AMD's Initial Production Run of Radeon VII Just 5,000 Pieces, Company Denies it

[Midland Dog]

Yes, because rasterization and ray-tracing are the only two functions of a video card
Both Radeon (VII) and RTX-series offers features the other doesn't

but the RTX card gives you more features, im personally happy on pascal yet i would love to hop on turing, if dlss gets support at all resolutions and for the new h265 encoder are reasons that i will jump, not %100 sold on RT, i will wait and see when we have more titles to decide that, as long as it results in a nice looking game im happy

 

[Kaotik]

but the RTX card gives you more features, im personally happy on pascal yet i would love to hop on turing, if dlss gets support at all resolutions and for the new h265 encoder are reasons that i will jump, not %100 sold on RT, i will wait and see when we have more titles to decide that, as long as it results in a nice looking game im happy

It's funny, usually same people are "we so need RT because faking things with rasterization sucks" and "we so need DLSS because rendering at lower resolution and faking higher is great"

Have you actually counted which one offers "more features" over the other?

 

[EarthDog]

Have you actually counted which one offers "more features" over the other?

What features does it have over RTX?


16GB isn't a feature for a gaming card (much over 8 isn't honestly). It is part of what makes this GPU pricey for very little/no reason (for gamers).

It can never do DLSS/RT. Regardless of market support in a year or two, never is forever here. RT may not be needed now, but for the similar performance and similar price...

It uses more power, makes more heat, and will need additional mitigation (noise) to remedy this.

Radeon can compute better by far... but, this is supposed to be for gaming. How does that help gaming?

AMD lost its way (price to performance ratio) with this card.

 

[moproblems99]

It can never do DLSS/RT.

Is that true? I suspect the answer to be yes but is there enough unused grunt that could be re-purposed or is this an architectural restriction? I am suspecting the latter or they would have done so by now.

 

[EarthDog]

I don't think its the grunt so much as it is the specialized hardware to handle it with aplomb.

 

[lexluthermiester]

AMD lost its way (price to performance ratio) with this card.

I have to disagree with this. The projected performance numbers strongly suggest that this series of GPU will fill a gap in AMD's lineup in a good way.

 

[EarthDog]

I have to disagree with this. The projected performance numbers strongly suggest that this series of GPU will fill a gap in AMD's lineup in a good way.

It will, but I'm talking price to performance ratio... not a gap in their lineup. Different metrics.

 

[lexluthermiester]

It will, but I'm talking price to performance ratio...

I still can't agree. As long as the performance is there, the price is reasonable. Not great but good.

 

[EarthDog]

I still can't agree. As long as the performance is there, the price is reasonable. Not great but good.

People complain about the 2080s price/performance ratio (nvidia in general) and this thing (supposedly) performs on par with it and priced with it. This is NOT typical AMD who destroys in price to performance ratio. We dont even see the AMD people parroting this metric.

This would be a reasonable and more typical AMD card at 500. Too much vram and compute.

 

[lexluthermiester]

This is NOT typical AMD who destroys in price to performance ratio.

True, but the price is not set it stone yet, so it's a waiting game.

This would be a reasonable and more typical amd card at 500. Too much vram and compute.

But that's the point, with this card AMD is aiming both at the gaming market which will give some future-proofing and in the compute market which tends to be memory intensive. With this one card AMD wants to compete with Titan/Quadro but at a much better price point. And in these capacities I think they will succeed.

 

[hat]

h265 encoder? Don't recent Intel chips already do that?

 

[EarthDog]

True, but the price is not set it stone yet, so it's a waiting game.

But that's the point, with this card AMD is aiming both at the gaming market which will give some future-proofing and in the compute market which tends to be memory intensive. With this one card AMD wants to compete with Titan/Quadro but at a much better price point. And in these capacities I think they will succeed.

The compute ability seems like a byproduct, honestly. It is chopped down from MI50 (of course). But if you want compute, how does Vega64 compare to this card at ~2/3 the price?

 

[efikkan]

People complain about the 2080s price/performance ratio (nvidia in general) and this thing (supposedly) performs on par with it and priced with it. This is NOT typical AMD who destroys in price to performance ratio. We dont even see the AMD people parroting this metric.

This would be a reasonable and more typical AMD card at 500. Too much vram and compute.

Typical? I haven't seen AMD destroy Nvidia in price to performance ratio in general in years. Recent higher models from AMD seem to follow Nvidia. Fury X was priced exactly like GTX 980 Ti, and Vega 64 was price matched at launch with GTX 1080. GTX 1060 and GTX 1070 were better buys than RX 580 and Vega 56 respectively. In the past 2-3 years, it's only been the sub $200 range that AMD has offered better value than Nvidia. So unless you are talking about the old ATI days, AMD haven't really been that much cheaper lately.

We still don't know the actual performance of Radeon VII. How it's supposed to match RTX 2080 with just 25% more performance than Vega 64 remains a mystery to me. And we have to remember that AMD never said explicitly that it was matching RTX 2080, but just showed some cherry-picked benchmarks and the price.

 

[EarthDog]

Yes. Typical. RX series cards... also, 1080 was 599/699 while V64 was 499/599. So, yes. Perhaps destroy was a bit too much... however they typically win across the board in price to performance. This is nothing new to hear.

We still don't know the actual performance of Radeon VII. How it's supposed to match RTX 2080 with just 25% more performance than Vega 64 remains a mystery to me. And we have to remember that AMD never said explicitly that it was matching RTX 2080, but just showed some cherry-picked benchmarks and the price.

Tell me something I don't know.

 

[efikkan]

Yes. Typical. RX series cards... also, 1080 was 599/699 while V64 was 499/599. So, yes. Perhaps destroy was a bit too much... however they typically win across the board in price to performance. This is nothing new to hear.

No, not really. The MSRP of GTX 1080 was lowered to $499 months ahead of Vega 64.

 

[Kaotik]

What features does it have over RTX?

Radeon Chill, FRTC, Radeon Overlay which includes now every damn bell and whistle from Radeon Software, automatic OC, automatic undervolt, built-in benchmark function for any game, different AA modes etc etc
I'm aware that some of these are available for NVIDIA via 3rd party solutions, but NVIDIA doesn't offer them itself and thus shouldn't be counted as features for their cards.

It can never do DLSS/RT.

It supports DirectML and ROCm based machine learning APIs which can be utilized to do DLSS (or rather, DLSS-like tricks, of course NVIDIA won't give their exact implementation over to others)
It also supports RadeonRays which is RT, it could support DXR given the drivers (it might or might not be fast enough, we won't know until such drivers arrive if ever)

Radeon can compute better by far... but, this is supposed to be for gaming. How does that help gaming?

Games use more and more compute each day over traditional methods, so of course it helps.

 

[EarthDog]

Radeon Chill, FRTC, Radeon Overlay which includes now every damn bell and whistle from Radeon Software, automatic OC, automatic undervolt, built-in benchmark function for any game, different AA modes etc etc
I'm aware that some of these are available for NVIDIA via 3rd party solutions, but NVIDIA doesn't offer them itself and thus shouldn't be counted as features for their cards.

Wattman and Chill are pieces of additional software. Why does it matter that AMD made it or 3rd party. Not a feature over NV...

It supports DirectML and ROCm based machine learning APIs which can be utilized to do DLSS (or rather, DLSS-like tricks, of course NVIDIA won't give their exact implementation over to others)
It also supports RadeonRays which is RT, it could support DXR given the drivers (it might or might not be fast enough, we won't know until such drivers arrive if ever)

Does it have the separate HW to support these functions? No.

Games use more and more compute each day over traditional methods, so of course it helps.

I wait with bated breath to see this in action...

No, not really. The MSRP of GTX 1080 was lowered to $499 months ahead of Vega 64.

Really. MSRP from Day 1 is the thing here. How the market reacts to other entries (like lowering the price because of the pricing of V64) really shouldn't be considered IMO. It was a market response to a card that had a much better price to performance ratio. Typical of AMD.

 

[Kaotik]

Wattman and Chill are pieces of additional software. Why does it matter that AMD made it or 3rd party. Not a feature over NV...

Software, (partly) yes, but still features available for the cards. Of course if moving goalposts is a thing one could argue that nothing can be counted as feature over the other.

Does it have the separate HW to support these functions? No.

Does it need to have separate HW to support those functions? No.

I wait with bated breath to see this in action...

Which part? Compute? You've already seen it for years in action, the amount just keeps climbing because compute is more flexible solution than traditional shaders etc

 

[EarthDog]

LOL nobody is moving the goal posts... I simply don't call software (the same software functionality), via inhouse or 3rd party to be any different. They both have software that allows for full control over the cards. You remind me of the marketing guys for the AMD software presentations.

It may function, but does it function as well as dedicated hardware to it? No (likely not). Apologies there as I know I didn't mention HW initially. But yeah, no dedicated hardware to do it likely = worse performance.

If compute mattered as much in games as you say it does, AMD cards would be cleaning the joint up. They aren't.

 

[Kaotik]

LOL nobody is moving the goal posts... I simply don't call software (the same software functionality), via inhouse or 3rd party to be any different. They both have software that allows for full control over the cards. You remind me of the marketing guys for the AMD software presentations.

Most features are software in the end, where do you draw the line? Dedicated hardware for that specific function when your GPU is full of general purpose hardware? Quite limited view.

It may function, but does it function as well as dedicated hardware to it? No (likely not). Apologies there as I know I didn't mention HW initially. But yeah, no dedicated hardware to do it likely = worse performance.

Possibly worse performance (it really isn't that clear cut difference, it's dependent on several factors), but also more flexibility.

If compute mattered as much in games as you say it does, AMD cards would be cleaning the joint up. They aren't.

I didn't say how much it matters in games, I said it's used more and more all the time which plays into AMD's strenght more than NVIDIAs. You seem to think everything happens overnight and as soon as some feature benefits the other it should crush the competition on all fronts.

 

[EarthDog]

Clearly the line is drawn with software... lol, I don't see how NV cards having the same software functionality as AMD could be a feature for AMD over NVIDIA. Source isn't relevant... the ability is.

I'm a betting man... LIKELY. When its used and that flexibility can be leveraged... I'm with ya! When... as it stands they a non player in the gaming RT arena.

When it compute matters enough to make a difference, let me know.

Actually, I don't think things happen overnight. If you read any of my posts in the NV RTX threads, you can see I was one of few voices there who begged for patience on RT tech/implementation. So, like it typical with AMD we have to wait to see if how they do things pan out. I'm still waiting for Vulkan to get traction, for an example. Mantle before it? I realize these are APIs but my point remains. I also don't hedge my bets on the future either. I have always told users both here and my site not to get RTX cards for RT capabilities alone. Buy it on its current performance both with and without. Perhaps this year we will see more titles with it and better implementations. Only time will tell... but regardless of AMD/NV/Intel, etc... I don't hold my breath long.

I don't put any check boxes in future implementations. But to each their own and we will agree to disagree.

 

[Kaotik]

Clearly the line is drawn with software... lol, I don't see how NV cards having the same software functionality as AMD could be a feature for AMD over NVIDIA. Source isn't relevant... the ability is.

Some of the said features are available for NVIDIA, not all even with 3rd parties. And if you just think about the ability, then RT isn't one since Radeons can do that via RadeonRays (and other solutions as well, just not DXR at this time) and DLSS can be done as soon as someone writes the software to do it.

I'm a betting man... LIKELY. When its used and that flexibility can be leveraged... I'm with ya! When... as it stands they a non player in the gaming RT arena.

Gaming DXR arena, not Gaming RT arena. Claybook is heavily RT based game and runs just fine on Radeons (and older GeForces).

When it compute matters enough to make a difference, let me know.

It already makes a difference, how big the difference is could only be measured if someone did separate path for their game which would use just traditional graphics shaders, but obviously no-one would do that since there's no benefit from it.

 

[EarthDog]

Yaaaaaaaawn. Streeeeeetch...Wake me when its ready and benchmarked.

 

[lexluthermiester]

how does Vega64 compare to this card at ~2/3 the price?

And half the memory.. That's the key point. With a very large pool of ultra high speed memory, the GPU will need to access system RAM much less, thus increasing process speeds, dramatically in some cases. It's all about how you factor in certain variables.

 

[Kaotik]

@EarthDog related to earlier, NVIDIA has actually demonstrated DLSS-like solution running on DirectML, supposedly running on unnamed iGPU (meaning either Intel or AMD)
http://on-demand.gputechconf.com/si...-gpu-inferencing-directml-and-directx-12.html around 14:30 or so