Does the RTX series create an openning for AMD?

[amenx]

Others showed Vega more favorably:

https://www.computerbase.de/2017-08...#abschnitt_ultrahdbenchmarks_in_3840__2160_4k

But yes, Vega was still a disappointment from a gaming perspective. They threw a bunch of stuff in it for compute / professional use that bloated the die size and power use and some of the promised features to help the gaming side were MIA.

Vega was a smart design from a business standpoint. Killer mining card with a little gaming on the side. It served its purpose and AMD got their ROI on it. If it was a pure gaming card, may not have sold as well.

 

[amenx]

Nvidia not only needs to keep an eye on what AMD may come up with, but their traditional gaming base may one day just say to hell with it and consider consoles. See quite a few people think along those lines after seeing RTX prices.

 

[Head1985]

No. AMD doesn't have an opening. Mining and high prices have effectively killed high end PC gaming for me. Nvidia shows no signs of reigning in high prices with their new cards.

Consumers have shown they will never buy AMD cards in quantity no matter how competitive they are so why should they even try? Consumers want to overpay for Nvidia cards and overpay they shall.

I'm glad I got my 480 for $200 as I'll be stuck with it for some time. The result is my PC game purchases have dwindled to near zero (one. One game that isn't graphic intensive).

If AMD can deliver better architecture in every way just like ATI did in radeon 9700 to x800 period people will buy them.But since geforce 8800 launch they didnt have anything better than Nvidia.GCN is here already 8 years.They need be better in every way for atleast 2 generations just like ATI and they can have 60% market share.

 

[rainy]

GCN is here already 8 years.

Not exactly, first generation of GCN (HD 7000) was released on January 2012.
https://en.wikipedia.org/wiki/Radeon_HD_7000_Series

 

[Head1985]

Not exactly, first generation of GCN (HD 7000) was released on January 2012.
https://en.wikipedia.org/wiki/Radeon_HD_7000_Series

7970 launched in 2011.7 or 8 years that dont matter.GCN is old and is worse than nvidia architecture in every way for gaming.
Back in radeon 9700 to x800 era ATi was better in pretty much everything and thats why they were at 60% market share.

 

[PeterScott]

7970 launched in 2011.7 or 8 years that dont matter.GCN is old and is worse than nvidia architecture in every way for gaming.
Back in radeon 9700 to x800 era ATi was better in pretty much everything and thats why they were at 60% market share.

9700 Pro was my last ATI/AMD GPU, and it was quite awesome for the time.

 

[Geegeeoh]

GCN is here already 8 years.

GCN is a series of microarchitectures, like TeraScale and the Rx00 before it.
Vega it's not the same as Tahiti.

 

[Hans Gruber]

The radeon cards were not that far behind Nvidia cards compared to AMD CPU's before Ryzen for more than a decade.

 

[Tup3x]

If AMD can deliver better architecture in every way just like ATI did in radeon 9700 to x800 period people will buy them.But since geforce 8800 launch they didnt have anything better than Nvidia.GCN is here already 8 years.They need be better in every way for atleast 2 generations just like ATI and they can have 60% market share.

X800 wasn't anything special compared to GeForce 6 series. The lack of SM3.0 actually did bite back at them in the future. X1800 was rather meh product too. I unfortunately bought one... It used a lot of power and ran quite hot too. It broke after aa yea. Back then one year warranty was a thing and it happened to have one year warranty. X1900 series was much better though. But then Nvidia came out with GeForce 8...

 

[Hans Gruber]

i had a 9700pro, 9800pro,x1950pro, x1950XT and a 7950. Everything else has been Nvidia starting with the original Geforce 256 card.

 

[Guru]

Yes, but in the medium term, no in the short term. AMD's 7nm is going to all go towards Vega 20 that is means for the professional market at first, then their EPYC processors and thus we are only going to see a consumer level graphic card in Q2 2019. For Nvidia they can access some 7nm in Q2 2019, so at the earliest 7nm products from Nvidia in late Q3 2019.

 

[Genx87]

AMD will feel the pressure to implement hardware acceleration for Ray Tracing like NVIDIA, so many developers are getting excited for this tech and is being implemented in all of the major game engines .. If AMD doesn't take the step now, they will be outflanked by NVIDIA for the foreseeable future. AMD can't simply afford to fall behind on Ray Tracing the same way they did on Tessellation/Geometry or CUDA or proper compute or even AI. They are going to have to step up or rest in the dust.

This is pretty much it right here. AMD cant wait a generation on Ray Tracing. Even if games wont really take advantage of it. The perception of being behind will hurt them.

 

[Genx87]

If AMD can deliver better architecture in every way just like ATI did in radeon 9700 to x800 period people will buy them.But since geforce 8800 launch they didnt have anything better than Nvidia.GCN is here already 8 years.They need be better in every way for atleast 2 generations just like ATI and they can have 60% market share.

That is eye opening. I didnt realize the 4000 series while popular didnt even crack the 40% market share mark.

 

[Genx87]

X800 wasn't anything special compared to GeForce 6 series. The lack of SM3.0 actually did bite back at them in the future. X1800 was rather meh product too. I unfortunately bought one... It used a lot of power and ran quite hot too. It broke after aa yea. Back then one year warranty was a thing and it happened to have one year warranty. X1900 series was much better though. But then Nvidia came out with GeForce 8...

AIT\AMD threw away a lot of good will with the X800 production issues they didnt tell the world about. I think some people still have back orders waiting to be filled.

 

[krumme]

Amd is going to launch a vega gpu on a highend 7nm process node this year. With a borked ngg path that arch efficiently for consumer gaming should have brought them on par with nv 28nm Maxwell.
What a waste, but with nv crazy prices all over the b2b market is certainly going to be happy to have some compettition. Goes to show the insane margins on this market that it makes sense to launch a beeding edge 7nm vega gpu solely for that pro market.
Hopefully navi is not long away.

 

[Guru]

If AMD can deliver better architecture in every way just like ATI did in radeon 9700 to x800 period people will buy them.But since geforce 8800 launch they didnt have anything better than Nvidia.GCN is here already 8 years.They need be better in every way for atleast 2 generations just like ATI and they can have 60% market share.

Stop using fallacies and lies already! Nvidia is using the same Fermi architecture at the core, just because they call it different name doesn't mean it is. GCN has been massively redesigned over and over again. GCN 4 has almost nothing in common with GCN 1.

Stop saying AMD are using the same architecture, you are purposely lying. Stop claiming Nvidia are using a different one each time as well, at the core its based on the Fermi architecture. Even Turing shares its roots with Fermi. So stop with your absurd lies!

Fact of the matter is that AMD had better graphic cards all the time. Even the 200 series were better than the Nvidia 600 and 700 series. Nvidia's 400 series sucked, 500 improved them a bit, but ultimately it still sucked, 600 and 700 were much better, but AMD did much better with their series.

The only good graphic cards between Nvidia's 400 to 700 series were the GTX 460 768mb and 1GB. Later on the GTX 780 and 770 were good, but not great.

 

[Head1985]

Stop using fallacies and lies already! Nvidia is using the same Fermi architecture at the core, just because they call it different name doesn't mean it is. GCN has been massively redesigned over and over again. GCN 4 has almost nothing in common with GCN 1.

Stop saying AMD are using the same architecture, you are purposely lying. Stop claiming Nvidia are using a different one each time as well, at the core its based on the Fermi architecture. Even Turing shares its roots with Fermi. So stop with your absurd lies!

Fact of the matter is that AMD had better graphic cards all the time. Even the 200 series were better than the Nvidia 600 and 700 series. Nvidia's 400 series sucked, 500 improved them a bit, but ultimately it still sucked, 600 and 700 were much better, but AMD did much better with their series.

The only good graphic cards between Nvidia's 400 to 700 series were the GTX 460 768mb and 1GB. Later on the GTX 780 and 770 were good, but not great.

Haha lol
AMD didnt do anything sice hawaii to be honest.They are still on 4x Shader engines with 64rops and are stuck with 4096SP with huge bottlenecks everywhere .ALL that new stuff on vega like primitive shaders and draw stream binning rasterizer dont work and will never work.Sure with polaris they claim +15% IPC vs 290x but that was not true.In real life it was 6% since hawaii https://www.computerbase.de/2016-06...iagramm-performancerating-polaris-architektur
On NV side:Fermi to kepler they removed hotclocks and went from 512SP on GTX580 to 2808SP on 780TI(Big fermi vs big kepler) so more than 5x.They also increased TMU from 64 on GTX580 to 240 on 780TI.

Kepler to maxwell:They doubled Rops on all cards.Reduced Shaders per SM from 192 on kepler to 128 on maxwell.Reduced TMU per sm from 16 on kepler to 8 on maxwell.They introduced new tile based rendering and new delta color compression.They also increased number of GPC from 5x on 780TI to 6x on TITANX and 980TI.

Maxwell to pascal:again new delta color compression + focus on high clocks

 

[JustMe21]

Of course, the 80 series of cards didn't keep the same price and if you compare similarly spec'd cards, not model numbers, the jump in performance isn't as dramatic. I do hope AMD and Intel eventually have some strong offerings, because the Nvidia price creep is getting ridiculous.

 

[AtenRa]

Haha lol
AMD didnt do anything sice hawaii to be honest.They are still on 4x Shader engines with 64rops and are stuck with 4096SP with huge bottlenecks everywhere .ALL that new stuff on vega like primitive shaders and draw stream binning rasterizer dont work and will never work.Sure with polaris they claim +15% IPC vs 290x but that was not true.In real life it was 6% since hawaii https://www.computerbase.de/2016-06...iagramm-performancerating-polaris-architektur

I found you completely wrong here,

Just for Polaris, the changes over Hawaii were dramatic.
Polaris with LESS on everything except for higher clocks is faster than Hawaii.

R9 390X
2816 cores
176 TMUs
64 ROPs
512 bit Memory

RX 480
2304 cores
144 TMUs
32 ROPs
256 bit Memory

Also , for Vega 10 (directly from AT review)

https://www.anandtech.com/show/11717/the-amd-radeon-rx-vega-64-and-56-review/3

From an architectural standpoint, AMD’s engineers consider the Vega architecture to be their most sweeping architectural change in five years. And looking over everything that has been added to the architecture, it’s easy to see why. In terms of core graphics/compute features, Vega introduces more than any other iteration of GCN before it

Vega 10 half precision (FP16) even today is unmatched in mm2/perf by any NVIDIA offering. It doesnt have a lot of impact in gaming but for AI learning and other compute workloads that need half precision they are clearly in the lead as mm2/perf goes.

 

[Geegeeoh]

AMD didnt do anything sice hawaii to be honest.

This lack of knowledge is embarrassing...

 

[Head1985]

I found you completely wrong here,

Just for Polaris, the changes over Hawaii were dramatic.
Polaris with LESS on everything except for higher clocks is faster than Hawaii.

Yes by 6%.Did you read my post i posted link to this so for you again.So 6% since 2013..great...
https://www.computerbase.de/2016-06...iagramm-performancerating-polaris-architektur

Also did you read my post abou vega?So again: ALL that new stuff on vega like primitive shaders and draw stream binning rasterizer dont work and will never work

Sad true is that AMD is stuck on 4x shader engines/64rops/256TMU/4096SP.That is pretty unbalanced setup.
They need use 6x shader engines just like Nvidia using 6x GPC since maxwell.This will allow them increase rops to 96 and increase geometry power and pixel fillrate by 50%.
6x shader engines will also increase shader utilization because each shader engines will have less shaders in there so less bottlenecks.

Look at vega 56OC vs 64OC they have pretty much same performance.So what those 512SP doing?Well nothing because they are bottleneck by 4x shader engines.
https://www.gamersnexus.net/guides/3053-vega-64-vs-vega-56-clock-for-clock-shader-differences

This should amd do after hawaii yet we are in 2018/19 and they are still stuck at 4xshader engines.
Best thing they should do now is just create new architecture from scratch it is too late for gcn.

 

[AtenRa]

Yes by 6%.Did you read my post i posted link to this so for you again
https://www.computerbase.de/2016-06...iagramm-performancerating-polaris-architektur

Again, 6% higher performance with less Shaders, half ROPs and less TMUs doesnt translate in to only 6% IPC increase.

R9 390X
2816 cores
176 TMUs
64 ROPs
512 bit Memory

RX 480
2304 cores
144 TMUs
32 ROPs
256 bit Memory

You have to compare R9 390 vs RX 480 at the same clocks to be even a little closer to apples to apples.

R9 390
2560 cores
160 TMUs
64 ROPs
512 bit Memory

RX480 IPC is closer to 15% over Hawaii.

Also did you read my post abou vega?So again: ALL that new stuff on vega like primitive shaders and draw stream binning rasterizer dont work and will never work

Sad true is that AMD is stuck on 4x shader engines/64rops/256TMU/4096SP.That is pretty unbalanced setup.
They need use 6x shader engines just like Nvidia using 6x GPC since maxwell.This will allow them increase rops to 96 and increase geometry power and pixel fillrate by 50%.
6x shader engines will also increase shader utilization because each shader engines will have less shaders in there so less bottlenecks.

Look at vega 56OC vs 64OC they have pretty much same performance.So what those 512SP doing?Well nothing because they are bottleneck by 4x shader engines.
https://www.gamersnexus.net/guides/3053-vega-64-vs-vega-56-clock-for-clock-shader-differences

This should amd do after hawaii yet we are in 2018/19 and they are still stuck at 4xshader engines.
Best thing they should do now is just create new architecture from scratch it is too late for gcn.

VEGA 10 is not a 100% gaming only Chip, it is unfortunate that AMD didnt have the R&D resources to create VEGA 10 for the Professional and Server/Cloud market and a strip down more Gamer oriented Chip. All those extra features and Compute performance is not translated in to current gaming performance unfortunately. Perhaps Navi will be better, lets wait and see how the landscape will be in 2019/2020. Because I see NVIDIA is taking exactly the same road with the new RT line.

 

[Head1985]

Again, 6% higher performance with less Shaders, half ROPs and less TMUs

LOL again you didnt read that test.
390x at 1035/2000=5.8TF and 256gb/s
rx480 at 1266/4000=5.8TF and 256gb/s
You know they downclock 390x they are not idiots

 

[AtenRa]

LOL again you didnt read that test.
390x at 1035/2000=5.8TF and 256gb/s
rx480 at 1266/4000=5.8TF and 256gb/s
You know they downclock 390x they are not idiots

Why compare only TFs for gaming when ROPs are halved, TMUs are less and memory bandwidth almost half and yet Polaris is faster.

As i have said before, you have to compare against R9 390 at the same clocks vs RX480 and see how much faster Polaris is in gaming.
Comparing only TFLOPs for gaming is wrong.

 

[crisium]

Of course Polaris has performance-per-flop increases if everything else is equal. 470 vs 380X vs 7970 comparisons have shown this. The problem is it's not equal. 2304:32 is unbalanced. 256-bit bus with 8 Gbps memory is also not enough. Hawaii as configured out of box (2816:64:512-bit), still offers more performance-per-flop than any other GCN.

2816:64:384-bit (don't think 512-bit is needed thanks to compression) with Polaris architecture would likely be a 1070 competitor and a smaller, cheaper to build chip than Vega 56. Fiji and Vega show that GCN really cannot scale to efficiently benefit from anymore shaders. It's a shame they never tried to make this. Also it is odd they haven't bothered to make a dGPU based on the consoles. 2560:32:384-bit Xbox 1 X, even if very ROP unbalanced, would obviously be faster than the RX 580. It would still be far behind a 1070 but it could easily be a RX 590 and priced higher than a GTX 1060. Original Xbox 1 GPU found its way to cards, so it is possible.