AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

Anarchist4000 said:
Sampling 4 frames in a single refresh? That's probably a 30Hz stream sampling and 120fps render without vsync. If it was a 60Hz stream then it's impressive.


That wouldn't effect engineering samples as they could source faster memory. Even then the bandwidth shouldn't be that much of a concern. With presumably larger caches, tiling, and ROPs on L2, the bandwidth requirements should be significantly less. It's more likely they're concealing performance relative to Volta while finishing off the drivers. AMD still hasn't said much about the NCUs and caching beyond FP16 rates and there are only so many reasons to still keep that secret.
Click to expand...

You seriously think they're concealing information this late in the game? Come on...
 
Anarchist4000 said:
You have some solid benchmarks of Vega and architectural details to show me?
Click to expand...

Of course not, don't be facetious, you know exactly what I mean.

Do you seriously believe they are withholding information so as to not give away how it competes with *Volta*?

Maybe that was sarcasm that went over my head?
 
ieldra said:
You seriously think they're concealing information this late in the game? Come on...
Click to expand...

One plausible explanation would be drivers, maybe they are not up to par yet. The consumer cards are 2 months away from release and probably 3 months away from stock being healthy.
 
ieldra said:
Do you seriously believe they are withholding information so as to not give away how it competes with *Volta*?
Click to expand...
Well the feature set is more in line with Voltage than Pascal, so why not? Unified memory, theoretical TFLOPs, FP16 (possibly), so why not? All indications were they skipped high end Polaris to focus on high end Vega against Volta which was pushed back when HMC fell through.
 
Clukos said:
One plausible explanation would be drivers, maybe they are not up to par yet. The consumer cards are 2 months away from release and probably 3 months away from stock being healthy.
Click to expand...

Absolutely, but that's a far cry from claiming they're withholding information so as to not give away how it compares to *volta*, because it outperforming anything Pascal based is a given.

The point being that the reason they are not showing more straightforward performance numbers is because they are not impressive, and not that they are so impressive that they don't want to give nvidia a heads up, that seems like very wishful thinking .

Anarchist4000 said:
Well the feature set is more in line with Voltage than Pascal, so why not? Unified memory, theoretical TFLOPs, FP16 (possibly), so why not? All indications were they skipped high end Polaris to focus on high end Vega against Volta which was pushed back when HMC fell through.
Click to expand...


12.5 tflops is comfortably Pascal territory, a 1080Ti is around 14 tflops if you consider the aftermarket models that throttle far less.

The Volta based GP102 equivalent will likely be in the 16tflop range.

Unified memory is also available on Pascal ?

Like I said, seems like wishful thinking.
 
Clukos said:
One plausible explanation would be drivers, maybe they are not up to par yet. The consumer cards are 2 months away from release and probably 3 months away from stock being healthy.
Click to expand...
According to Raja, the dual-GPU setup was running with FPS "well above what any single graphics card can run today", and the point of the demo was just to show Threadripper's IOs and that Crossfire AFR is working on Vega's current driver.

It doesn't excuse the fact that said demo and everything else told about Vega in the Computex keynote was a complete PR flop towards the enthusiast market.
I have no doubts the sales for the 1080 and 1080 Ti skyrocketed between May 31 and now, as AMD really blew up their enthusiasts' patience that day.
Lisa Su seemed pretty nervous when she started to talk about Vega, as her voice started shaking a lot when the subject came up. She probably knew it was going to be bad.

In fact, most people at AMD probably knew the reaction at Computex was going to be really bad. I can't see anyone from RTG that I know by name (Raja, Robert Hallock, et al) thinking "Gee, that Vega demo we're showing at Computex is sure going make enthusiasts happy". Something probably went wrong, and we either won't ever know what happened or someone will talk about it in 5 years or so, like that little steam Raja let out about AMD's plans for discrete GPUs back in ~2012.
 
It really doesn't matter what they demoed if the card is 2 months away still. What else are they going to show? They already showed a single card demo multiple times.

Also nobody has ever done a GPU demo with an fps counter for a card that is 2 months from release, why does anyone expect AMD to start now?
 
ieldra said:
The point being that the reason they are not showing more straightforward performance numbers is because they are not impressive, and not that they are so impressive that they don't want to give nvidia a heads up, that seems like very wishful thinking .
Click to expand...
That may be part of it, but then why is the NCU configuration still unknown? That would be useful for any programmers designing for the hardware. If AMD wanted the server market, providing simple details would be a good idea to get people started.

ieldra said:
12.5 tflops is comfortably Pascal territory, a 1080Ti is around 14 tflops if you consider the aftermarket models that throttle far less.
Click to expand...
Yet it's faster than P100 at some tasks and TFLOPs can be misleading without knowing the NCU configuration.

ieldra said:
Unified memory is also available on Pascal ?
Click to expand...
Not the hardware form like what Volta advertised. There is a difference and I don't foresee Pascal paging data like the HBCC should enable. Still a question if the controller can work on x86.

If you want to talk Volta, take it to another thread, but my point stands. I'm sure driver work is part of the reason, all indications are that Vega targets Volta.
 
Anarchist4000 said:
all indications are that Vega targets Volta.
Click to expand...
Really? So now Vega will somehow leapfrog Pascal and achieve 100% performance over Fiji with what? The same ALU count and a frequency boost?

These kind of statements are dangerous, this just keeps repeating every time AMD is about to launch something. People kept believing Polaris is targetting GTX 1080 till the very last minute, also with Fiji before it and Hawaii too. The hype kept reaching unbelievable levels, till it came crashing down. If AMD had a winner on their hands they wouldn't hesitate to show it in a heartbeat. Afterall, their premium card is only a month away.
 
Anarchist4000 said:
Not the hardware form like what Volta advertised. There is a difference and I don't foresee Pascal paging data like the HBCC should enable. Still a question if the controller can work on x86.

If you want to talk Volta, take it to another thread, but my point stands. I'm sure driver work is part of the reason, all indications are that Vega targets Volta.
Click to expand...
This is from Nvidia Volta white paper:
Enhanced Unified Memory and Address Translation Services GV100 Unified Memory technology in Volta GV100 includes new access counters to allow more accurate migration of memory pages to the processor that accesses the pages most frequently, improving efficiency for accessing memory ranges shared between processors.
Click to expand...
IOW: improved virtual memory access, but not the first to have HW support virtual memory.

I don't understand the urge to pit Vega against this architecture but not the other. At the end of the day, what matters is that AMD took a break from competing against the high end for more than a year.
But one has to give credit to AMD marketing for realizing that putting "Poor Pasca..." in that video would have made them look even dumber.
 
Last edited:
Esrever said:
No they didn't...
Click to expand...
One of the demos that AMD had at CES was their new Vega architecture running DOOM with Vulkan on Ultra settings at 4K resolution.
Click to expand...

https://www.pcper.com/news/Graphics-Cards/CES-2017-AMD-Vega-Running-DOOM-4K

Yes, they did...

This is all besides the point, they HAVE run demos with settings+performance numbers before. Surely you can't have forgotten this

Zjd8kZF.jpg

 
You must log in or register to reply here.
Back
Top