News Nvidia Blackwell RTX 50-series Founders Edition graphics cards — Details on the new design, cooling, and features of the reference models

[Admin]

The upcoming Nvidia RTX 50-series Founders Edition GPUs will have a new industrial design and aesthetics, along with improved cooling. Here are the details on the 5090, 5080, and 5070 Founders Edition graphics cards

Nvidia Blackwell RTX 50-series Founders Edition graphics cards — Details on the new design, cooling, and features of the reference models : Read more

 

[derekullo]

Any chance on the RTX 5070 Ti being a single slot card?

NVIDIA GeForce RTX 5070 Family Graphics Cards

Enable new experiences with the power of AI.

www.nvidia.com

	RTX 5070 Ti	RTX 5070
Slots	Varies by manufacturer	2-Slot

I have feeling that means 2-3 and not 1-2 ... but I have hope!

 

[JarredWaltonGPU]

Any chance on the RTX 5070 Ti being a single slot card?

NVIDIA GeForce RTX 5070 Family Graphics Cards

Enable new experiences with the power of AI.

www.nvidia.com

	RTX 5070 Ti	RTX 5070
Slots	Varies by manufacturer	2-Slot

I have feeling that means 2-3 and not 1-2 ... but I have hope!

Highly unlikely, considering it's a 300W part, but someone could try to do a single slot version. It would probably need to be liquid cooled with an external pump and radiator, though.

 

[Jabberwocky79]

"there are rumors that the 5090 at least uses liquid metal. Don't disassemble your graphics card, in other words."

I wonder how this is going to work with water-cooled solutions? I personally plan to water cool a 5080, so I'm anxious to know.

 

[Gaidax]

"there are rumors that the 5090 at least uses liquid metal. Don't disassemble your graphics card, in other words."

I wonder how this is going to work with water-cooled solutions? I personally plan to water cool a 5080, so I'm anxious to know.

Liquid Metal is no rumor, that's what they use there. We have GamerNexus video with Nvidia engineer working on cooling.

I'm actually super interested to see what FE delivers on that front.

 

[JarredWaltonGPU]

Liquid Metal is no rumor, that's what they use there. We have GamerNexus video with Nvidia engineer working on cooling.

I'm actually super interested to see what FE delivers on that front.

I'm not certain if that's only the 5090, I guess is the issue. And other brands will do whatever. But yeah, don't try taking apart a 5090 FE. LOL. Probably why they went to GN, since he likes to tear graphics cards apart.

 

[spongiemaster]

"there are rumors that the 5090 at least uses liquid metal. Don't disassemble your graphics card, in other words."

I wonder how this is going to work with water-cooled solutions? I personally plan to water cool a 5080, so I'm anxious to know.

The 5090 FE's triple PCB design looks like a no-go for water cooling. Alphacool is the only company that has announced waterblocks so far, and they did not announce an 5090 FE compatible block. If the 5080 uses the same design, then you're going to want to look at another model.

 

[JTWrenn]

I hate that we use TGP and TBP now. Don't think Nvidia reported it as TGP they reported it as TBP....that is brain melting. I believe the TBP is actually 360 W for a 5080 for example which means the TGP should be less....as it shouldn't include fans and such. I believe Nvidia announce the TBP is 360 so...yeah these get confusing

My understanding is TGP is just the board and components on it, not fans or leds. TBP is board plus fans and leds. So really the TBP is likely a few watts lower.

TDP though could be different and is deprecated but comparing to 3000 series gets hard there. Just brain melty to me. Can't wait till we get to just using one which should be TBP so you can compare stock to after market I believe. I could be totally wrong.

 

[palladin9479]

Any chance on the RTX 5070 Ti being a single slot card?

NVIDIA GeForce RTX 5070 Family Graphics Cards

Enable new experiences with the power of AI.

www.nvidia.com

	RTX 5070 Ti	RTX 5070
Slots	Varies by manufacturer	2-Slot

I have feeling that means 2-3 and not 1-2 ... but I have hope!

Any card can be a single slot card with a GPU water block on it. Though you might want to have the GPU mounting brace be two slots just to have two screws holding it to the case.

 

[palladin9479]

I hate that we use TGP and TBP now. Don't think Nvidia reported it as TGP they reported it as TBP....that is brain melting. I believe the TBP is actually 360 W for a 5080 for example which means the TGP should be less....as it shouldn't include fans and such. I believe Nvidia announce the TBP is 360 so...yeah these get confusing

My understanding is TGP is just the board and components on it, not fans or leds. TBP is board plus fans and leds. So really the TBP is likely a few watts lower.

TDP though could be different and is deprecated but comparing to 3000 series gets hard there. Just brain melty to me. Can't wait till we get to just using one which should be TBP so you can compare stock to after market I believe. I could be totally wrong.

TDP is still used, it's the amount of thermal energy, aka heat, that the cooling solution needs to be capable of dissipating. If your device is outputting 400W of thermal energy and your cooling can only move approximately 370W, then the device is going to eventually run out of thermal capacity and have to reduce it's thermal output. The reason for the "other" numbers is that devices now do a form of automated overclocking called "Boost" so you can equip a better thermal solution to allow the device to stay in OC mode longer. In the old days thig was something we used to manually do by using third party cooling solutions or waterblocks.

 

[JTWrenn]

TDP is still used, it's the amount of thermal energy, aka heat, that the cooling solution needs to be capable of dissipating. If your device is outputting 400W of thermal energy and your cooling can only move approximately 370W, then the device is going to eventually run out of thermal capacity and have to reduce it's thermal output. The reason for the "other" numbers is that devices now do a form of automated overclocking called "Boost" so you can equip a better thermal solution to allow the device to stay in OC mode longer. In the old days thig was something we used to manually do by using third party cooling solutions or waterblocks.

Sure but it used to be use to describe the amount of power and they stopped doing that. Sometimes they have both but generally they just tell you the TGP as that should tell you what you need to know for cooling as well. I suppose there could be a situation where TDP would be better but if you are producing more wattage in heat then you are pumping into something I think someone is going to be calling about the breaking of physics though unless I am missing something.

Now they use TBP or TGP and mixing those up does cause some issues as they are not exactly the same. It's not huge but it is there. Then when you go back and compare it is slightly different if there is a TDP vs TBP issue. Again should be small but the shift is confusing when comparing more than one generational jump.

 

[palladin9479]

Sure but it used to be use to describe the amount of power and they stopped doing that. Sometimes they have both but generally they just tell you the TGP as that should tell you what you need to know for cooling as well. I suppose there could be a situation where TDP would be better but if you are producing more wattage in heat then you are pumping into something I think someone is going to be calling about the breaking of physics though unless I am missing something.

Now they use TBP or TGP and mixing those up does cause some issues as they are not exactly the same. It's not huge but it is there. Then when you go back and compare it is slightly different if there is a TDP vs TBP issue. Again should be small but the shift is confusing when comparing more than one generational jump.

So TDP was never a measurement of power usage, but I can understand the confusion.

Conservation of Energy means any Thermal energy generated needs to come from somewhere, and that somewhere would be the PSU. 400W of Thermal energy means 400W+ of power draw. Thing is, just because a device is designed for 400W doesn't mean it can't be made to draw more.

 

[JarredWaltonGPU]

Sure but it used to be use to describe the amount of power and they stopped doing that. Sometimes they have both but generally they just tell you the TGP as that should tell you what you need to know for cooling as well. I suppose there could be a situation where TDP would be better but if you are producing more wattage in heat then you are pumping into something I think someone is going to be calling about the breaking of physics though unless I am missing something.

Now they use TBP or TGP and mixing those up does cause some issues as they are not exactly the same. It's not huge but it is there. Then when you go back and compare it is slightly different if there is a TDP vs TBP issue. Again should be small but the shift is confusing when comparing more than one generational jump.

TDP, TGP, and TBP are often used as synonyms, but the actual power metrics can be very different. Sometimes, TGP is just the power from the GPU. Sometimes it's GPU + RAM + VRMs. Sometimes it's everything. I personally want to know the entire power use of the card, not just the GPU — because who cares if the GPU uses 250W if the RAM and everything else needs another 50W or whatever?

Software often gets information about just the GPU, depending on the GPU manufacturer and drivers. Running RTSS for example, I can see that Arc B-series for sure isn't reporting total board power. It will show 100W while the PCAT will show fluctuations from 120~150 watts.

I've switched to mostly saying "TBP" because I want to report the total power of the entire graphics board. Now, some companies might call it by a different name, or report different power information, but I'm doing my best to report power accurately. 🤷‍♂️

 

[JTWrenn]

TDP, TGP, and TBP are often used as synonyms, but the actual power metrics can be very different. Sometimes, TGP is just the power from the GPU. Sometimes it's GPU + RAM + VRMs. Sometimes it's everything. I personally want to know the entire power use of the card, not just the GPU — because who cares if the GPU uses 250W if the RAM and everything else needs another 50W or whatever?

Software often gets information about just the GPU, depending on the GPU manufacturer and drivers. Running RTSS for example, I can see that Arc B-series for sure isn't reporting total board power. It will show 100W while the PCAT will show fluctuations from 120~150 watts.

I've switched to mostly saying "TBP" because I want to report the total power of the entire graphics board. Now, some companies might call it by a different name, or report different power information, but I'm doing my best to report power accurately. 🤷‍♂️

But isn't TBP the board plus cooling and TGP just the board? Those really shouldn't be used interchangeably should they? Especially when the manufacturer picks one specifically. Right now my understanding is they are calling it TBP power. Both TGP and TBP are supposed to be everything on the board, just one adds the cooling and such right?

So
TDP thermal power usually equating to TGP
TGP board power total, ie no leds or cooling
TBP board plus lighting and cooling.

Don't think any of these are chip only acronyms when used right.

Am I off on that? I just feel like if they report TBP it should be reported as TBP until there is a hands on test and things get messed with. This conversation really highlights my issue with a lot of it in the industry though...too many things used interchangeably seems to start to blur them all together even though they have specific meanings.

 

[derekullo]

Was for an upcoming LLM project

5090
Dual slot
32 gigabytes of ram (0.016 gigabytes per dollar)
1792 GB/s bandwidth (0.896 Gigabytes per second /$)
$2000 hopefully

A4000
single slot
16 gigabytes of ram (0.022 gigabytes per dollar)
448 GB/s bandwidth (0.629 Gigabytes per second /$)
$712 on amazon

Such hard choices!

3 - A4000s would have 1344GB/s of bandwidth, 48 Gigabytes of GDDR6 and cost roughly $2,136

GB/$ versus Gigabytes per second /$

Really leaning towards the a4000s still because performance really suffers if you can't fit the model in your gpu's ram ... some popular models are over 50 gigabytes

 

[JarredWaltonGPU]

But isn't TBP the board plus cooling and TGP just the board? Those really shouldn't be used interchangeably should they? Especially when the manufacturer picks one specifically. Right now my understanding is they are calling it TBP power. Both TGP and TBP are supposed to be everything on the board, just one adds the cooling and such right?

So
TDP thermal power usually equating to TGP
TGP board power total, ie no leds or cooling
TBP board plus lighting and cooling.

Don't think any of these are chip only acronyms when used right.

Am I off on that? I just feel like if they report TBP it should be reported as TBP until there is a hands on test and things get messed with. This conversation really highlights my issue with a lot of it in the industry though...too many things used interchangeably seems to start to blur them all together even though they have specific meanings.

I believe originally Nvidia created the acronym TGP for Total Graphics Power. And when they say that, they mean the power of the entire graphics device. For over a decade, the software reported power use of Nvidia cards has been very close to what you get when using a PCAT or my old Powenetics device to measure power.

AMD created the term TBP originally, and it stood for “Typical Board Power,” which was problematic because what is “typical?” But it can also mean “Total Board Power” which is basically the same as Nvidia’s TGP in practice.

What it really comes down to is that the two companies came up with competing terms, TDP wasn’t well defined, and now neither one will adopt the other’s acronym.

 

[uplink-svk]

600W >32 - 33 dBa 2-slot design. Yeah, I wanna see that in reality...

 

[toooooot]

Does anyone want to make predictions regarding the FE cards' temperature.
They most definitely produce more heat, each depending on the amount of power they consume.
How much better does this cooler do?
Would 5090 reach 90c under load?
I am skeptical regarding the temperature, honestly.
But that is why I am even more eager to see the tests.

 

[JarredWaltonGPU]

600W >32 - 33 dBa 2-slot design. Yeah, I wanna see that in reality...

You and me both! But it's technically 575W for the 5090. I just know in the past that we were promised great cooling and noise levels from RTX Founders Edition cards only to end up being disappointed. The RTX 40-series was actually great, but the 3080 and especially 3080 Ti FE cards were loud and hot, and the RTX 20-series was, if anything, worse.

 

[Amdlova]

You can belive in nvidia two slots can keep the 575w...

You only need one AC unit from fujitsu they normal reach 8 degree celcius...

 

[Kindaian]

Can we get benchmarks per watt? Because the numbers will be meaningless otherwise. Current GPUs are pulling more watts than some room heaters!

Performance per second is all nice and dandy, but it skews the issue to a "pump more power to the problem", instead of solving in a reasonable amount of time with the least power used (ex. generating 240 frames per second with the least power used).

 

[WonkoTheSaneUK]

"there are rumors that the 5090 at least uses liquid metal. Don't disassemble your graphics card, in other words."

I wonder how this is going to work with water-cooled solutions? I personally plan to water cool a 5080, so I'm anxious to know.

I believe it's only the FE cards that come with liquid metal.

Gigabyte have announced pre-blocked 5080 & 5090 cards, both with an AIO & for your own custom loop.
MSI are offering AIO cards.
Alphacool have announced a block for AIB cards only.
All these AIO cards come with a 360 radiator.

As I understand it, 5090 reviews will appear on the 24th and 5080 on the 29th, with both FE cards being available* from 30th.
AIB cards might be a week or 3 later.

* - Subject to F5 lottery, of course!

 

[Dean0919]

"as noted above, coupled with a $1,999 price tag."

So, basically we need to sell our kidney to buy this card, got it.

 

[JarredWaltonGPU]

"as noted above, coupled with a $1,999 price tag."

So, basically we need to sell our kidney to buy this card, got it.

And I've got bad news for you: There's a long line of people and companies who will be more than happy to fork over $2000. Very likely not for gaming purposes in most cases. Or at least, not purely for gaming.

 

[Dean0919]

And I've got bad news for you: There's a long line of people and companies who will be more than happy to fork over $2000. Very likely not for gaming purposes in most cases. Or at least, not purely for gaming.

They can pay 2000 and even 10k if they want, I don't care. I know that I'm happy with my RTX 3080 and not gonna waste money on something I don't even need. If they want to throw away such a big money, it's their choice. I'm happy with mine.