Radeon R9 290X Review, new GPU
 

[url]http://www.tomshardware.com/reviews/radeon-r9-290x-hawaii-review,3650.html[/url]
28nm,6.2B transistors, 1Ghz GPU,2816 shaders, supposed to have 5.6 Tflops (titan/780 is supposed to have 4.5/4.0) and many other goodies (pricetag of 550$, 100 less than 780 and about half of titan).

Since this card throttles (correct term?) back its clock so that it doesn't exceed 95C (for the uninformed: AMD says it's supposed to get that hot), I'm eagerly looking forward to seeing how it performs with third-party coolers and especially with water cooling. It looks like this card has amazing potential for gaming, but unless the stats for it on mersenne.ca are accurate I guess we'll have to wait to see how it performs with mfakto.

Sadly, 7990s have shot back up in price with the release of the 290x. :sad:

I'm still learning about this stuff so please forgive any incorrect terminology I've used.

With the release of the 290X, it seems that thereis a new API, called Mantle will be implemented ([url]http://en.wikipedia.org/wiki/Mantle_(API[/url])).
This might get us some help with mfackto :
Direct GPU memory access..
Performance increase over high-level APIs such as DirectX and OpenGL.

Anantech simple worded take :[url]http://www.anandtech.com/show/7371/understanding-amds-mantle-a-lowlevel-graphics-api-for-gcn[/url]
The two point above might get us some improvement with OpenGL but it seems to be only graphic one.

I find it curious that the Toms piece did not mention the wattage at full load - for these kinds of beasts that is a nontrivial issue.

[QUOTE=ewmayer;357540]I find it curious that the Toms piece did not mention the wattage at full load - for these kinds of beasts that is a nontrivial issue.[/QUOTE]

I think Anand did mention it, and it was quite high, IIRC.

AMD/ATI hardware always looks good value, but Mantle is games focused and comes too late. I've been faithfully sticking to nVIDIA for the last 3-4 years. Mainly because the distributed computing projects I'm interested in tend to work better on nVIDIA. Many years ago, nVIDIA already had the foresight to develop CUDA, making it easier for scientists to develop software.

For example, GPUGrid, which is nVIDIA only because it requires CUDA. There doesn't seem to be any work going on to run it on AMD hardware. I've contibuted towards 4 research papers on Cancer & HIV already:
Buch et al, PNAS 2011 methods
Buch et al, J. Chem. Theory Comput. 2011 cancer
Giorgino et al, J. Chem. Theory Comput, 2012 cancer
Sadiq et al, PNAS 2012 hiv

And Folding@home, which started off in early days on ATI (still remember those 9700 cards!) hardware. It all changed 3+ years back with a new client which enabled the nVIDIA cards to crunch the work much faster. It was only recently that the work output of AMD/ATI cards caught up for Folding@home (actually still lags behind 10% or so). Unless you're buying only for a gaming machine, good hardware needs to be matched together with software development.

I think AMD has a bit too much going on with CPUs and GPUs. On the other hand, the console deal they've struck is going to feed them for a LONG time. If they spend that money wisely, their R&D is going to look to spike them back up into a competitive mindset.

I'll be interested in seeing some benchmarks for mfakto later when the drivers stabilize a bit. AMD can still make a good video card, and the R9 290X is proof of that. If it gets a bit more compute-ability, we could be looking at a real winner. I'm not sure of the nomenclature, but if the 280X or whatever is 75% of the price with 90% of the performance, some seriously dollar efficient products could be coming out.

This card is begging for liquid cooling. It's interesting that the testers believe they could have pushed past 300W if they wanted to. With a decent waterblock and good high flow rate, a person could easily dissipate 300W with a 60C temperature gradient. The R9 280X might get better results per watt, though...

This rehash to me is luke warm... mainly a rebrand of the GCN architecture and packaging more of the same stream processors into the 290X. Process is also still stuck with 28nm, so power efficiency per watt is similar. The strategy across the board was to drop the prices with a 33% discount to wow people with the value factor.

Bearing in mind, the old 7xxx series cards have come down in price over time, whereas the 33% discount is vs the original recommended retail price from last year. In practice the "progress" is a much smaller discount over current prices (in-shop after rebates).

In truth, AMD's not alone as the competition is doing the exact same thing. Geforce 7xx are also rebrands of the GK104/GK106 chips. We'll need the next generation in mid-2014 to see any real technological or efficiency progress.

Yes, +280X and 700 series are rebrands. the 290(X) is GCN 1.1, not sure how much more they have improved there. Frankly, both nVidia and AMD have been milking Kepler and GCN for quite some time....

Assuming the 7970 GHz has 460 GHz/d, I would say very safe around or over 500 GHz. [SIZE="1"](Anyone have one of them? :p)[/SIZE]

[QUOTE=kracker;357730]Yes, +280X and 700 series are rebrands. the 290(X) is GCN 1.1, not sure how much more they have improved there. Frankly, both nVidia and AMD have been milking Kepler and GCN for quite some time....

Assuming the 7970 GHz has 460 GHz/d, I would say very safe around or over 500 GHz. [SIZE="1"](Anyone have one of them? :p)[/SIZE][/QUOTE]

I am very tempted to order 2 of these, but I dont have any more nodes to run them on here, so I will check if I can get a used node first.

It may be better if you wait for the non-stock coolers to come in.