot of settings to get 3900X chips running at "stock" and not overclocked in any way.

142W is the limit for the 105W TDP CPUs, and you cant see a figure past, even with PBO on and CPU power management settings on Auto, unless you cool the CPU really low (like <60C). But I completely forgot the "PowerReportDeviation" issue. The subject that almost every board lies about its power draw to the CPU itself. So, yes partially you are right. It could reach 150W true power.

 

[Bones]

Alot of different takes on how to go about it here and TBH none are really "Wrong" so much but it's the fact your setup is/will be close to the edge that bothers me with the current PSU, esp since it has a bit of age on it.

All I'm wanting to do is suggest something that will work, not be too expensive, have a minimum of worry with it too so you'll at least be able to use the system once built up until the time to upgrade comes around again.

I normally spec things to about 80% of it's power rating as a nominal figure with all the expected wattage load and go from there to accomidate for any load spikes the system may and will have at times. This way it doesn't make the PSU itself run under what would be a heavily loaded condition all the time - That will place unneccesary strain on the PSU and tend to shorten it's life.

The big thing here to me is to be sure once the system is built you won't run into problems with it and also to help avoid a sudden PSU failure which can and normally will take out the rest when it goes.

It may not happen for sometime to come but once it does it's already too late and expensive to deal with, certainly costing more than the extra spent on one that's little higher in wattage rating. I'm also aware they can even go out to the point of flaming death and that's never a good thing for it and your own safety.

Just ask @Mr.Scott about that sometime - He'll tell you what happened with one of his when it did just that.

Your call.

 

[AusWolf]

Alot of different takes on how to go about it here and TBH none are really "Wrong" so much but it's the fact your setup is/will be close to the edge that bothers me with the current PSU, esp since it has a bit of age on it.

All I'm wanting to do is suggest something that will work, not be too expensive, have a minimum of worry with it too so you'll at least be able to use the system once built up until the time to upgrade comes around again.

I normally spec things to about 80% of it's power rating as a nominal figure with all the expected wattage load and go from there to accomidate for any load spikes the system may and will have at times. This way it doesn't make the PSU itself run under what would be a heavily loaded condition all the time - That will place unneccesary strain on the PSU and tend to shorten it's life.

The big thing here to me is to be sure once the system is built you won't run into problems with it and also to help avoid a sudden PSU failure which can and normally will take out the rest when it goes.

It may not happen for sometime to come but once it does it's already too late and expensive to deal with, certainly costing more than the extra spent on one that's little higher in wattage rating. I'm also aware they can even go out to the point of flaming death and that's never a good thing for it and your own safety.

Just ask @Mr.Scott about that sometime - He'll tell you what happened with one of his when it did just that.

Your call.

It sounds a little bit extreme, but of course everything's possible. I'll try to squeeze a new PSU into the budget, or wait with buying the graphics card until I have the money for both the GPU and PSU... or go with the original plan: try my luck with the current one, and start saving for a new one anyhow. We'll see.

Dominator is expensive RAM, and if you're going for premium RAM then you might as well get the best speed for Ryzen - 3600 isn't much more expensive and the extra 200MHz on the infinity fabric is worth more performance than fancy heatspreaders or tighter timings at the lower 3200MHz speed.

I've just spent a little bit of time looking into this. As far as I see, 3200 MHz is more widely available in my area, and is also fairly cheaper (depending on the model of course). On the other hand, the difference in performance appears to be marginal. It seems to me like the "nvme vs SATA" situation: even though one is faster than the other on paper, you have to focus hard to spot any real world difference. Although, I've never owned a Ryzen 3000 system, so feel free to correct me if I'm wrong.

 

[Bones]

Doesn't have to be anything way over the expected wattage useage, just something you know will handle it if anything will.
Even a good quality 650W is better than what you've got ATM so you're not constantly bumping the ceiling with what it can handle.

Just add up the expected total wattage (Nominal) useage and figure what 10% of that would be, then simply double it to get the 20% cushion value you'd want.

Add this 20% to the original expected useage (Nominal) value and that will give an approximate wattage to get in a PSU and that's all you'd need to worry about.

 

[AusWolf]

Doesn't have to be anything way over the expected wattage useage, just something you know will handle it if anything will.
Even a good quality 650W is better than what you've got ATM so you're not constantly bumping the ceiling with what it can handle.

Just add up the expected total wattage (Nominal) useage and figure what 10% of that would be, then simply double it to get the 20% cushion value you'd want.

Add this 20% to the original expected useage (Nominal) value and that will give an approximate wattage to get in a PSU and that's all you'd need to worry about.

Interesting recipe. What I'd do is: CPU TDP + GPU TDP + 100 W (for the rest of the system and a little headroom). Based on pre-launch estimates, this gives me 505 Watts, so my Seasonic may or may not be OK... hence my OP. With your method it's 105 W + 300 W + 20% = 486 W. Again, it's close to the ceiling, which may end up OK, or may not.

 

[Bones]

Interesting recipe. What I'd do is: CPU TDP + GPU TDP + 100 W (for the rest of the system and a little headroom). Based on pre-launch estimates, this gives me 505 Watts, so my Seasonic may or may not be OK... hence my OP. With your method it's 105 W + 300 W + 20% = 486 W. Again, it's close to the ceiling, which may end up OK, or may not.

That's correct but also take into account the age of the PSU, that does matter to an extent.

You can exceed that value if you want to accomidate the age of the current unit. I know any PSU technically can handle spikes of 100%+ of it's value but that will change with age of course as you'd expect.
With that even a 600W unit would be the ticket here as long as it has all the connectors you need.

 

[Chrispy_]

Interesting recipe. What I'd do is: CPU TDP + GPU TDP + 100 W (for the rest of the system and a little headroom). Based on pre-launch estimates, this gives me 505 Watts, so my Seasonic may or may not be OK... hence my OP. With your method it's 105 W + 300 W + 20% = 486 W. Again, it's close to the ceiling, which may end up OK, or may not.

I'm curious where you're getting TDP figures for the unannounced 3070Ti and remember that the TDP figures for the 3000-series are not accurate to their real-world power draw.

I've just spent a little bit of time looking into this. As far as I see, 3200 MHz is more widely available in my area, and is also fairly cheaper (depending on the model of course). On the other hand, the difference in performance appears to be marginal. It seems to me like the "nvme vs SATA" situation: even though one is faster than the other on paper, you have to focus hard to spot any real world difference. Although, I've never owned a Ryzen 3000 system, so feel free to correct me if I'm wrong.

For Zen2 the gaming performance increase moving from 3200 C16 to 3600 C16 is about 4-5%, same for encoding. It doesn't make much difference to rendering though.

Take the total cost of your system upgrade, (probably ~$1000) add 4-5% (so $40-50) and if the DDR4-3600 is less than a $40-50 premium then it is both faster and better value for money.

Don't get me wrong, 3200MHz RAM is fine but you're leaving some performance on the table by not picking it. Even if you have to drop down to a cheaper kit with looser timings, it's worth doing. Expensive 3200 CL14 is still inferior for Zen2 (and presumably Zen3) than a cheapo 3600 CL17 kit because It's not about RAM bandwidth or RAM latency, it's about running your CPU's infinity fabric* at its maximum clock, which is still going to be of huge importance on Zen3 if you're using a CPU with two chiplets (so the 12-core and 16-core variants). Zen 3's change to CCX cache means that 8-core and 6-core models won't be as sensitve to RAM speeds for Zen3 as they are with Zen2.

* - Edit: In case you aren't aware, your CPU's infinity fabric is locked to the RAM speed. You can't run the Infinity fabric at it's maximum (official) 1800MHz whilst using DDR4-3200 that only runs at 1600MHz. The 1600MHz RAM clock will drag your CPU's internal fabric clock (FCLK) down by 12% and whilst this is only a minor performance hit, it's still a performance hit.

 

[AusWolf]

I'm curious where you're getting TDP figures for the unannounced 3070Ti

I used the preliminary data on the RX 6900 XT page of TPU's GPU database. Needless to say that everything is speculation as of now.

remember that the TDP figures for the 3000-series are not accurate to their real-world power draw.

True. I might be better off postponing on the graphics card a couple months, and buying it together with a bigger PSU. I will also have more (real world) data to go by at that point.

 

[Bones]

I'm curious where you're getting TDP figures for the unannounced 3070Ti and remember that the TDP figures for the 3000-series are not accurate to their real-world power draw.

Have to agree - You have to know all that before doing a final figuring of the expected wattage draw to get it right. Was going on the values the OP provided but if that's not accurate then perhaps a minimum 100W cushion would do based on the proposed build.
Whatever it winds up being is what to base it all on.

 

[AusWolf]

I'm curious where you're getting TDP figures for the unannounced 3070Ti and remember that the TDP figures for the 3000-series are not accurate to their real-world power draw.


For Zen2 the gaming performance increase moving from 3200 C16 to 3600 C16 is about 4-5%, same for encoding. It doesn't make much difference to rendering though.

Take the total cost of your system upgrade, (probably ~$1000) add 4-5% (so $40-50) and if the DDR4-3600 is less than a $40-50 premium then it is both faster and better value for money.

Don't get me wrong, 3200MHz RAM is fine but you're leaving some performance on the table by not picking it. Even if you have to drop down to a cheaper kit with looser timings, it's worth doing. Expensive 3200 CL14 is still inferior for Zen2 (and presumably Zen3) than a cheapo 3600 CL17 kit because It's not about RAM bandwidth or RAM latency, it's about running your CPU's infinity fabric* at its maximum clock, which is still going to be of huge importance on Zen3 if you're using a CPU with two chiplets (so the 12-core and 16-core variants). Zen 3's change to CCX cache means that 8-core and 6-core models won't be as sensitve to RAM speeds for Zen3 as they are with Zen2.

* - Edit: In case you aren't aware, your CPU's infinity fabric is locked to the RAM speed. You can't run the Infinity fabric at it's maximum (official) 1800MHz whilst using DDR4-3200 that only runs at 1600MHz. The 1600MHz RAM clock will drag your CPU's internal fabric clock (FCLK) down by 12% and whilst this is only a minor performance hit, it's still a performance hit.

Fair enough. I thought 3200 MHz was the fastest officially supported speed (and thus 1600 MHz fclk), anything above was overclock.

 

[Chrispy_]

Fair enough. I thought 3200 MHz was the fastest officially supported speed (and thus 1600 MHz fclk), anything above was overclock.

That, I think, is because the fastest official JEDEC standard for DDR4 is 3200, there is no such thing as JEDEC 3600. Technically, every DDR4-3200 kit is also an overclock because the XMP timings will be CL16 or similar - the official JEDEC timings are a ridiculously-loose CL22 or even CL24 I think....

At Zen2 launch, AMD put on several presentations about the architecture and went into detail about the fabric speeds and RAM divider. The 1:1 divider between RAM and FCLK automatically drops to 1:2 at DDR4-3733, so AMD officially caps its own FCLK at 1833MHz and are on record recommending 3600MHz DDR4 and 1800MHz FCLK as the sweet spot for Zen2 performance.

It turns out that if you bought a bargain-basement 3600 or 3100 that's low-yield in the silicon lottery, you may find that 1800FCLK isn't actually possible and you have to drop down to 1766MHz - but that's also usually a factor of cheap motherboard and cheap power supply all playing their part too. Realistically, the Zen2 I/O die will handle 1800MHz+ as long as you don't buy low-end parts and that's why AMD set the FCLK:RAM divider to cap FCLK at 1800MHz. You can override this AMD-official divider with many motherboards to push your FLCK beyond 1800MHz, but up to 1800 FCLK is 100% 'stock' as far as AMD are concerned.

 

[AusWolf]

That, I think, is because the fastest official JEDEC standard for DDR4 is 3200, there is no such thing as JEDEC 3600. Technically, every DDR4-3200 kit is also an overclock because the XMP timings will be CL16 or similar - the official JEDEC timings are a ridiculously-loose CL22 or even CL24 I think....

At Zen2 launch, AMD put on several presentations about the architecture and went into detail about the fabric speeds and RAM divider.

The 1:1 divider between RAM and FCLK automatically drops to 1:2 at DDR4-3733, so AMD officially caps its own FCLK at 1833MHz and are on record recommending 3600MHz DDR4 and 1800MHz FCLK for Zen2.

It turns out that if you bought a bargain-basement 3600 or 3100 that's low-yield in the silicon lottery, you may find that 1800FCLK isn't actually possible and you have to drop down to 1766MHz - but that's also usually a factor of cheap motherboard and cheap power supply all playing their part too. Realistically, the Zen2 I/O die will handle 1800MHz+ as long as you don't buy low-end parts and that's why AMD set the FCLK:RAM divider to kick in at 1800MHz.

It makes a lot more sense now. So basically, 3200 MHz RAM / 1600 fclk is the fastest officially supported and guaranteed speed on all Zen 2 CPUs, even though 3600/1800 MHz technically works with most chips, right? Then I guess they class 3600 MHz memory as overclock just to avoid a potential lawsuit coming from 0.1% of the users who couldn't make it work with their 3100X CPUs (again, correct me if I'm wrong).

 

[xtreemchaos]

i allways like to have a high wattage PSU i run a 850w its loads morethan i need near double but its better to have it and not need it than to need it and not have it same rules as a condom . + ive a hybred and the fan only comes on when im pulling 500w or more is which almost never.

 

[tabascosauz]

It makes a lot more sense now. So basically, 3200 MHz RAM / 1600 fclk is the fastest officially supported and guaranteed speed on all Zen 2 CPUs, even though 3600/1800 MHz technically works with most chips, right? Then I guess they class 3600 MHz memory as overclock just to avoid a potential lawsuit coming from 0.1% of the users who couldn't make it work with their 3100X CPUs (again, correct me if I'm wrong).

Yeah, that's just about the number of people having trouble sustaining 1800MHz FCLK. At launch (when quad-core Zen 2 didn't exist), there were some early production 3600 chips that had trouble running 1800; I would imagine that with the new positioning, the 3100 is the only SKU that's possibly more vulnerable than the rest.

I would probably make a distinction between the on-I/O die memory controller and the on-substrate Infinity Fabric. Whatever Infinity Fabric connections exist on the chiplets themselves aren't the issue, and neither is the memory controller (which is plenty capable of 4000MT/s on its own, even in lower SKUs); the problem exists with trying to get the routing located on the substrate itself up to speeds beyond 1800/1833/1866/1900/1933MHz.

The other issue there is that when an average users encounters "difficulty running a 3600MT/s kit", there exists a billion other places from which the instability can stem. Which memory ICs they use and the timings settled on by the board outside of the few XMP-dictated ones, the binning of the ICs within that die revision type alone, appropriate DRAM voltage, appropriate SOC voltage for the speed, stock procODT settings dictated by the board, etc. So you'd be able to verify whether an inability to run FCLK 1800MHz is actually just that, only after hours or days of testing that your average Joe would never undertake. You'll never know if those users really had a CPU incapable of running FCLK 1800, or if there were other issues at play.

If you're looking towards an 8-core or dual-chiplet CPU, I don't see there being a single issue with setting your sights on a decent 3600CL16 kit. Even with 3200MT/s XMP kits, full stable compatibility is hardly guaranteed (just take a look around the web at all the troubleshooting threads); there are countless things that can come in the way of RAM stability, none of which are 1:1 FCLK speed concerns.

The only higher speed kits that I would consider to be largely fail-safe from everything would be less common 3200MT/s kits that run at CL22 timings at strictly the baseline DDR4 1.2V. These generally don't rely on XMP profiles to hit those speeds out of the box, and are IFAK covered by JEDEC 3200. As you can probably tell, you probably don't want these for their performance.

 

[rbgc]

Recomended PSU Table for GPU a CPU (Source: ASUS, Manuals, PSU)

 

[300BaudBob]

For anyone interested in why keeping a power supply for many years is not really a great idea this PDF from NASA is worth a read.

 

[Selaya]

The system is very unlikely to load both the CPU and GPU to the max at the same time - games that are CPU (multithread) bound/bottlenecked, if they do exist and max the CPU will in all likeliness drop the GPUs power, and vice versa.
I would assume that both the 3070 and the 5950X's power consumption will be within the ballpark of the 2080Ti and 3900X/3950X's, and W1zzard was kind enough to do total power consumption tests for the latter - the system will max out at about 400W draw at the wall.
While you may want to consider ordering a higher-specced PSU (if you don't already have one), the 550W-rated one will do just fine - it's certainly not worth spending another $200+ on a new PSU just for like, a 10% efficiency upgrade at 400W - you'd have to run your system at close to maximum power drawn for several years straight for your PSU investment to amortise.

Also, a 550W PSU is probaly safe enough to operate even for at 600-650W for a short timeframe in the unlikely event that your system really maxes out both the CPU and the GPU's power draw for whatever reason.