Radeon VII

[Gerard]

Your right, AMD basically make their data cards backwards compatible for gamers. It’s a bit like strapping a truck engine onto a skateboard and entering it into an F1 race. Yes it will keep up but you wouldn’t design it that way unless you had no other choice.

Well they did that for Vega (presumably because of the stricter Development budget), I don't think it was the case for cards a few years beforehand?

 

[CAT-THE-FIFTH]

I would certainly expect GTX1070/GTX1070TI level performance under £300 by now. Looks like it won't be coming too soon,especially since we seem to be getting GTX1080TI competitors at similar price from AMD and Nvidia around 2 years later.

 

[LePhuronn]

I highly doubt that considering Radeon VII is Vega+25% performance there is no way they would sell their own card with only 10% less performance than their top tier for half the price

This is why the prices in the leak were the most dubious part. But also bear in mind that this leak is about 6 weeks old and before there was any renewed talk about a 7nm Vega gaming card.

Also consider the Radeon 7 is just a space filler. Right now $699 is just an arbitrary figure AMD have slapped onto a card that would've just ended up in the bin. As 7nm Vega yields improve we could see the price come down, if AMD even bother to keep selling it. Who knows, the occasional mention of a big Navi might be something too which replaces the Radeon 7.

I should imagine we'll have Navi information around Computex, it'd be nice to talk new graphics cards to go with the new CPUs. And if we do get more on Navi at Computex then I should imagine we'll get some more substantial leaks before then.

 

[dragonslayer]

This is why the prices in the leak were the most dubious part. But also bear in mind that this leak is about 6 weeks old and before there was any renewed talk about a 7nm Vega gaming card.

Also consider the Radeon 7 is just a space filler. Right now $699 is just an arbitrary figure AMD have slapped onto a card that would've just ended up in the bin. As 7nm Vega yields improve we could see the price come down, if AMD even bother to keep selling it. Who knows, the occasional mention of a big Navi might be something too which replaces the Radeon 7.

I should imagine we'll have Navi information around Computex, it'd be nice to talk new graphics cards to go with the new CPUs. And if we do get more on Navi at Computex then I should imagine we'll get some more substantial leaks before then.

I agree that Radeon VII won't last long personally I'm only expecting it to be around for 6-8monthsand I get the feeling they only released it so they could pretend to be competitive with Nvidia especially after they had a rocky launch with the 20xx series

The only thing I disagree on is the price and performance I think it'll perform far better than we are all expecting (35%+ better than vega64) and it'll cost £400-500 for the top SKU

 

[LePhuronn]

they only released it so they could pretend to be competitive with Nvidia

There's nothing pretend about it. We'll get confirmation obviously when reviews come out, but if the Radeon 7 matches the RTX 2080 in performance then it's not pretending, it is competitive. As to AMD as a company pretending to compete? Well that's a fallacy too given Vega 64 and downwards matches or beats the Nvidia equivalent. It's only the obscene halo parts where AMD can't bring anything to the table, but let's be honest here why should they bother? AMD are in the business to make money, not bring Nvidia's prices down, and if no money can be made by fighting for the top spot then why actually try, especially as history has proven time and again AMD's superior products still get outsold by Nvidia.

he only thing I disagree on is the price and performance I think it'll perform far better than we are all expecting (35%+ better than vega64) and it'll cost £400-500 for the top SKU

You talking about the RX 3080? If AMD can wring another 35% out of GCN for GCN 6 then awesome, but £400 notes is ludicrous money. £500 will make it DOA. Let's say it happens this way, we'll get exactly the same questions being asked as those about the Radeon 7 now: if you can get the potential of ray tracing and DLSS for an extra £50, would you bother with the AMD card?

That's what £400 will do to the RX 3080; you're matching the RTX 2070 in raster performance for only £50-80 less. Now to some that is a good saving, to some it's nothing so they'll jump on RTX's potential. If it's £500 then that's more expensive than the RTX 2070 so literally no point in bothering with Navi, especially as the FreeSync advantage is starting to get eroded by Pascal and Turing supporting Adaptive Sync (after a fashion). Look at it this way:

If raster performance was identical, but AMD was pitched at £50 less than RTX, would you go Nvidia for the extra features and potential?
But if raster performance was identical and AMD was pitched at £100-150 less than RTX, would you still go Nvidia?

There's no way that leaked RX 3080 is coming out at £250, but even as high as £350 it is still a much more attractive proposition against the RTX 2070, which is the target competitor. The minute you go north of £400 then the argument starts swining in Nvidia's favour. At £500 you're smacking your head against the wall, lamenting AMD screwing up again.

Part of Ryzen's success has been the price/performance ratio (as is always the case with AMD). You get near Intel performance without the price tag and it's shaken things up a bit. If Ryzen 3000 proves to be as good as demonstrated, the CPU market gets torn asunder. AMD need to do the same with Navi. It's not going to set the world alight, but aggressive, realistic pricing will be massively disruptive. Hell, why do you think these GTX 11xx rumours have started up?

 

[Panos]

Your right, AMD basically make their data cards backwards compatible for gamers. It’s a bit like strapping a truck engine onto a skateboard and entering it into an F1 race. Yes it will keep up but you wouldn’t design it that way unless you had no other choice.

But we use those cards in PCs not consoles..... So all the extra non gaming horse power is needed to many of us, who cannot afford a £2000+ professional card, which cannot be used on gaming.

And also more future proof with larger and faster memory

Not that much future proof
For two reasons.

a) Blender & Unity eat up 16GB before you even realize it. At least AMD supporting HBCC and direct RAM access through it's GPGPU stuff, makes the issue almost non existent even with the 8GB Vega 64.

Something we cannot say for Nvidia on the other hand, which has all these features locked to their professional cards only, making their "consumer" products useless on anything but gaming and some light work.

b) Skyrim + graphical mods with ENBoost to remove the 10GB cap
(cap is 10GB of VRAM+RAM)

 

[D.P.]

You have written gazillion times over the past year "AMD lacks Tensor cores cannot do XYZ". Yet you completely ignore the fact that AMD doesn't need to have dedicated cores to do such things. For it's design over GPGPU is great.
And the best example is TensorFlow matrix calculations (like CIFAR10) with ROCm vs CUDA. Where the CUDA implementation is using heavily the Tensor Cores on a card like the TV100 (top of the range Volta for business), a humble Vega 64 with it's pitiful 8GB VRAM, keeps up at 90% of it's performance. And wrote humble because it costs 1/8 the price of the TV100.

Stop blindly believing what ever Nvidia marketing says please.

You keep referring to flawed benchmarks, and you keep ignoring the fact that the lack of dedicated tensor cores gives Nvidia a 10x speed up over AMD GPus in tests like imagnet. Nvidia also doesn't need tensor cores to be competitive, their CUDA cores combined with the CUDA platform is significant faster and more efficient than ROCm.


But the biggest issue is that simply no wants to use ROCm, everyone far prefers TensorFlow and CUDANN, ROCm support is just lightyears behind CUDA, and nvidia basically define this industry.

 

[Brazo]

And also more future proof with larger and faster memory

Again by accident not design.

The truck engine on a skateboard will offer better mpg. Again by accident not design!
A problem of this accident is that AMD will struggle to dip below 16gb for future cards (with similar performance)as they have set a precedent!

 

[Brazo]

Well they did that for Vega (presumably because of the stricter Development budget), I don't think it was the case for cards a few years beforehand?

Your right it’s only recent card that follow this trend.

 

[Brazo]

I would certainly expect GTX1070/GTX1070TI level performance under £300 by now. Looks like it won't be coming too soon,especially since we seem to be getting GTX1080TI competitors at similar price from AMD and Nvidia around 2 years later.

We did briefly get v56 @ £299

 

[Illuminist]

We got Fury X which was £600 + and was 1060 performance a few years back, its not that different to now except there is a tier above the "top" card now which is basically a gaming titan called the 2080ti now

 

[greatash]

We did briefly get v56 @ £299

Yep and they are £310-320 now for some models. Sell the games that comes with it and it comes to £250 or less, best price/performance card out there for 1440p by far.

I dont know why most people ignore it.

 

[Illuminist]

Yep and they are £310-320 now for some models. Sell the games that comes with it and it comes to £250 or less, best price/performance card out there for 1440p by far.

I dont know why most people ignore it.

Plus flash it to a 64 bios and have almost the same performance as a 64 for no cost, i personally think the vega 56 is genius if only nvidia owners and their brainwashing realised they could be playing at 1080 levels for £300 ish with 3 free AAA games

 

[Ian Evey]

Plus flash it to a 64 bios and have almost the same performance as a 64 for no cost, i personally think the vega 56 is genius if only nvidia owners and their brainwashing realised they could be playing at 1080 levels for £300 ish with 3 free AAA games

Old news I’m afraid, by the time the 56 dropped in price enough the 1070/1080 were better options so not many people with either of those cards would buy a 56.

 

[Panos]

You keep referring to flawed benchmarks, and you keep ignoring the fact that the lack of dedicated tensor cores gives Nvidia a 10x speed up over AMD GPus in tests like imagnet. Nvidia also doesn't need tensor cores to be competitive, their CUDA cores combined with the CUDA platform is significant faster and more efficient than ROCm.


But the biggest issue is that simply no wants to use ROCm, everyone far prefers TensorFlow and CUDANN, ROCm support is just lightyears behind CUDA, and nvidia basically define this industry.

Just an FYI because you are confused. There is TensorFlow ROCm port, where the benchmarks posted before coming from (old version from April last year) comparing the initial release between TensorFlow CUDA and TensorFlow ROCm. Currently supports 1.9.2 and next release 2.0.

Please keep up with the tech

 

[D.P.]

Just an FYI because you are confused. There is TensorFlow ROCm port, where the benchmarks posted before coming from (old version from April last year) comparing the initial release between TensorFlow CUDA and TensorFlow ROCm. Currently supports 1.9.2 and next release 2.0.

Please keep up with the tech

FYI, I run Tensorflow use 32-way NV100 clusters and TF1.12

 

[Illuminist]

Just an FYI because you are confused. There is TensorFlow ROCm port, where the benchmarks posted before coming from (old version from April last year) comparing the initial release between TensorFlow CUDA and TensorFlow ROCm. Currently supports 1.9.2 and next release 2.0.

Please keep up with the tech

FYI, I run Tensorflow use 32-way NV100 clusters and TF1.12

You guys crack me up XD

 

[CAT-THE-FIFTH]

We did briefly get v56 @ £299

Briefly being the issue with short term pricing. We should be getting that level of performance at £170 to £250 by now as standard. Just look at what you could have got from 2015 onwards at £170 to £250,its R9 290/R9 390/GTX970 level performance and we still get that level of performance now,and its been years.

You guys crack me up XD

The one-upmanship is hilarious. It reminds me of this:
https://www.youtube.com/watch?v=VKHFZBUTA4k

 

[Panos]

FYI, I run Tensorflow use 32-way NV100 clusters and TF1.12

Unfortunately due to budget 2 Vega 64s

 

[Panos]

That might explain why from "No Vega 20 for consumers" to having Radeon VII.