AMD Radeon VII Benchmark/Launch Mega Thread

[cj09beira]

4 minutes ago, mr moose said:

Wasn't that also at an inflated price due to mining/low stocks?

not always only at the end, before that it was obvious which was selling like hot cup cakes 

[mr moose]

35 minutes ago, cj09beira said:

not always only at the end, before that it was obvious which was selling like hot cup cakes 

Given the 470 was released 2 months before the 1050ti and the shortages started the following year I can't imagine there'd being much more than 4-5 months where the 470 was the better option.  It's not like now the mining boom has died almost completely that you'd be a fucking moron to get a 1050ti when the 570 is the same price.

[Taf the Ghost]

37 minutes ago, cj09beira said:

not always only at the end, before that it was obvious which was selling like hot cup cakes 

The RX400 series drivers weren't really in great shape for about 6 months, but the early prices were higher than MSRP for a while. Might have been an early part of the mining boom before it kicked into a bubble that drove it. ETH mining had already started to get going by then.

[Taf the Ghost]

2 minutes ago, mr moose said:

Given the 470 was released 2 months before the 1050ti and the shortages started the following year I can't imagine there'd being much more than 4-5 months where the 470 was the better option.  It's not like now the mining boom has died almost completely that you'd be a fucking moron to get a 1050ti when the 570 is the same price.

AMD also has cut prices pretty far, recently. GDDR5 getting cheaper and their costs on 14nm going down certainly helps in that regard.

[JoostinOnline]

2 minutes ago, mr moose said:

It's not like now the mining boom has died almost completely that you'd be a fucking moron to get a 1050ti when the 570 is the same price.

Brand loyalty was stronger before Ryzen. I'll admit I made some stupid decisions several years back, going with Nvidia over AMD just because I'd heard bad stuff about AMD. Public perception means a lot more when you don't understand the details yourself.

[mr moose]

1 minute ago, JoostinOnline said:

Brand loyalty was stronger before Ryzen. I'll admit I made some stupid decisions several years back, going with Nvidia over AMD just because I'd heard bad stuff about AMD. Public perception means a lot more when you don't understand the details yourself.

I get that, I'm simply pointing out that any perception of an over adoption of the 1050 over the 570 has other effectors that make the decision legitimate. Not every purchase was brand loyalty driven when the 1050 did provide a cheaper option for a while there.   

[i_build_nanosuits]

9 hours ago, ATFink said:

 Radeon VII is worse than a $700 rtx 2080 (until drivers are fixed; then some people will leverage the FP64 compute capability). Even after driver fixes if the prices are similar the rtx 2080 will still be better for everyone not utilizing FP64.

 

except the gimped the 16Tflops of FP64 from the actual compute card the VII is based on down to only 1.6Tflops on this card making it basically useless for double precision floating point compute.

[xAcid9]

5 minutes ago, i_build_nanosuits said:

except the gimped the 16Tflops of FP64 from the actual compute card the VII is based on down to only 1.6Tflops on this card making it basically useless for double precision floating point compute.

3.5TFLOPS actually. 1/4

[leadeater]

10 minutes ago, i_build_nanosuits said:

except the gimped the 16Tflops of FP64 from the actual compute card the VII is based on down to only 1.6Tflops on this card making it basically useless for double precision floating point compute.

The MI50 it's based off is 6.7TFLOPs FP64 so 1/2 rate of FP32. The Radeon VII is 3.46 TFLOPs FP64 so 1/4 rate FP32. Put another way the Radeon VII is half the FP64 performance for the MI50.

 

None of the figures you mentioned are actually correct.

 

An RTX 2080 for comparison is 0.3 TFLOPS FP64, or put another way... utterly useless and not worth considering as a valid option for FP64. Don't worry though a Titan RTX can surely fix that.... 0.5 TFLOPs FP64.... nope.

[Taf the Ghost]

5 minutes ago, leadeater said:

The MI50 it's based off is 6.7TFLOPs FP64 so 1/2 rate of FP32. The Radeon VII is 3.46 TFLOPs FP64 so 1/4 rate FP32. Put another way the Radeon VII is half the FP64 performance for the MI50.

 

None of the figures you mentioned are actually correct.

 

An RTX 2080 for comparison is 0.3 TFLOPS FP64, or put another way... utterly useless and not worth considering as a valid option for FP64.

Science Department King Radeon VII.

[cj09beira]

1 minute ago, Taf the Ghost said:

Science Department King Radeon VII.

if those science departments aren't nvidia departments, mine is, such a sad thing to see

[Stefan Payne]

6 hours ago, xAcid9 said:

your reason is similar to Buildzoid and i agreed. 

No, its completely different because one is the Corporate side, wich is what Bullzoid was talking about.

I was talking about the Consumer side. 

 

And that is the real Problem, that it makes little sense for AMD to target "Gamers" because they won't buy their products, even if they are good...

Thus it makes more sense to target the ones that are not brand loyal - wich are usually at the lower end - and go for them...

 

But still, look at the Steam Charts. Its rediculous...

[Taf the Ghost]

And I'm really not joking about the Science Department King bit. These cards are going to be living very long lives in builds at Science Departments of Universities the world over for at least a decade. Especially if anyone figures out how to flash a BIOS that can enable 1:2 FP64.

[leadeater]

1 minute ago, Taf the Ghost said:

Science Department King Radeon VII.

Ninja edited in the Titan RTX just to show how comical even that is for FP64. Not even the Quadro RTX cards have usable FP64, exact same meager performance.

 

Without price checking I think the cheapest Nvidia card with usable FP64 performance is the Titan V which doesn't beat out the Radeon 7 by much, for 4 times the price.

[Taf the Ghost]

2 minutes ago, cj09beira said:

if those science departments aren't nvidia departments, mine is, such a sad thing to see

To an extent, but you can stack Compute Cards in the same system. That's one of the fun tricks with them. Also, you're going to see professors buying them up themselves. These are spot-type purchases users can get, which is why the secondary market for these is going to last a long time.  Same thing with the Hawaii cards.

[Taf the Ghost]

1 minute ago, leadeater said:

Ninja edited in the Titan RTX just to show how comical even that is for FP64. Not even the Quadro RTX cards have usable FP64, exact same meager performance.

 

Without price checking I think the cheapest Nvidia card with usable FP64 performance is the Titan V which doesn't beat out the Radeon 7 by much, for 4 times the price.

Pretty sure the Tesla parts are the only ones with FP64. 

 

From here: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#Tesla

 

It looks like you need either a GP100 or GV100 based card to have better FP64.

[cj09beira]

5 minutes ago, Taf the Ghost said:

And I'm really not joking about the Science Department King bit. These cards are going to be living very long lives in builds at Science Departments of Universities the world over for at least a decade. Especially if anyone figures out how to flash a BIOS that can enable 1:2 FP64.

i always thought that there is a good chance that the Wx bios should be able to run on the reference pcb cards

[cj09beira]

btw have you guys seen this 

i find it quite hard to believe, we will see^TM

 

[Taf the Ghost]

2 minutes ago, cj09beira said:

i always thought that there is a good chance that the Wx bios should be able to run on the reference pcb cards

The Radeon VII is cheap enough someone is going to try. If they can really get 1:2 FP64, these things are going to be the utter kings of a Niche. 

[Taf the Ghost]

Just now, cj09beira said:

btw have you guys seen this 

i find it quite hard to believe, we will see^TM

 

It's one rumor, though we know multiple Navi products are coming. One being released in October is extremely likely, but the launch is probably August. That's AMD's normal window for GPUs.

[Lathlaer]

9 minutes ago, Taf the Ghost said:

Pretty sure the Tesla parts are the only ones with FP64.

Doesn't Titan V offer about 6.1TFLOPS of double precision?

[Taf the Ghost]

Just now, VegetableStu said:

oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooof ._.

(or maybe with the same timeframe as Threadripper 3000? o_o)

AMD should be dropping a lot of info at GDC in March, plus a lot of the industry will be meeting together. We should have firm time lines after that point.

[Taf the Ghost]

1 minute ago, Lathlaer said:

Doesn't Titan V offer about 6.1TFLOPS of double precision?

Yes, but it's their only normally available type of GPU that does. It's also 3K USD. You could buy 2 Radeon VII + an entire system for the same cost.

[leadeater]

27 minutes ago, Taf the Ghost said:

Pretty sure the Tesla parts are the only ones with FP64. 

 

From here: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#Tesla

 

It looks like you need either a GP100 or GV100 based card to have better FP64.

There's non Telsa parts like the Titan V and Quadro P100 and V100, Quadro variants have video outs and Telsa variants don't.

[JoostinOnline]

23 hours ago, GoldenLag said:

even if it was 100$ cheaper. people would still not buy it. its just how GPU market works

 

I honestly think it would if they made an 8GB version for $100 less. It seems like the performance would be similar in gaming.