Rx Vega hits SiSoft Sandra, 35% faster than gtx 1080.

[Memories4K]

Just now, Ryan_Vickers said:

what myth?  Who ever believed that?

I keep seeing people repeat it:

"Nah man, 4GB of HBM is just like 8GB of GDDR5, no worries. Don't believe these Nvidia shills who tell you different."

[vanished]

Just now, Memories4K said:

I keep seeing people repeat it:

"Nah man, 4GB of HBM is just like 8GB of GDDR5, no worries. Don't believe these Nvidia shills who tell you different."

By that logic if I overclock my 32 GB of DDR3 enough I can simulate having 48 GB of it

 

[AdmiralMeowmix]

2 minutes ago, Memories4K said:

I keep seeing people repeat it:

"Nah man, 4GB of HBM is just like 8GB of GDDR5, no worries. Don't believe these Nvidia shills who tell you different."

Plus 16 GB's of HBM isnt gonna run as fast as 8 GB's.

 

Nvidia tried and failed. They couldn't get it to run as fast as they needed it too over 8 GB's which is why they went back to GDDR5X for the 1080ti which they improved upon and upped the speed on.

 

If AMD got 16 GB's of HBM to run just as fast as 8 GB's while Nvidia couldn't I'll surely be impressed.

[VagabondWraith]

1 hour ago, AluminiumTech said:

Fully active chip? WTF are you on about?

 

The 1080Ti has all SMs enabled using GP102.

Nah, the 1080 Ti (and TXP) are cut-down GP102 chips.

[corsairian]

35% faster than a 1080 is nothing to scoff at. Especially if the TDP is a bit lower than a 1080 is. That kind of performance in something the size of a Fury will be amazing. I want to see a Dual GPU version of their flagship VEGA card. 

[ivan134]

9 minutes ago, corsairian said:

35% faster than a 1080 is nothing to scoff at. Especially if the TDP is a bit lower than a 1080 is. That kind of performance in something the size of a Fury will be amazing. I want to see a Dual GPU version of their flagship VEGA card. 

We've seen 3 games already that have great mGPU implementation. If we get more, I would actually get excited for dual GPUs.

[Tech Inquisition]

2 hours ago, VagabondWraith said:

Nah, the 1080 Ti (and TXP) are cut-down GP102 chips.

Technically GP102 is a cut GP100 with alternative memory controllers for G5X and not HBM2, and slimmed down for more gaming performance...

1080Ti and TXp both have different versions of the GP102.

[Prysin]

4 hours ago, App4that said:

Hold on, I'm overclocking my 1800x to 5GHz like the leaked benchmarks. Don't want to mess this up. 

5.8GHz m8. It leaks all the heat out!!

[zMeul]

47 minutes ago, Tech Inquisition said:

Technically GP102 is a cut GP100 with alternative memory controllers for G5X and not HBM2, and slimmed down for more gaming performance...

1080Ti and TXp both have different versions of the GP102.

the GPU for Tesla are quite different, they have different FP precision

and ps: the GP100 in the Tesla P100 is not the full Pascal - this was revealed back when nVidia unveiled the Tesla P100

 

I don't think we'll ever see the GP100 on desktop anytime soon, if ever - it will be a bitch to cool

[Delicieuxz]

4 hours ago, Ryan_Vickers said:

By that logic if I overclock my 32 GB of DDR3 enough I can simulate having 48 GB of it

 

I don't think the reasoning has to do with its speed, but with efficiency. It sounds to me like Raja is saying that games request and are allocated a lot more VRAM than they actually use (Raja suggests most games don't use more than 50% of what they request and receive allocation for), but that with Vega, the handling of VRAM allocation is improved, so that a game is only allocated exactly what it makes use of, thereby freeing up all the VRAM that has traditionally been tied up in allocation, while not actually being needed by the game that's requested it.

 

 

[Coaxialgamer]

5 hours ago, AluminiumTech said:

Fully active chip? WTF are you on about?

 

The 1080Ti has all SMs enabled using GP102.

Nope. Full gp102 has 3840 cuda cores. It can be found in the quadro p6000 for example. 

[Coaxialgamer]

5 hours ago, Memories4K said:

If i remember correctly, it's not that much more to manufacture a 480 than a 460.

In this case, it'll be a bit more since HBM2 but i'm pretty sure AMD can still make a hefty profit and more importantly eat at the market share that Nvidia has a hold of.

Both of those chips are fairly small at 123 and 233mm² respectively. Once you get to the really large chip, small differences can make a huge difference in dies per wafer and yields. 

[Space Reptile]

well can someone translate these benchmarks to somewhat real world performance?

I don't like sisoft Sandra marks as they are just gibberish to the common mortal

[AlTech]

5 hours ago, VagabondWraith said:

Nah, the 1080 Ti (and TXP) are cut-down GP102 chips.

I just realized.

 

Both have 56 out of 60(?) SMs enabled. For some reason I thought that the Titan XP was fully enabled....

[AlTech]

6 hours ago, Doobeedoo said:

The 8GB HBM2 though. Can't wait to see that running. Though I wonder will there be 16GB model too.

If there is one then there would be an unreasonable price hike for the 16GB model.

 

Because the 16GB model would need to be using a 4096 Bit HBM2 memory bus with 1TB/s bandwidth which is a lot more expensive than the 2048 Bit HBM2 bus.

[PlayStation 2]

I'm going to hold salt just so I don't have to hold an L if the hype train busts.

[juri-han]

6 hours ago, Memories4K said:

This myth needs to die

not a myth if AMD said it, in fact it was a major part of their capsaicin and cream presentation.....

 

they talked abut how vega would effectively use a higher percentage of VRAM where other cards only take advantage of only 40% vega would use up to 80%, so doubling the the VRAMS performance, you wont need as much to do the same job..... so its ether false or true and we wont know till vega launches and this is why you dont preorder but its NOT a rumor 

[TotaLInsanity]

Even tho i would love vega to be a success i still think this is too little too late.

[suicidalfranco]

9 hours ago, RagnarokDel said:

People who use numbers instead of words should be banned from these forums.

u mus7 b3 fun 47 p4r7135

[Valentyn]

8 hours ago, App4that said:

Hold on, I'm overclocking my 1800x to 5GHz like the leaked benchmarks. Don't want to mess this up. 

That's not a RyZen cpu. The benchmark results lists an FX 970 chipset. Bonkers stuff. 

[Humbug]

16 minutes ago, TotaLInsanity said:

Even tho i would love vega to be a success i still think this is too little too late.

yes it is

The GTX 1070 and GTX 1080 have had a good run for about a year. And now they are even being price cut and superseded by the GTX 1080ti. All while AMD's highest end competitive card is the rx 480. Just not good planning/execution from AMD...

[DoctorZeus]

How Do i run this damn benchmark

[Doobeedoo]

2 hours ago, AluminiumTech said:

If there is one then there would be an unreasonable price hike for the 16GB model.

 

Because the 16GB model would need to be using a 4096 Bit HBM2 memory bus with 1TB/s bandwidth which is a lot more expensive than the 2048 Bit HBM2 bus.

Yeah true. We'll see, though I guess 8GB may be the only version. Cause I remember them mentioning something it can keep up with more amount of GDDR anyway.

[DoctorZeus]

http://ranker.sisoftware.net/show_run.php?q=c2ffcdf4d2b3d2efdbe9deecd9ebcdbf82b395f095a898becdf0c0

 

I need an adult. I'm rank #-1

[TidaLWaveZ]

14 hours ago, Ryan_Vickers said:

So Vega is 35% faster than a 1080, and the 1080 Ti is 35% faster than a 1080, but the 1080 Ti is 25% faster than Vega?  Am I missing something...?

 

I see this is causing a lot of confusion, it all makes perfect sense if you account for the D2F or "Dick to Floor" ratio.