With the GTX 1070 typically commanding a 10% lead over the Fury X, in this particular title both are neck and neck.
In that video he said that developers won't have to worry about memory size limitations when creating higher resolution assets for 4K due to HBCC, not that games would need to have developer support in order to benefit from it.
If you were referring to me:
https://www.techpowerup.com/reviews/Zotac/GeForce_GTX_1080_Ti_Amp_Extreme/31.html
I don't see a 10% lead there
But yes, the 1070 drops off heavily @ 4k in this game
https://www.techpowerup.com/reviews/Performance_Analysis/Sniper_Elite_4/4.html
Still a good game FPS though and the game is very well optimized and good use of Async Compute.
http://www.guru3d.com/articles-pages/msi-gtx-1080-ti-gaming-x-review,12.html
They approached at different angles so hard to compare, but when they both are at the same place the FPS is actually very similar.
https://youtu.be/590h3XIUfHg
And that is against a 1962MHz 1080 Ti! Boost is under 1600 MHz. So its an OC'd 1080 Ti vs Vega and Vega is performing very well.
Reddit is already awash with people believing that Sniper Elite demo was run on 2gb vram.
Fml.
Which one version of the RX Vega GPU?It looks like Vega will be 10% slower than 1080 TI in titles skewed heavily in AMD's favor like Sniper Elite, which means it'll be 20% slower in a much larger swath of titles. If this ends up the case, it'll average 15+% slower than 1080 TI. I'm still curious as to the die size and power consumption metrics since neither was discussed. All in all though, as a halo gaming card it's too late and going to be VERY short lived with Volta likely hitting in Q4/Q1 next ear.
May be they are 2 different cards.It appears to me we have AMD's wooden screw fiasco in front of us. It's more likely the card Raja held was a mock up than all the slides and circulating info courtesy of AMD is false in regards to the PCIe connectors.
He took it out of an anti-ESD bag. Who'd go to that extent if it was just a mock-up?It appears to me we have AMD's wooden screw fiasco in front of us. It's more likely the card Raja held was a mock up than all the slides and circulating info courtesy of AMD is false in regards to the PCIe connectors.
LARGE CAVEAT: We only know it was done at 4K, not actual settings used
The only Vega Raja talked about is the Radeon Vega Frontier edition for professional market. That does not mean no Radeon RX VEGA launch in Q2. I think we will know for sure by Computex time if RX VEGA is coming in June. If it doesn't then surely its very disappointing AMD are almost a generation behind Nvidia. I don't know what AMD has planned for 2018 but I have a feeling it will be a repeat of Maxwell vs Radeon 300/Fury X as I think (actually almost sure) AMD might not have the funds to launch a full lineup of Vega GPUs at 14nm+ because they would be allocating their meagre cash resources to 7nm Navi GPUs due for H1 2019. AMD will have to find a way to generate more money and try to keep up with Nvidia's generations which are now not tied to process nodes and happen every 18-24 months.
http://pro.radeon.com/en-us/vega-frontier-edition/The Radeon Vega Frontier Edition graphics card is going to empower the pioneers creating the next generation of gaming experiences, but it does beg one question: Can you game on a Radeon Vega Frontier Edition? The answer is yes, absolutely. But because this graphics card is optimized for professional use cases (and priced accordingly), if gaming is your primary reason for buying a GPU, I’d suggest waiting just a little while longer for the lower-priced, gaming-optimized Radeon RX Vega graphics card. You’ll be glad you did.
http://www.anandtech.com/show/11403/amd-unveils-the-radeon-vega-frontier-edition
16GB HBM2, targeting pros first...
Already abbreviating Frontier Edition to FE. LOL
Your bias is so incredibly sad. Nvidia literally did the EXACT SAME THING with Volta and you offered up extreme praise at the marvels of nvidia's brilliance. You then actually, unbelievably, incredibly, offer damning criticism because it has MORE ram than people expected AMD to be able to offer with two stacks of hbm2.How ridiculous, the only Vega they seem to be readying for Q2 is Vega 10 aimed at the professional/prosumer market, and that 16GB of HBM2 probably indicates that it's going to be quite expensive.
For the consumer market, I don't get a good feeling about Vega.
Just have a look on the text on their Frontier Edition Page:
http://pro.radeon.com/en-us/vega-frontier-edition/
Frontier Edition comes Mid-End June. Waiting a while longer means no consumer cards in Q2.
Your bias is so incredibly sad. Nvidia literally did the EXACT SAME THING with Volta and you offered up extreme praise at the marvels of nvidia's brilliance. You then actually, unbelievably, incredibly, offer damning criticism because it has MORE ram than people expected AMD to be able to offer with two stacks of hbm2.
AMD previews Vega for the precise market that they need to win in order to actually make money and all you do is criticize it. What is wrong with you?
In the Volta thread you claimed to like ALL new technology and here AMD surprises with a stronger Vega that anticipated amd you call it "ridiculous".
Amazing.
For that to happen there must be some indication that Vega has dedicated FP16 units, and so far there is nothing that hints at such a possibility.To go out on a limb and speculate a bit here: might AMD be copying the Nvidia playbook even more than we're seeing here? How? I'll get to that in a minute, but here's my reasoning:
Previously, AMD/ATI has been a "gaming first, everything else second, third, or whatever really, we don't care, how about some games?" type of company. Before the Radeon Pro push, the FirePro line saw little adoption compared to Quadro cards, despite being far cheaper. Nvidia has since the Kepler days been pushing "pro-ish" cards (Titans) first on new architectures, and recently relegated the truly pro-level cards to dedicated silicon - due to the use/need for DP compute and the like, which takes up silicon area with no gains for most users.
Now, we see something similar from AMD: A Radeon Pro card being the first from a new architecture, with a somewhat inexplicably large die size (despite having 2x the SPs of an RX470, the die is significantly larger than 2x the 480/470 die - and that's without accounting for the die area savings due to HBM vs. GDDR5). AFAIK, the DP compute rate of Vega hasn't been disclosed, but I'd gather the extra hardware required for packed math does represent at least some of the die size increase, and it would be logical for a pro-level card to target something better than 1:32 DP compute rate - which would explain even more of the large die.
There used to be rumors of two Vega dice, yet in recent months, all we've been hearing of is the one presented to us yesterday in the Vega FE. Would it be entirely crazy to think the other Vega die might be a consumer-focused die, ditching packed math and whatever increased DP compute rate the VFE has, and using the freed-up die space and increased power efficiency to put in more CUs? After all, packed math has near zero value in the consumer space, as does DP compute. If they could push power efficiency to a point where we could have a 5-6000 SP Vega gaming card at tolerable power consumption at <600mm2, that would trounce anything Nvidia has to offer, at least if they can push it to 1500MHz or thereabouts. It would also make sense to launch it later, as a consumer focused "response" to Volta, while in the meantime selling partially disabled Vega ... 10? whatever the VFE has cores as a lower-end (and still reasonably competitive) stopgap until then. After all, it looks like even the VFE is reasonably competitive for gaming.
Of course, again, this is entirely unfounded speculation. But some times, dreaming is fun.
Reason is extremely simple. In emergin market like Deep Learning there is both very high margin, and very high volume. Focusing your efforts on this market with innovation is very good business decision.To go out on a limb and speculate a bit here: might AMD be copying the Nvidia playbook even more than we're seeing here? How? I'll get to that in a minute, but here's my reasoning:
Previously, AMD/ATI has been a "gaming first, everything else second, third, or whatever really, we don't care, how about some games?" type of company. Before the Radeon Pro push, the FirePro line saw little adoption compared to Quadro cards, despite being far cheaper. Nvidia has since the Kepler days been pushing "pro-ish" cards (Titans) first on new architectures, and recently relegated the truly pro-level cards to dedicated silicon - due to the use/need for DP compute and the like, which takes up silicon area with no gains for most users.
Now, we see something similar from AMD: A Radeon Pro card being the first from a new architecture, with a somewhat inexplicably large die size (despite having 2x the SPs of an RX470, the die is significantly larger than 2x the 480/470 die - and that's without accounting for the die area savings due to HBM vs. GDDR5). AFAIK, the DP compute rate of Vega hasn't been disclosed, but I'd gather the extra hardware required for packed math does represent at least some of the die size increase, and it would be logical for a pro-level card to target something better than 1:32 DP compute rate - which would explain even more of the large die.
There used to be rumors of two Vega dice, yet in recent months, all we've been hearing of is the one presented to us yesterday in the Vega FE. Would it be entirely crazy to think the other Vega die might be a consumer-focused die, ditching packed math and whatever increased DP compute rate the VFE has, and using the freed-up die space and increased power efficiency to put in more CUs? After all, packed math has near zero value in the consumer space, as does DP compute. If they could push power efficiency to a point where we could have a 5-6000 SP Vega gaming card at tolerable power consumption at <600mm2, that would trounce anything Nvidia has to offer, at least if they can push it to 1500MHz or thereabouts. It would also make sense to launch it later, as a consumer focused "response" to Volta, while in the meantime selling partially disabled Vega ... 10? whatever the VFE has cores as a lower-end (and still reasonably competitive) stopgap until then. After all, it looks like even the VFE is reasonably competitive for gaming.
Of course, again, this is entirely unfounded speculation. But some times, dreaming is fun.
I agree.One thing that was nice about this was confirmation of 1525MHz+. RIP 1200MHz leaks. Unless this Pro Vega card is specially binned and cherry picked from a bin full of lemons, we can expect consumer Vega will reach at least this.
Seeing 480 GB/s HBM2 is good and bad. Good: RIP 350 GB/s leaks. Bad: Still no sign of 512 GB/s. While surely not much of a difference in performance, from a brand strength standpoint it looks weak to have less than your predecessor. Consumer Vega, especially if it comes later could have faster.
16GB is good. Even though consumer will likely have 8GB, it means it is possible. My expectations: 8GB consumer Vega this year, with a refresh to 16GB next year to compete with Volta (note that I am not saying which Volta, since performance is still up in the air on Vega and Volta both).
Still so many questions. Hopefully Computex makes things more clear.
Nope.One thing that was nice about this was confirmation of 1525MHz+. RIP 1200MHz leaks.