Vega/Navi Rumors (Updated)

[Crumpet]

I certainly don't expect it to be a super expensive GPU because Raja seemed to make multiple points about Nvidia's prices.

 

[Head1985]

I certainly don't expect it to be a super expensive GPU because Raja seemed to made multiple points about Nvidia's prices.

Yeah and fiji was overpriced and didnt have success because of that.Both furyx and fury was disaster.

 

[Crumpet]

Yeah and fiji was overpriced and didnt have success because of that.Both furyx and fury was disaster.

Indeed, though with Raja ALSO clearly stating that Vega is a high end gpu and is NOT a budget GPU I also don't want people to think they're going to be getting some miracle card at ridiculously low prices.

I personally think best case scenario we're looking at $450+, worst case scenario $700.

 

[EnzoLT]

I hope it's around $400-$500. There's a big gap there that needs to be filled

Sent from my SM-G930F using Tapatalk

 

[antihelten]

Who here is talking about half precision math? Outside of some speculation on its possible adoption in VR, how is that relevant for gaming performance?

AMD is the ones talking about half precision math, seeing as the "25" in MI25 refers to half precision. Obviously we can derive the full precision performance from this number (12.5 GFLOPS), but the only number we have direct knowledge of from Vega is the FP16 one.

And no one is talking about gaming performance, we're talking about theoretical floating point performance, and using this to calculate the clock rate. Remember my initial reply to you was about your comment on the clock rate of Vega and the various numbers that have popped up and what evidence exists to support them, not about gaming performance.

I'm talking about single precision, which is why I was making calculations based on 12.5TFlops.

Doesn't matter, you still end up at 1.5 GHz, since the theoretical FP32 of Vega is well known (and there are zero IPC improvements here).

Of course it's possible that all the IPC improvements AMD is talking about is in half precision. That is all they've demonstrated and talked about directly, at least. But that would be "IPC improvements*" with a really frickin' huge asterisk next to it. From the general/non-specific language used, I'm hesitant to believe that all the IPC improvements they're talking about are in packed math. After all, for any and all regular end user use, that is zero IPC improvement - as I showed quite clearly with my extremely simple calculation. It doesn't matter in 2017 if VR games and experiences start using half-precision math in 2019.

I never said that packed math is the only IPC improvement, only that's it's the only IPC improvement that affects the theoretical floating point performance of Vega. And since we know the theoretical FP16 performance of MI25 (25 GFLOPS), that is all we need to calculate the clock rate (again unless MI25 has more than 4096 shaders).

This has nothing to do with gaming performance directly, obviously we cannot accurately calculate this, even though we know that MI25 is 4096 shaders at 1.5 GHz, since there are many other factors that affect this (including IPC improvements in other areas).

 

[piesquared]

GP102 on 16nm: 2.1GHz
GP104 on 16nm: 2.1GHz
GP106 on 16nm: 2.1GHz
GP107 on 14nm: 1.9GHz



You cannot look at CPU clockspeeds to say what clockspeeds GPU's on a process will achieve.

Of course you can't, but it proves that there isn't a process limitation for a higher clocked GPU. Polaris and lower is not optimized for higher clocks, but the design competes with an architecture that runs with a 30% higher clock.

The GP107 has to fit inside a 75W TDP. We don't know what clocks it could hit at 75W or less on 16nm. The power/frequency curve won't be the same between the 2 processes, so a GP107 at 1.9 GHz on 14nm tells us nothing about how the same GPU with the same power limits would clock on 16nm.

 

[Bouowmx]

The 1.9-GHz figure comes from cards with additional power connector (not restricted to 75 W): MSI GAMING X (Guru 3D) and ASUS ROG STRIX (TechPowerUp). Otherwise, restricted, GeForce GTX 1050 can overclock to 1.7-1.8 GHz: ZOTAC Mini (Guru 3D).

 

[Harmaaviini]

The 1.9-GHz figure comes from cards with additional power connector (not restricted to 75 W): MSI GAMING X (Guru 3D) and ASUS ROG STRIX (TechPowerUp). Otherwise, restricted, GeForce GTX 1050 can overclock to 1.7-1.8 GHz: ZOTAC Mini (Guru 3D).

That TPU link is interesting. On overclocking page they say:
GPU overclocking is capped too, at 1911 MHz actual frequency after GPU Boost, but the implementation seems buggy. The actual clock can still be increased even though 1911 MHz is displayed; at beyond 1911 MHz displayed, the card will run at higher performance, and at even higher clocks, it will get increasingly unstable. Maximum overclock of our sample was +185 MHz to the GPU clock, a 13% increase.

Looking at the perf/W chart we see it right up there with other 10-series cards. It wouldn't be if Samsung process was really terrible. https://www.techpowerup.com/reviews/ASUS/GTX_1050_Ti_Strix_OC/30.html

 

[Mopetar]

I certainly don't expect it to be a super expensive GPU because Raja seemed to make multiple points about Nvidia's prices.

It still really depends at where they come in at in terms of performance. If by some small miracle AMD has something that's slightly above Titan Xp level, then then even $800 can be said to be reasonable.

 

[guskline]

Oh for the Love of the Almighty, VENT the VEGA!!!!!!!!!!!!!!!!

 

[Snarf Snarf]

The tension this AMD silence is creating is palpable. I would literally take a screen shot of RX VEGA in Radeon font right now with a "coming soon," I'm that anxious about this, I WANT A NEW CARD AND A NEW ADAPTIVE SYNC MONITOR.

4096 SP's at 1500 MHz with DX11 performance around 5-10% behind 1080ti and Vulkan/DX12 performance 15% faster for $650 would have me preordering a big Vega on paper launch day and a 1440p 144Hz FreeSync monitor shortly after.

If it turns out to be 1080 +10% and about 20% behind 1080ti, it would need to priced in the 499-550 range with aftermarket options available on day one.

 

[guskline]

Like Wayne and Garth, my twin RX480s are in my Ryzen 1800x rig crying out to Vega. They cry "WE ARE NOT WORTHY!"

 

[tential]

The tension this AMD silence is creating is palpable. I would literally take a screen shot of RX VEGA in Radeon font right now with a "coming soon," I'm that anxious about this, I WANT A NEW CARD AND A NEW ADAPTIVE SYNC MONITOR.

4096 SP's at 1500 MHz with DX11 performance around 5-10% behind 1080ti and Vulkan/DX12 performance 15% faster for $650 would have me preordering a big Vega on paper launch day and a 1440p 144Hz FreeSync monitor shortly after.

If it turns out to be 1080 +10% and about 20% behind 1080ti, it would need to priced in the 499-550 range with aftermarket options available on day one.

I already got my monitor. I was too afraid of all the cheap Korean freesync monitors selling out. I can't find the uhd550 much anywhere which was my main monitor I want.

So now I'm just really sad that I can't use the monitor.

 

[zinfamous]

At this point, I am starting to think that the silence means only good things. Because if Vega is flops, or is merely adequate, then this kind of anticipation leading up to a loser will really hurt AMD, imo.

At least, that's what I tell myself

 

[zinfamous]

I already got my monitor. I was too afraid of all the cheap Korean freesync monitors selling out. I can't find the uhd550 much anywhere which was my main monitor I want.

So now I'm just really sad that I can't use the monitor.

I kept refreshing and was planning on buying that Microboard when it was going for $700..over the last 3 months, with still no ability to use it, lol. Glad I didn't get it, because that AOC will be out in a few weeks, and I think is a wee bit better, maybe.

 

[tential]

I kept refreshing and was planning on buying that Microboard when it was going for $700..over the last 3 months, with still no ability to use it, lol. Glad I didn't get it, because that AOC will be out in a few weeks, and I think is a wee bit better, maybe.

What's the cost though? Primary reason I'm with amd is cost. It's stupid cheap to build a good system with amd at the high end if you're willing to wait. On top of this, sli is dead, cf isn't. I hate it when users say they're both bad. No, sli sucks. It's terrible in comparison to cf. So that's another reason I'm waiting for amd. Is cf potential. Sli scaling was horrid with Nvidia high end.

I can get a 4k and 1440p monitor with freesync cheaper than 1 gsync monitor for my purposes. That's stupid.

I'm happy to be in the amd ecosystem only because it feels like monitor tech is changing fast now. If amd could get their time to market down, and work with TV vendors to get freesync on tvs ASAP this would be game over for me personally. TV with freesync? You can have all my money.

 

[IEC]

Indeed, though with Raja ALSO clearly stating that Vega is a high end gpu and is NOT a budget GPU I also don't want people to think they're going to be getting some miracle card at ridiculously low prices.

I personally think best case scenario we're looking at $450+, worst case scenario $700.

You could actually have both those scenarios be true simultaneously. A full fat, uncut Vega plus a 1-step cut Vega. AMD did it for 290/X, Fury/X, and RX470/RX480. Assuming the performance is there, of course. In all those cases the slightly cut-down card was the best bang for the gaming buck for a long time.

 

[Paratus]

At this point, I am starting to think that the silence means only good things. Because if Vega is flops, or is merely adequate, then this kind of anticipation leading up to a loser will really hurt AMD, imo.

At least, that's what I tell myself

Oh it does. I know because I'm going to upgrade my entire system late this year and NV never has a competitive product when I'm in the market.

I bought:

• 9600xt vs FX5700
• AGP x1950 Pro vs 7800GS
• HD 5870 vs GTX 285
• HD 7970 Ghz vs GTX 680

So I fully expect Vega to crush the 1080TI.

 

[EnzoLT]

Oh it does. I know because I'm going to upgrade my entire system late this year and NV never has a competitive product when I'm in the market.

I bought:

• 9600xt vs FX5700
• AGP x1950 Pro vs 7800GS
• HD 5870 vs GTX 285
• HD 7970 Ghz vs GTX 680

So I fully expect Vega to crush the 1080TI.

Delusional optimism lol but hopefully nonetheless.

Sent from my SM-G930F using Tapatalk

 

[Valantar]

Oh it does. I know because I'm going to upgrade my entire system late this year and NV never has a competitive product when I'm in the market.

I love tour logic.

 

[sandorski]

AMD has invited some Press to attend an AMD/MSI event in Austin. Vega related?

 

[zinfamous]

Oh it does. I know because I'm going to upgrade my entire system late this year and NV never has a competitive product when I'm in the market.

I bought:

• 9600xt vs FX5700
• AGP x1950 Pro vs 7800GS
• HD 5870 vs GTX 285
• HD 7970 Ghz vs GTX 680

So I fully expect Vega to crush the 1080TI.

haha. that is kinda my history as well. I tend to think I will go nVidia, but in the lengths of upgrade cycles for me, I tend to hit when AMD happens to have the overall better options at the plate (generally performance/price)

 

[SighTurtle]

AMD has invited some Press to attend an AMD/MSI event in Austin. Vega related?

I'm not seeing much press. I think MSI and AMD are gonna host another one of those gaming showoffs they did for Ryzen, streamers playing on Vega computers. Tomorrow, they do it, with teaser info from press outlets.

this is merely a guess.

 

[krumme]

What kind of perf is needed to play bf1 at 4k High setting 70fps min?

 

[alcoholbob]

What kind of perf is needed to play bf1 at 4k High setting 70fps min?

You only get about 10 percent more perf out of high versus ultra. The big deal with frostbite is whether or not you have MSAA on. With just FXAA or no AA you should get about 20+ percent more performance. At High with no AA you should get about 35 percent better performance than Ultra with AA.

A 1080 will get you ~55fps with Ultra/MSAA, at High with no AA you are looking at 75fps averages.

I think an OCed 1080 (to 2GHz or so) should be able to hit your goals of 70fps min at 4K assuming you are running High/no AA.