RTX 3080 power supply requirements - 850W?

[Space Lynx]

The TDP really tells little about the power:

The AMD Ryzen 9 3950X Review: 16 Cores on 7nm with PCIe 4.0

www.anandtech.com

.
My guess is it will be comparable to 3950X, so this maybe a good reference.

I think I am going to aim for 5600x. going to skip 5900x otherwise i might have to get a new psu. i have 700w gold evga

 

[londiste]

The TDP really tells little about the power:

Ryzen 3000 power limit is at 135% TDP in generic conditions. Anandtech has a nice page in 3000-series review explaining the complexities but 135% is generally correct. CPU does not have to pull that much power but that is the limit and heavy loads are reaching that.
105W TDP = 141W Power Limit. 65W TDP = 88W Power Limit.

 

[R-T-B]

I think I am going to aim for 5600x. going to skip 5900x otherwise i might have to get a new psu. i have 700w gold evga

I really don't see any one single GPU and CPU using more than 700W right now.

 

[rbgc]

Asus recomended PSU table for NVIDIA GPU and CPU (source: ASUS, manuals, PSU)

 

[NesteaZen]

I never know what ryzen 7 and ryzen 9 is.

 

[rbgc]

Most power supplies hit their peak efficiency levels with loads in the range of 40 to 80 percent. Building to about 50 to 60 percent of a PSU's capacity is advisable to achieve maximum efficiency and yet leave room for future expansion and overclocking.

Good idea is also to see fan noise curve because some PSU and also GPU can switch fan off.

For example

 

[R-T-B]

Asus recomended PSU table for NVIDIA GPU and CPU (source: ASUS, manuals, PSU)

View attachment 173149

to be fair, that's assuming a pretty garbage PSU. There is no way you actually need a well built PSU that high up.

Most power supplies hit their peak efficiency levels with loads in the range of 40 to 80 percent. Building to about 50 to 60 percent of a PSU's capacity is advisable to achieve maximum efficiency and yet leave room for future expansion and overclocking.

The difference is in practice usually in the realm of 1-3%. So no, it's really not sensible given the cost difference. It's similar to when I was mining, and tested 230V vs 120V. You gain so little for your trouble.

 

[rbgc]

to be fair, that's assuming a pretty garbage PSU. There is no way you actually need a well built PSU that high up. The difference is in practice usually in the realm of 1-3%. So no, it's really not sensible given the cost difference. It's similar to when I was mining, and tested 230V vs 120V. You gain so little for your trouble.

To be fair, I don't understand obsession with PSU and count every watt to choose PSU "for GPU or CPU". PSU is component for years and reused in new builds.

 

[EarthDog]

To be fair, I don't understand obsession with PSU and count every watt to choose PSU "for GPU or CPU". PSU is component for years and reused in new builds.

right, but there is a difference in cost for buying a properly sized model versus overkill. This 50-60% capacity thing is a waste of money. As was said the difference in efficiency is AT MAX 1-3%. Hardly noticeable iver A YEAR if you do the math. Yet, if we go by the ~50% rule, you're paying $50 to $100 more dollars on a psu for no reason (well, quiet perhaps). So yeah, you can disregard watts and over buy to be 'efficient'.. its your wallet.

FTR, I'm running an i9-10980xe with all c/t at 4.5 ghz with a rog strix 3080... no issues here with a quality 750w unit. I load around 550W while gaming.

 

[Bill_Bright]

Most power supplies hit their peak efficiency levels with loads in the range of 40 to 80 percent. Building to about 50 to 60 percent of a PSU's capacity is advisable to achieve maximum efficiency and yet leave room for future expansion and overclocking.

Nah! Many MANY years ago, before "flat" efficiency curves and Plug Load Solutions came up with their 80PLUS certification process, that was true. But as ED suggests, not today.

To be fair, I don't understand obsession with PSU and count every watt to choose PSU "for GPU or CPU". PSU is component for years and reused in new builds.

In a round about way, you are almost contradicting yourself.

Basic power supplies naturally have a bell shaped efficiency curve where X is the load and is typically between 50 and 70% of the total load capacity. That is the point where the efficiency peaks, then drops rapidly on either side. And "cheap" basic power supplies, even at peak efficiencies, have pretty lousy efficiencies. Peaking at 70% or even less is not uncommon.

Yet those type supplies are still widely used today with devices that pose a fairly consistent load on the power supply. A TV for example, presents a fairly consistent load on the supply. So engineers choose and match a supply where the typical demand falls near the power supply's peak efficiency range. Hopefully, they have a little tree-hugger in them and choose a better quality supply that peaks higher than 70% efficiency.

Computer power demands, of course, vary widely, from just a handful of watts at idle to 100s of watts when tasked. Hence the desire for a relatively "flat" efficiency curve. But not just a flat curve, but a relatively high efficiency rating too. Enter 80 PLUS certifications. If you look at the requirements for the various rating (hover over the 80 PLUS logos), you will see a maximum of just 4% difference between the 50% efficiency requirement and the 100% requirement. So as rbgc suggested, there just is no need to obsess over such tiny differences.

It would take many years of savings in energy costs to make up a couple percentage points. This is why paying more for Platinum or Titanium over Gold is rarely an efficient use of your $$$ - unless you just happen to stumble on a "too-good-to-pass-up" deal.

I agree with rbgc and when sizing up the PSU to purchase, you should plan ahead and think about possible upgrades that may require more power. So go ahead and pad the results to give you a little extra head room. But buying 200% above your needs is way overkill and as ED noted, is just wasting money.

These days, with a quality PSU, the only real reason to buy way more than you need is to ensure the PSUs fan barely spins or doesn't spin at all in order to minimize or eliminate any fan noise. But frankly, a quality PSU from a reputable maker will have a quality fan. And when installed in quality case is likely to be near silent (or totally silent - below ambient noise levels) anyway. So if you need 500W, don't waste your money on a 1KW supply. 600 - 650W will be plenty and still provide a nice buffer for future expansion/upgrades.

 

[R-T-B]

right, but there is a difference in cost for buying a properly sized model versus overkill. This 50-60% capacity thing is a waste of money. As was said the difference in efficiency is AT MAX 1-3%. Hardly noticeable iver A YEAR if you do the math. Yet, if we go by the ~50% rule, you're paying $50 to $100 more dollars on a psu for no reason (well, quiet perhaps). So yeah, you can disregard watts and over buy to be 'efficient'.. its your wallet.

FTR, I'm running an i9-10980xe with all c/t at 4.5 ghz with a rog strix 3080... no issues here with a quality 750w unit. I load around 550W while gaming.

As I said, I've yet to encounter a single CPU and GPU that will fully load a good 750W unit.

That includes HEDT.

 

[Sandbo]

I think I am going to aim for 5600x. going to skip 5900x otherwise i might have to get a new psu. i have 700w gold evga

Wasn't expect to upgrade, but then knowing Zen 3 are last on AM4, my X570 board asked me to get 5950X.

Ryzen 3000 power limit is at 135% TDP in generic conditions. Anandtech has a nice page in 3000-series review explaining the complexities but 135% is generally correct. CPU does not have to pull that much power but that is the limit and heavy loads are reaching that.
105W TDP = 141W Power Limit. 65W TDP = 88W Power Limit.

It might work for this group of CPUs, but then Anandtech also reviewed some 35W TDP CPUs and find pretty inconsistent results:

The $60 CPU Question: AMD Athlon 200GE or Intel Pentium Gold G5400? A Review

www.anandtech.com

To be fair, I don't understand obsession with PSU and count every watt to choose PSU "for GPU or CPU". PSU is component for years and reused in new builds.

Exactly and that's why I don't save on that; My Seasonic X-560 bought in 2011, has been running almost non-stop for 10 years, going through 2-3 systems.

 

[londiste]

It might work for this group of CPUs, but then Anandtech also reviewed some 35W TDP CPUs and find pretty inconsistent results:

The $60 CPU Question: AMD Athlon 200GE or Intel Pentium Gold G5400? A Review

www.anandtech.com

Athlon GE200 is Zen+ where power limit is at TDP. So are Ryzen 3000 mobile and APUs.
AMD increased power limit for Zen2 - Ryzen 3000 series and Ryzen 4000 APUs.

 

[Bill_Bright]

As I said, I've yet to encounter a single CPU and GPU that will fully load a good 750W unit.

Me either but... ...in reality whether or not the PSU is a "good" 750W unit or a sub-par unit really is immaterial when properly sizing a PSU for purchase. We must assume 750W is 750W even though one may be full of ripple, poorly regulated and out of tolerance. Without actual testing, we cannot assume a "good" 750W PSU can deliver more power than a sub-par 750W PSU with the same published specs on the measured rails. We can only assume the "good" PSU's output voltages will be cleaner and perhaps more stable with less wasted energy.

Of course, nobody wants instability cause by sub-par "dirty" power. I would never buy a brand new Porsche then fill it up with questionable fuel from the corner Tobacco and Bait Shop.

I think the problem is recommended vs actual needs. No table like that ASUS table, no GPU or graphics card maker, no calculator is ever going to recommend an underpowered PSU. They all pad the results. And that makes sense. And with the exception of the eXtreme OuterVision PSU Calculator, all those tables, GPU and card makers, and (AFAIK) all other calculators use arbitrary numbers for the other components the PSUs will power. Or at least, they typically don't state how many sticks and what type RAM, how many and what size case fans, if using a conventional CPU cooler or some alternative, how many and what type drives, etc. So they pad the results even more.

When sizing up a PSU for purchase, "IF" (and yes, now I am using a "what if") the maximum demand of the CPU and the maximum demand of the graphics solution added together total more than 750W (setting aside the demands of the mobo, RAM, drives, fans, etc.) we must assume it is possible both power hungry devices "might" max out demands at the same point in time, and make our purchase accordingly.

The problem is, we typically don't really know the true maximum power consumption of these components. And therein lies the rub. Without actual testing, we have to assume the manufacturers' published specs are fairly accurate.

So even though it is unlikely both the CPU and the graphics solution will reach peak demands at the same point in time, it is still possible. And the PSU must be able to provide it. And for that matter, we must assume at that same point in time, all fans will be spinning at full speed, all drives are being heavily accessed, all RAM are maxed out too.

If anything all this built-in padding/headroom just shows us is the advice to size your PSU to run at about "50 to 60 percent of a PSU's capacity" really makes no sense with today's PSUs and their relatively "flat efficiency curves". All the extra headroom is already in there. There's no need to double. The only exceptions I can see would be if you are seriously planning on adding a second graphics card in the near future or, maybe, moving from integrated graphics to a power hungry card.

***

Side question - anyone aware of any stress or benchmark program that fully taxes the CPU, GPU, RAM and drives at the same point in time? It seems to me an "artificial" scenario (like a stress or benchmark test) would be the only scenario where that might actually happen - but I am not aware of any that do that. Typical "real-world" scenarios tend to be CPU or GPU intensive, not both at the same time - at least not for more than a few clock-cycles at once.

 

[R-T-B]

Side question - anyone aware of any stress or benchmark program that fully taxes the CPU, GPU, RAM and drives at the same point in time?

I know there's a tool that does this, but the name is escaping me.

It is a pretty artificial scenario so I never used it.

 

[Sandbo]

I know there's a tool that does this, but the name is escaping me.

It is a pretty artificial scenario so I never used it.

I wonder if you could just run Prime95+Furmark at the same time.

 

[EarthDog]

I wonder if you could just run Prime95+Furmark at the same time.

Not furmark...please not furmark...

But maybe Unigine Superposition or something.... That will give you a good idea................see below.

BUt holy shiza, I am not planning my rig around 100% nameplate values. There ISN'T a situation, that I can think of), on a desktop PC where that happens. On boot, your HDDs spin up... but the CPU/GPU isn't at 100%, not close. Gaming, nope...HDDs are at rest most of the time and gaming loads on a CPU are nothing compared to rendering/prime 95, etc. So, just something simple like a GPU test and CPU test is plenty overkill and easily 'good enough'.

 

[R-T-B]

BUt holy shiza, I am not planning my rig around 100% nameplate values. There ISN'T a situation, that I can think of), on a desktop PC where that happens.

I can. But it isn't what you call a normal use case.

 

[EarthDog]

I can. But it isn't what you call a normal use case.

Thats my point. You can try to fake it... but why? There isnt a realistic use case (desktop) where the cpu, gpu, hdds, etc are full tilt....

 

[Rei]

Thats my point. You can try to fake it... but why? There isnt a realistic use case (desktop) where the cpu, gpu, hdds, etc are full tilt....

May be a good stresser for PSU and/or UPS power load benchmark....

 

[EarthDog]

May be a good stresser for PSU and/or UPS power load benchmark....

Thats a reason to do it... sure... but not a use case for a desktop.

 

[R-T-B]

Thats a reason to do it... sure... but not a use case for a desktop.

The only use case I have that comes close is mining, and I'm not sure that's "Desktop" though some Desktop users have certainly attempted it.

 

[GeeBee]

Hello to all the tech gurus here lol.

Didn't want to start a new thread for something trivial, so...

I would like to get into gaming (most likely sim racing), so I'm waiting for the dust to sort of settle down and (hopefully!) get this card for 4K 60Hz gaming (my monitor specs).

AORUS GeForce RTX™ 3080 MASTER 10G (rev. 1.0) Gallery | Graphics Card - GIGABYTE U.S.A.

Discover AORUS premium graphics cards, ft. WINDFORCE cooling, RGB lighting, PCB protection, and VR friendly features for the best gaming and VR experience!

www.gigabyte.com

My CPU is a 9700k and my Power Supply a Seasonic 650W Titanium (a few months old with minimal usage).

Do ya think my Power Supply will be up for it or do I need to go for the 850W equivalent?

Thanks for your time reading my 1st post.

 

[Rei]

Hello to all the tech gurus here lol.

Didn't want to start a new thread for something trivial, so...

I would like to get into gaming (most likely sim racing), so I'm waiting for the dust to sort of settle down and (hopefully!) get this card for 4K 60Hz gaming (my monitor specs).

AORUS GeForce RTX™ 3080 MASTER 10G (rev. 1.0) Gallery | Graphics Card - GIGABYTE U.S.A.

Discover AORUS premium graphics cards, ft. WINDFORCE cooling, RGB lighting, PCB protection, and VR friendly features for the best gaming and VR experience!

www.gigabyte.com

My CPU is a 9700k and my Power Supply a Seasonic 650W Titanium (a few months old with minimal usage).

Do ya think my Power Supply will be up for it or do I need to go for the 850W equivalent?

Thanks for your time reading my 1st post.

Welcome to TPU!
I think it would've been best had you started your own thread as you will find there will be alotta inputs here.

As for your question, your SeaSonic is a Titanium rating? Anyway, your PSU, especially if it's new somewhat new, should be good enough even for your CPU + RTX 3080.

 

[GeeBee]

Ah, ok then.

Noticed that Gigabyte recommends a 750W power supply for their 3080 - hence the ask.

Cheers