3080 + 3090 - Most power hungry cards in recent memory

[OldFatherTime]

I think you've completely missed the point for the criticism which was more to do with the insane power and heat for the size of the die and the performance lets not get too fanboyish here now.

I dont think anyone should really care about power consumption providing the performance is good enough I mean you can run the 3080 on a 650w power supply which isn't a issue for most people at all unless theres a underlying agenda which seems to be the case.

I'm giving up. This is bizarre. Power consumption = Heat. There's nothing more to say.

I think we just wait and see what happens and see what the actual mainstream users think... not after a day but after a month or two. And I predict they won't be overjoyed.

BTW - a 650w Power Supply better not be some cheap ****** clone - and be at least Bronze, preferably higher. Even then people are going to experience problems - especially if they have latest Intel Processors.

 

[Rroff]

Are you sure about that? It is my understanding that the team that works on one node isn't allowed to work on a competitor's node due to legal reasons and signed legal documents which makes sense since TSMC doesn't want to share node information with another Fab and wise versa.

I've never encountered anything like that aside from some very specific stuff i.e. when working with Micron on developing memory.

AFAIK stuff like their failure analysis labs, etc. don't have separate teams per where the cores are being made with several of the more experienced people working across a range of products.

EDIT: In general though there isn't that level of secrecy - I've looked at having custom ASICs made up at TSMC before (shuttle runs) and been somewhat involved in that in a previous job and aside from stuff like risk production there isn't that much in the way of that kind of legal stuff involved - just your usual Confidential Information handling terms/clauses.

 

[Zeed]

I think you've completely missed the point for the criticism which was more to do with the insane power and heat for the size of the die and the performance lets not get too fanboyish here now.

I dont think anyone should really care about power consumption providing the performance is good enough I mean you can run the 3080 on a 650w power supply which isn't a issue for most people at all unless theres a underlying agenda which seems to be the case.

Power consumption Could be 500w i got cooling power as long as there is processing power I DONT CARE.

 

[Phixsator]

I've never encountered anything like that aside from some very specific stuff i.e. when working with Micron on developing memory.

AFAIK stuff like their failure analysis labs, etc. don't have separate teams per where the cores are being made with several of the more experienced people working across a range of products.

EDIT: In general though there isn't that level of secrecy - I've looked at having custom ASICs made up at TSMC before (shuttle runs) and been somewhat involved in that in a previous job and aside from stuff like risk production there isn't that much in the way of that kind of legal stuff involved - just your usual Confidential Information handling terms/clauses.

I see, but could that be different if we are talking work with the bleeding edge nodes/technology?

 

[CuriousTomCat]

Power requirements are relative. You could easily overclock the most power efficient GPU in history until it becomes power hungry.

Nvidia have clearly just clocked these higher than usual that's all. If they hadn't done that, the enthusiast would have overclocked them anyway.

Nvidia haven't actually created a power hungry architecture.

 

[Rroff]

I see, but could that be different if we are talking work with the bleeding edge nodes/technology?

I don't know - I have a general idea of their setup but not day to day operations.

 

[OldFatherTime]

Power requirements are relative. You could easily overclock the most power efficient GPU in history until it becomes power hungry.

Nvidia have clearly just clocked these higher than usual that's all. If they hadn't done that, the enthusiast would have overclocked them anyway.

Nvidia haven't actually created a power hungry architecture.

Delusional.

 

[OldFatherTime]

I don't know - I have a general idea of their setup but not day to day operations.

More fascinating insight.

 

[SimonOcean]

With great power, comes great coil whine lol.
It was mentioned in a couple of video reviews as well.

Oh hoooo. Coil whine is a deal breaker for me. I've got a totally silent rig. High frequency coil whine would be a huge step backward. In that case I would rather wait for Big Navi.

 

[clv101]

For me the high power consumption makes this release a bit of a failure, engineering wise. It seems like most of the performance increase has simply come from burning more power. Thats not progress. Progress needs to come from architecture and silicon, not from designing a better cooler!

CPUs have managed to live with ~100W for 20+ years, Nvidia are mucking around with 320W and 350W parts. I'm really hoping AMD can be competitive with 3080 for 220-250W.

 

[Rroff]

More fascinating insight.

Dunno what your problem is.

 

[CuriousTomCat]

Delusional.

Well not really. The Samsung 8nm node is efficient on power. Not as power efficient as tsmc but that's a smaller node so it's understandable. It's just that Nvidia have cracked the clockspeed to a stupidly high level which naturally consumes a lot of power and generates a lot of heat.

 

[K1LaW8tt5]

It's just that Nvidia have cracked the clockspeed to a stupidly high level which naturally consumes a lot of power and generates a lot of heat.

because they had to in order to get a performance increase. The card has no headroom and operates at the power limit by default. It screams of Vega and yes I have one too.

You get the impression Nvidia has something else lined up to replace within 12 months. I can't see them creating a mobile gpu out of these either.(without moving from samsung)

 

[five8five]

30W higher power consumption compared to a 2080Ti for a 32% performance increase at 4k seems reasonable.

Cooler seems to do a decent job as well.

 

[clv101]

30W higher power consumption compared to a 2080Ti for a 32% performance increase at 4k seems reasonable.

No I don't see that as reasonable at all. Where does that thinking take us? 400W cards next year, then 500W? If performance had scaled with power the last decade we'd be pumping hundreds of kW into our systems by now. Raw performance needs to come with comparable efficiency improvements.

 

[Harlequin]

Law of Thermo Dynamics 101 - 320w of electrical power, converted to heat actually has to go somewhere......

 

[five8five]

No I don't see that as reasonable at all. Where does that thinking take us? 400W cards next year, then 500W? If performance had scaled with power the last decade we'd be pumping hundreds of kW into our systems by now. Raw performance needs to come with comparable efficiency improvements.

It's only 30W! For 4k gaming now it's a fair trade off imo. Looking at the list it doesn't seem too bad.......
Btw, did you look at the Performance per Watt chart?

The rumour is that AMD cards are going to be quite poor hungry as well. The revealed cards, three fan and two fan cards are 2x 8pin - 375W max.

 

[K1LaW8tt5]

30W higher power consumption compared to a 2080Ti for a 32% performance increase at 4k seems reasonable.

Cooler seems to do a decent job as well.

You will find to get the 30% it will be more than 30w.

 

[Cyber-Mav]

The rumour is that AMD cards are going to be quite poor hungry as well. The revealed cards, three fan and two fan cards are 2x 8pin - 375W max.

Worse part is that new amd card will be trying to still take on the 2080ti.

 

[Rroff]

The rumour is that AMD cards are going to be quite poor hungry as well. The revealed cards, three fan and two fan cards are 2x 8pin - 375W max.

I would laugh if after all this AMD came out with a new line of cards on 8nm.