Nvidia Ampere Discussion [2020-05-14]

[CarstenS]

You forgot the countdown - I don't remember when it startet though.

 

[trinibwoy]

You forgot the countdown - I don't remember when it startet though.

Yeah I didn't forget, just don't consider it "info". It was really just an announcement of an announcement.

 

[pharma]

The "new" cooler seems to be very effective. Hope some review really test the claims.

 

[CarstenS]

The differences for temperature and noise seem useless, since the 2000-series' coolers are far beyond their design points (+100 watts)

Yeah I didn't forget, just don't consider it "info". It was really just an announcement of an announcement.

Fair enough.

 

[Jawed]

3090 is cooler than 3080. That's worth an extra $800 isn't it?

 

[Geeforcer]

oh god, Nvidia actually made their own unboxing video. That's just sad.

If they had a sense of humor, they should have filmed a "boxing" video at their contractor's plant packaging line, with a shot of a worker handing it to a the media crew who immediately start unboxing it while trying not to get burnt on still-hot shrink-wrap.

 

[pharma]

That RTX 3090 is one BIG card!

 

[trinibwoy]

3090 is cooler than 3080. That's worth an extra $800 isn't it?

Eh, there's no way the 3090 is at 70C / 30dB when pushing 350w. Nobody would buy AIB cards if that's true.

 

[PSman1700]

That RTX 3090 is one BIG card!

Damn i want one

 

[ECH]

• Firestrike - 200fps+, 31919 score, 25% faster than 2080Ti, 43% faster than the 2080S
• Firestrike Extreme - 1440p, 20101 score, 24% faster than 2080Ti , 45% faster than 2080S
• Firestrike Ultra - 2160p, 11049 score, 36% faster than 2080Ti, 64% faster than 2080S
• Timespy - 1440p, DX12, 17428 score, 28% faster than 2080Ti, 49% faster than 2080S
• Timespy Extreme - 4K DX12, 8548 score, 38% faster than 2080Ti, 59% faster than 2080S
• Port Royal - 1440p, 11455 score, 45% faster than 2080Ti, 64% faster than 2080S

A decent set of AAA games was also run on the purported RTX 3080 system and yielded the following RTX 2080 Super comparisons:

• Farcry 5 - 4K, GPU maxed at 99% (throttling), 5GB Video Memory used, 62% faster than 2080S
• Borderlands 3 - 4K, DX12, GPU maxed at 100% (throttling), 7.4GB Video Memory used, 56% faster than 2080S
• Horizon Zero Dawn - 4K, GPU maxed at 99% (throttling), 9.9G Video Memory used, 76FPS average
• Assasins Creed Odyssey - 4K, GPU at 97%, 7GB Video mem, 67fps average, 48% faster than 2080S
• Forza Horizon 4 - 4K, 143fps, GPU at 97%, 48% faster than 2080S

https://hexus.net/tech/news/graphic...a-geforce-rtx-3080-benchmarks-video-released/

 

[DegustatoR]

It's the same benchmark which was making the rounds yesterday.
The heavier is the workload on shading - the higher is the gain over 20 series.
I'm not sure what they mean by "throttling" though.

 

[PSman1700]

The heavier is the workload on shading - the higher is the gain over 20 series.

RTX gonna love UE5 games.

 

[pharma]

It's the same benchmark which was making the rounds yesterday.

The Hexus link also includes the "review" benchmark video.

 

[techuse]

Those numbers are much more sobering. Wonder why their borderlands numbers are so much lower than Nvidia’ s.

 

[pharma]

Same numbers as before, though will wait for "legitimate" review site results before concluding the accuracy of these numbers. This "leak" site apparently has a history of fudging.

 

[agent_x007]

I'm not sure what they mean by "throttling" though.

I'm sure it's about power throttling when GPU get's 1V+ on die.
With more efficient use of resources, power usage get's higher.
Fully unlocked high end Turings (TU104/TU102), are always power limited under most GPU heavy scenarios (assuming not CPU/platform limited).

NV should learn how to lock default max. boost voltage to be around 0,9V.
Otherwise, they will only make even bigger heater next time.
Default max. doesn't have to be maximum allowed under OC.

 

[DegustatoR]

I'm sure it's about power throttling when GPU get's 1V+ on die.

It's not "throttling", it's normal GPU boost operation.

And I kinda wonder recently since when does NV of all companies on the market should suddenly learn how not to make "heaters"?
We're assuming way too much right now.

 

[SimBy]

It's the same benchmark which was making the rounds yesterday.
The heavier is the workload on shading - the higher is the gain over 20 series.
I'm not sure what they mean by "throttling" though.

What it means is the clock drops because of power or heat constraints.

 

[DegustatoR]

What it means is the clock drops because of power or heat constraints.

Again, which is a normal mode of GPU boost operation.

There's thermal throttling which happens if the card hits above max temp but I doubt that these FEs will throttle like this. If they will then these weird looking new coolers on them are actually bad.

 

[agent_x007]

It's not "throttling", it's normal GPU boost operation.

Throttling or limit, doesn't matter how you name it.
GPU Boost works from limit to limit because it's not meant for GPU to have stable performance.
Meaning : Theoretical value at set clock speed are 100% inaccurate for trying to make/calculate a end performance value.
Still, everything is fine... as long, as you don't compare performance with clock speeds of those type of cards.