AMD Radeon R9 300 Series MegaThread: FAQ and Resources

Page 39 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
the point is not about how many unit being sold. the list were ranked by it's performance. why that top 500 machine usually use nvidia tesla and intel phi and not AMD firepro? AMD launch S9150 last year and since then the list has been updated twice. why AMD still not dominate the list with their superior S9150? nvidia already have contract for super computer that aiming for world no.1 performance in 2017 time frame. and there were another super computer that aiming for similar position. and intel won the contract with their phi. why not AMD?
 
When discussing HPC, it's not about the overall performance. That is a key aspect, no questioning about it, but most scientists don't go for the biggest baddest hungriest machine they can find to crack numbers. They go with what they need. I know it's murky to say it like that, but the point is that being a "top 500" does not mean gross numbers and I agree with Vog in that regard. It's like saying Toyota doesn't sell much because they are no longer in F1, WRC or other big races (sans Le Mans 😛). If a FirePro will now have CUDA and it's cheaper for what I *need*, then why not? I believe that is the angle AMD finally acknowledged defeat on. OCL did not catch up to CUDA in most scientific tasks, research and big name number crunchers support out there. Kind of the same as OGL and DX back in the end of the 90s.

Well, the positive spin is that will mean more sales from AMD, and nVidia will keep royalties from it, so I guess they all win.

Cheers!
 
i don't think AMD will use CUDA directly. but from what i can understand they will try to provide tools so it can be much easier to translate CUDA into OpenCL or using wrapper so CUDA can run on their hardware (like those Glide wrapper). and no nvidia are not getting paid in this process. anandtech mention that AMD legal team has made sure they will not get into trouble with nvidia. they said google is doing something similar (about the wrapper thingy) but nvidia did nothing about it. though anandtech mention that things could be different since AMD compete directly with nvidia selling accelerator hardware.

and while it is true fireGL can be cheaper nvidia has actually has no problem competing with price if there is a need for it. nvidia after all can afford to do it more than AMD did. i read somewhere that nvidia once sell their tesla far cheaper that their official price just so they win the bid with amazon. but in the end nvidia win against other solution because they provide guaranteed 1:1 replacement (no question ask) at much cheaper price.
 
we all know firepro gpu's are simply overpriced R9's and R7's. if amd wanted to destroy tesla they will simply lower price to their radeon brothers price and watch as nobody sensible buys the expensive options that are equally fast just because they are slightly better at cuda than amd's transcoding to open CL and then computing method. price to perf will be gone. that's what they look at when designing a data center. electricy cost yearly, initial cost to build, performance for computing what they compute in flops. NVidia and intel are winning the grand scheme of this at the moment. they are calculating that at the end of the life of the computer (say 5 years) the amd system will have a worse cost to perf ratio.
 
Fire pro cards offer full dp performance and open gl optimised drivers compared to Radeon, and they are considerably better value for money than quadro most of the time. Nvidia quadro and tesler cards are based on GeForce silicon as well. There are differences though, which is why they carry a premium.

I work with 3d cad allot, and trust me the performance you get out of an entry workstation card is much better than considerably more expensive gaming cards in that type of software.

The underlying hardware might be the same but the software makes a real difference...
 


that is something that not even AMD dare to do. because if it really that easy AMD had done it long ago. but why they did not did it yet? and surprisingly AMD actually is copying what nvidia did when it comes to gaming card vs workstation card: purposely gimping the gaming card so people will buy the professional line up. just check the test that done by anandtech. when it comes to double precision 280X handily beat 290X. but S9510 which is based on hawaii is really a beast. when it comes to DP not even nvidia best chip at DP can compete with it. well in theory anyway.

btw AMD has always been selling their professional card at much cheaper price than nvidia. but most company still choose nvidia more expensive solution (hence nvidia pretty much dominate the market since long ago). when it comes to professional world price/performance will come second after the result. the most important metric is you can get the job done.
 


all cards get the job done some just take longer than others...

Yes drivers make all the difference (both amd and NVidia make you pay $$$ more for some DP drivers 🙁 )

In my job I cant disagree with you more. every decision we make is price to performance based. we do a justification of bids weighing all bids for a project for benefits and costs to justify if a more expensive solution is worth it. 90% of the time its not. All businesses are only in business to make money. you claim they are simply buying one brand based off preference and paying more for it. I can tell you at least in my field this is not true. we will work with cheaper vendors if they truly have a better deal any day of the week. No business would buy a $500,000 server without first doing a full month long investigation of bids and purchase decision process. its too much money to simply go with what somebody recommended. I guarantee that the CEO's would require a presentation on how one card performs faster than another and uses less power over the life of the part and will have a better work done to power used delta, all the while being fast enough to do the work they need. They would also need to collect bids from multiple company's and being as there are only three available for server GPU's this becomes a pretty easy task, and execs would want to see bids from all three. procurement would be fired if they didn't get all this information around before making a purchasing decision.

 


True, although nvidia isn't actually outright faster. It's a bit of a case by case basis, but usually the same price Fire Pro *is faster* than the quadro across a range of software.

I think in this regard the problem is AMD and their marketing (or lack thereof). Most large companies use a computer vendor due to wanting technical support and so on. Most (if not all) hardware vendors I've ever spoke to *simply don't offer* any AMD hardware at all unless by special request, and usually their staff no nothing about the tech simply stating 'nvidia is a better brand' or something like that.

The thing is, AMD do offer better results in many big name CAD packages but you'll still only get offered nVidia. I mean maybe nVidia's top end quadro is faster than the fastest Fire Pro overall, but as with desktop cards when you look further down the stack where *most people* are buying it is seldom black and white like that.
 
Why would 380X will get HBM? I thought AMD was very clear about only Fiji will have HBM for this gen.

@MM

I still think they are the same. Just being renamed into something else just like what happen to Hawaii -> Grenada.

 
Uh 380x is full Tonga and no it was never going to have hbm, it's a mid range card that doesn't have enough shaders to need it...

Edit: as for rebrands, personally I don't usually mind as the latter revisions usually offer improvements.

Where I do question it is like with the Radeon 370, which is based on a slower, cut down version of a gpu compared to previous gen 270. That's just misleading to consumers imo.

No issue with Hawaii rebrands as they offer higher speeds and double the memory so actually constitute an upgrade from previous card.
 
To me as long as they get the job done i don't really mind either. But the whole noise about this rebrands is more about ethical or something like that. The problem is not all people have deep knowledge about pc hardware or even if they have reasonable amount or knowledge they did not follow the lateast info closely. So when buying from newer series people expecting performance improvement. And sometimes they expect noticeable jump in performance. For example the guy already have 270. Then he buy 370 expecting more performance only to be dissapointed when he know the exact performance. I think this is what most people don't like with rebrands. Stuff like this might happen.

Then throw in the fanboyism stuff to the mix :lol:

So you get the general consensus like i mention above haha.
 
I just bought an MSI 390X, downclocked version of their flagship one. $350 on newegg with mail in rebate in tandem with the discount. Pretty good value for a card that trades blows with the GTX 980 in my opinion. I was already impressed with the 290Xs when I had those with my 780Tis (primary and secondary systems). I just hope the driver situation has improved since I had the 290Xs; though to be fair the last few Nvidia drivers have been pretty awful for me, especially the ones for the GTX 880M. I do wish AMD had priced the Nano more kindly or I would have gone with that, but the performance difference with a 390X was so minuscule it was hardly worth $200 more https://www.techpowerup.com/reviews/AMD/R9_Nano/14.html. Had it been $100 cheaper I would have seriously considered it. On the rebrands I'm fine with it as long as it provides some decent improvement.
 


HBM is expensive. but nano did not priced at 650 because of that reason alone. AMD actually asking premium on nano size and power efficiency just like nvidia charging a premium price on 750Ti because of it's efficiency.
 


True, the standard Fury is much better value than either Nano or Fury X, although as with most top end cards, it's still less value for money than the tier below.

I'm curious actually as what is the best perf / $ card in AMD's lineup? I'd guess the R9 380 is probably the sweet spot, whilst I think with nVidia it's probably the GTX 950?
 
Just a few comments! After being happy with AMD 6970 for 3-4 years now, it has run my games well in FullHD. Moving up to 2560' pixels, and the release of Fallout4 made me buy a new graphics card. Actually the FuryX was interested, until I realised the high price would remain for a while ahead. 390 was my second favourite, but a little expensive too, when I tested an ASUS R9 390strix I found out it was too big for my case. Was afraid the 380 might not be enough, it was still somewhat affordable, and I found a smaller sized card, and bought it. After testing FALLOUT4 I am very satisfied mow with the 380. It runs Fallout4 FULLHD without problems, and runs very well also in 2560* but if u have a 4k screen, u can´t run FA4 smoothly with full settings, then the Fps WILL DROP, but if you can play with lower settings it may run OK.
So for fullHD 380 should be enough for most players, 390 for hardcorers.
Only if u use a higher resolution screen 2560 380 is ok, 390 is perfect
For 4k I would suggest 390 or Fury.
But again it depends on games u play. WOT, WOW and stuff like that played fine with my OLD 6970. It even run FA4 even if it was about 20% slower than minimum recomendations, but yeah graphics were on minimum.
 
380x is a win because of its price. $240.

Its 10 fps faster than 960 in all games

only $40 more.

970 is $300 and gets another 10 fps over the 380x.

price to performance win goes to the 380x.

Full tonga gpu with 4gb of high speed ram and 2048 shaders? its a fully leaned out max speed 7970 for excellent price to perf ratio.

Technically a "rebrand" but rebranded from the R9 295m which was only available in the apple imac 5K, which nobody bought...

and it is one of the "rebrands" to use third gen (1.2) gcn arch so its a "rebrand" that is for the first time actually available as a desktop part.
 


A rebadge is still a rebadge is still an old GPU being marketed as a new one.
 


I pointed that out, but also pointed out you cant buy last gen's full tonga gpu for your desktop because it never came to market. Its the only rebrand that wasn't branded in the first place. I also pointed out that it while being a gpu from last gen, used current gen Uarch (gcn 1.2).

Soooo yes its an old gpu, but nobody ever got the old one before. essentially making it a new product.

this is like the 970 ram argument. Yes it has 4 gb of ram but only 3.5 gb gives you good performance soooo technically you can say it has 4 gb of ram, but nobody can use it...

what im saying? to call the 380x a rebrand is like calling the 970 a 4 gb card.
 


I don't care what you call the 970, it still doesn't change the fact that AMD have just been tarting up their old GPU's and marketing them as new ones for the last couple of generations.
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom