Sapphire Radeon RX 480 Nitro 4GB and 8GB OC

[CAT-THE-FIFTH]

SNIP

I have a GTX960 which never got the async fix either,and Nvidia was selling the abilities of Vulkan,before AMD earlier in the year:

https://blogs.nvidia.com/blog/2016/0...-graphics-api/

The Vulkan Graphics API Is Here—and Your NVIDIA GPU Is Ready

That was in February - Maxwell was the current Nvidia GPU uarch back then!

I am running Doom on my card currently - I would love to know when I can actually run Vulkan and get async used.

I doubt it. OpenGL works better in practice too ATM from my own experience(not the only one to notice it on Maxwell).

The people on numerous forums kept on saying Maxwell can do async,and yet Nvidia has gone all quiet about it.

These are some Maxwell slides:

http://www.geforce.com/sites/default...port-640px.png



http://international.download.nvidia...pi-support.png



It says async compute is a feature of Maxwell,just like when Fermi and Kepler were getting DX12 support.

Another one a few months ago:

http://www.geforce.co.uk/whats-new/a...s-using-vulkan

Watch The GeForce GTX 1080 Run Doom at up to 200 FPS Using Vulkan
By Andrew Burnes on 2016-05-0

That was in May.

Its hilarious considering Bethesda and iD are more Nvidia biased devs and if Nvidia was shouting about Doom and Vulkan yonks ago,Vulkan must run fine,right??

Yet,still no updates.

For me that means something is not quite right still.

It was the same bumpf Nvidia said about the FX5800 being DX9 capable - technically yes but in reality not so good. Some of us recognise stalling from previous GPU launches.

What Nvidia has with Pascal is not really async,but two methods - load balancing and improved context switching.

Anybody following the roadmaps knows Pascal was a late addition - Volta was meant to replace Maxwell.

Pascal looks like a die shrunk Maxwell with some minor DX12 fixes and higher clockspeeds. Polaris is essentially the same uarch as the Fiji GPUs but die shrunk and with some DX11 fixes.

They both are going to both have similar strengths and weaknesses as the previous generations they are based on IMHO.

I have owned BOTH the GTX660 and GTX960 and they traded blows with the equivalent AMD cards of their generations,ie,the HD7870 and R9 285/R9 380. BOTH,the HD7870 and R9 285/R9 380 seemed to have lasted longer.

I expect the same of the RX480.

The only way Nvidia can look better in DX12,is by sponsoring more games and trying to gimp async compute to some degree to lower the advantage on AMD hardware,just like them using their higher tessellation throughput to damage AMD in Nvidia sponsored games.

But I also expect like with Maxwell having increasing issues with DX12,people kept saying for yonks a fix would come,until Pascal launched and then it was Pascal will get some fixes whilst Maxwell was forgotten. I expect the same with Pascal,and then it might or might not come and by then Volta will be launched. Volta will be probably fine.

After many of us experienced Rollo during the HD5000 and Fermi launches,no wonder many of us are dubious about how these things turned out.

[CAT-THE-FIFTH]

Regarding these cards,it would be nice if AMD had released for review a few days earlier - the 4GB Nitro for £200 looks a pretty decent card in terms of value for money compared to even the 8GB version. Better late than never,AMD??

Edit!!

Apparently,Gibbo on OcUK has said that the R9 480 Nitro is one of the most sucessful(if not the most sucessful) card model they have launched in years.

Also,part of one of his comments:

AMD also inform me over 100k units of RX 480 are shipping to AIB's in the next 10-12 days, so I think by middle of August (2 weeks), we shall be in a much better position and would of cleared the majority of backorders.

Another one:

Sapphire have nailed it, the sales show this as RX 480 Nitro 8G OC is our best selling card of all time for the time window it has being online, if the sales remain it will outsell our best selling SKU of all time which was the MSI 970 Gaming which sold close to 10,000 units in the 2 years it was available.

The Sapphire has a cracking start as it is already beyond 1000 units and sales momentum in last 48hrs is actually increasing because several reviews have being published that are all raving about how amazing it is.

So we are pushing to get this card to sell 10,000+ units within year 1.

[Tarinder]

Want a bit of inside info?

I have it on very good rumour that AMD charges AIBs more for RX 480 GPUs when used with 8GB memory. The same GPU, coupled with 4GB, is cheaper to buy. This is the GPU price, not the memory price.

This may help explain why there's a reasonable degree of price difference from 4GB to 8GB - it isn't just the AIB and etailer eating more profit.

[CAT-THE-FIFTH]

Are the GPUs for the 8GB version binned any better??

If not it seems very weird,as it gives the GTX1060 some more breathing space.

[Terbinator]

How... odd.

Surely AMD would want to complete the better-value angle by having 8GB cards in the hands of people rather than a 1060?

[kalniel]

Want a bit of inside info?

I have it on very good rumour that AMD charges AIBs more for RX 480 GPUs when used with 8GB memory. The same GPU, coupled with 4GB, is cheaper to buy. This is the GPU price, not the memory price.

This may help explain why there's a reasonable degree of price difference from 4GB to 8GB - it isn't just the AIB and etailer eating more profit.

How the flip do AMD know what memory the processor is going to be paired with though? Unless they aren't actually the same GPU and one has a memory controller better optimised for 8GB of faster RAM...

[CAT-THE-FIFTH]

How the flip do AMD know what memory the processor is going to be paired with though? Unless they aren't actually the same GPU and one has a memory controller better optimised for 8GB of faster RAM...

It could be that the GPUs which can use higher speed RAM are binned.

[Tarinder]

How the flip do AMD know what memory the processor is going to be paired with though? Unless they aren't actually the same GPU and one has a memory controller better optimised for 8GB of faster RAM...

AIBs have to tell AMD the intended framebuffer for each GPU.

[cheesemp]

That doesn't really answer the question now does it? Plenty of people murder...So I might as well too? ROFL. So because other sites don't get it right, hexus shouldn't either? My point was, if there seems to be a driver issue, it should always be reported. All sites turn off hairworks despite the fact it works on both sides. Does an NV user go home and do that (or even AMD when they can easily run 4x?)? No. I wouldn't go as far as testing in 64x tessellation but all cards perform about the same (~7-10% hit) at 4-8x and it's in just about the same amount of games as dx12 now. Maxwell2/pascal just happen to perform far better in this area, and that is a bonus of those cards, yet all sites ignore it also.

Nothing wrong with showing ALL sides, but acting like certain features for one side or the other doesn't exist? I say that as a current radeon 5850 owner, and I'd like to know how these things work when using every advantage each side has before I buy. That is the point of a review correct? Bad reporting is just bad reporting. If AMD's drivers aren't working in game X, then report it too. If there's a patch out and you're not testing with it, explain why. It's kind of hard to ignore the fact that bethesda says pascal doesn't work right yet in doom. ROTT only evens up above 1080p, and considering the mins, I'm pretty sure everything above that will go below 30fps. Neither of these cards are 1440p cards unless you like turning stuff down IMHO (and I don't, which is why I've been playing GOG games for a while...LOL). There is literally 1fps difference between OpenGL/vulkan on 1060 here. Would you want to buy a card thinking vulkan was better on AMD then get burned once NV turns on Async in doom etc? This site reported both with vulkan and without, so not sure why you bring in other sites. My comment on vulkan/doom is about this site not reporting a malfunction (perhaps they didn't know, hence the question of what is being reported).

I'd also like to see some Adobe apps tested since a lot of us use them. Pick whatever runs faster and report both sides (OpenGL, OpenCL or Cuda). It would be nice if 3dsmax/maya/blender was included too. There are a LOT of indie devs/content creators out there so it's a shame to ignore that these cards can be used for that stuff too. Personally I'm waiting for Vega/1080ti/Titan tests in this stuff before I make my move. Again, it's a shame so many sites act as though the only purpose of these cards is gaming. Students can get the entire Adobe collection for $20 a month, and even non students can get single apps for $10-20 (IE photoshop/lightroom $9.99/mo). Blender is free, and again great for Indie people. Heck a review of the coming Titan next week without this stuff is almost meaningless as it's aimed AT these people. That is why they removed the GTX moniker. Vega/1080ti should be tested in this stuff too. Can you save $400-600 (assuming vega/1080ti are far cheaper than titan) by going with these and still do well with content creation stuff? Do any of these get close to quadro M6000's in this stuff? I know I'm not the only one who wants to do more than game on my next card.

I think you missed my point that reviews happen with what is available at the time...

[kalniel]

AIBs have to tell AMD the intended framebuffer for each GPU.

Which makes sense if the GPUs are different (or have different firmware). Otherwise it's just licensing and all the associated bureaucracy that comes with it...

[cheesemp]

Which makes sense if the GPUs are different (or have different firmware). Otherwise it's just licensing and all the associated bureaucracy that comes with it...

No different from most software then! I work on very specialised software. The lite version is the same software just licensed differently to cut a few extras but is 1/10th the cost. Not surprised the same would start happening with hardware. Margins are tight for AMD so every penny will count...

[ET3D]

Kind of important to have the right stuff working before claiming a victor in Doom or Vulkan correct?

Reviews can only reflect the state at time of reviewing. AMD's latest driver claims 10% performance improvement in Rise of the Tomb Raider. Does that mean that reviews shouldn't have reviewed the game previously because of expected performance increase, or gamers not made a decision based on it?

Things always change. It's hard to predict in advance what the result will be. 'Working with NVIDIA' doesn't mean there will be a huge breakthrough in the future, it just means what it says. Perhaps we'll get a 5% increase in the future, who knows. That's fine, that's normal. There's no point in crying 'OMG, that's not representative'.

[scaryjim]

It could be that the GPUs which can use higher speed RAM are binned.

I suspect this is the case, tbh. We've seen 8GB RX 480s hit 8800Gbps on the memory. Potentially AMD are binning chips where the controller is borderline for 8Gbps to go to the 4GB cards, where the stock speed is lower. Couple that with smaller AND slower GDDR5 chips and you can see how that price differential can start building up...

[canopus72]

I genuinely like AMD and since custom models of 290/X were released, I have always been using AMD gpu's. I pre-ordered a pair of the RX480 nitro-OC from overclockers but I'm kinda disappointed. In fact, the powercolor red devil RX480 performs much better than the reference model RX480 (typically 15%) improvement. So why is the sapphire nitro RX480 not performing as well?. Also, the author didn't state if he was using standard nitro RX480 (8GB) or the nitro-OC RX480. I may cancel my order for the sapphire nitro and get the custom XFX RX480 (£270 and 5yrs warranty), but that will depend on benchmark results for XFX custom model. I don't want to get red devil because it has hexagrams and other satanic images on the gpu and on the box, which is disgusting.

[Tarinder]

The 8GB card is the OC version running at 1,342MHz core.

[gupsterg]

Initially was keen on Sapphire RX 480 Nitro, now preferring the XFX RX 480 DD. Mainly due to the DirectFets used, also quite liking the cooler on it.