AMD Radeon R9 300 Series MegaThread: FAQ and Resources

Page 19 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar


That's a very interesting analysis. What I can conclude from it, is that current game engines are really not optimized for VRAM use at higher resolutions. Applying AA should not double or nearly triple your VRAM usage, lol. Case in point, the Frostbite engine, which is one of the best out there (I like it a tad better than UE), it really well optimized and it shows.

Sadly, GTA V won't be optimized anytime soon, nor in the future, haha. Rockstar has been very lazy about it and we can't expect miracles. In any case, at 4K, would you use AA? I know even at 1080p, I don't use AA.

Cheers!
 

i never use AA as i favor more fps. but i suspect nvidia and fanbois will lead with vram capacity limitation as a "major issue" with fury cards. expect this "issue" spammed and nailed into people's brains in the coming days.
historically, new memory has always launched with less capacity and higher price (but much better memory performance) ever since SDRAM vs DDR days. when nvidia gets HBM the propaganda will change to how "nvidia's" HBM2 is much better than "amd's" HBM1.
 


Wasn't AMD already getting stuff done with HBM2? My point is, for all we want Fury to succeed, I'd say it's a stopgap in terms of memory tech. The next GCN card should sport HBM2, right?

I mean, that's why they had to eat the cost of the rebrands isntead of getting a full old-gpu + new-memory swap. Although making the latter would have been hell expensive I guess. Better wait for a refined Fury-based GCN card and start from there. Which, I think, Fury nano will be that "watered down" card we're waiting for.

Let me fix my thoughts, haha.

1.- AMD working on HBM2 at the moment, so when nVidia gets Pascal out, AMD will also have HBM2 equipped cards at least in the works; although I'd expect them to have something either at the same time or a tad earlier.
2.- Fury nano is a premise for a watered down version of Fury's GCN card? That also implies HBM1 for the masses? (notice it's a question).
3.- Fury might be a stop gap until HBM2, new manufacturing process and refined GCN comes all in place.

Thoughts?

Cheers!
 
I don't see the point fanbois want to talk about HBM1 vs HBM2 because around the same time AMD should also have access to HBM2 unless AMD intend to keep Fury and 300 series for the entire 2016 and will only come up with new gpu in 2017. Nvidia did not use HBM1 most likely because of time and cost. Also that 4GB limitation. Imagine Quadro or tesla card that only limited to 4GB. Hence i was interesting how AMD will go with their pro line up. Will they just go with that 4GB or wait until HBM2 before making Firepro card using Fiji. On gaming space nvidia most likely will be able to hold out a bit even if fury end up faster than titan x. Expect nvidia will ramp up their software team (ahem, gameworks) so the advantage will tip to their side. Right now the limiting factor most likely how fast TSMC can be ready with their 16nm finfet. There are already rumor talking GP100 and GP104 being tape out. If the rumor are true then we probably going to see something moving through zauba listing as early 6 months before actual launch. If amd can push nvidia to be desperate enough maybe we will going to see 'broken' GP100 being pushed into actual product.
 

yes:
these cards are stopgaps. been droning about that for months. 😛
amd has HBM2 equipped gpus in the works, prolly finished internal tapeouts.
both amd and nvidia will have access to HBM2 next year as well as same finfet processes.
r9 nano is watered down fiji. the card sample shown at at expo is looked much cheaper than the fury ones in terms of build and construction (but not necessarily of lesser quality). i guess the CLC on fury cards alone would cost from 60-120+ bucks. less shaders would mean much higher yields and on a die this big - easier to cool.

now a conspiracy theory based on wild speculation and borderline misinformation: let's assume nvidia has pissed off everyone with it's patent trolling shenanigans. let's assume apple dropped nvidia gfx cards for this. let's assume amd would never do something like this (and they aren't in any position to do so). let's assume others (qualcomm et al) want to get even.
when the next processes ramp up production, both apple and qualcomm take up priority fab space from tsmc, glofo and samsung. i don't know yet if mediatek is among nvidia's targets. basically, all big, relevant arm players fab-block nvidia from early access. let's assume amd is smart enough to take advantage of this and gets next gpus fabbed early and starts selling them. only then amd will have a very good shot at getting better of nvidia. i don't think pascal gpus (GP100) will lose in terms of performance should all other factors like yields, tapeouts, HBM2, foundries etc. become equal.
 
Apple and others could do that but i'm not sure if TSMC would allow that. If anything TSMC would be afraid to lose nvidia. I think there are talk last year about apple want to pay TSMC to be their exclusive fab but TSMC rejecting the idea even as a rumor. And back in 2012 TSMC take Qualcomm and AMD capacity to satisfy nvidia need. So i doubt apple and qualcomm can attack nvidia from fab angle since TSMC would be afraid to lose nvidia.
 
The whole thing about comparing anything to the Titan X, as if that's Nvidia's fastest card, is a bit of a misdirect. Since there are no custom design Titan X's, any custom GTX 980 Ti is going to easily outpace a Titan X out of the box and overclocked.

Since it appears there's no room for a custom Fury X, a custom GTX 980 Ti is still going to be the mark to beat. With the Fury X's water cooler being better than any Windforce, DirectCU, etc. cooler, what you see in those reference reviews is probably going to be all there is. We'll see, but maybe there's some room for factory overclocked models.

perfrel_2560.gif

http://www.techpowerup.com/reviews/Gigabyte/GTX_980_Ti_G1_Gaming/30.html
 


Those statements quoted abot suggest both that high-end makes more money and that it doesn't. Not sure what to make of that.

But back to Lisa Su's interview, I'm pretty sure she was talking about GPUs as well, not just CPUs. When someone says "$300 AMD CPUs," how many CPUs would they even be talking about anyway? Tom's interview of So mentions in the writeup that even Nvidia's CEO, Jen-Hsun Huang, said the same thing; namely that high- and low-end chips yield high- and low-profits. Nvidia only does GPUs. So I'm pretty sure that statement was also about GPUs. Here's the Jen-Hsun Huang quote, which came up in a discussion about how Nvidia's high profits stem from Tesla, Quadro, and high-end GTX only.



http://www.bit-tech.net/news/hardware/2013/05/10/nvidia-profits/1

Aside from what they say, that makes total sense. Manufacturing a chip will cost roughly the same amount of money, no matter how fast it is. You just put silicon through the process and slap things on a PCB. To be fair, I know it's complicated, but that's across the board and that keeps production costs predictable. With production costs fairly set, the primary variable becomes the selling price.

Assume it's $30 to manufacture a graphics card (not meant to be an accurate assumption), and let' X be the amount you sell the card for. The profit will roughly be X - 30. So when X is smaller (i.e. you sell a card for less), you have less profit. Conversely, when X is larger (i.e. you sell a card for more), you have more profit. . . . Oversimplified, but it makes the relevant point. I'm not even sure what the point of contention is on the profit issue.

As for volume, you'll just have to look at sales figures. Market capitalization isn't a good way to look at things for that. Nvidia's bigger. Of course a bigger company will sell more than a smaller company. What really matters is how sales impacts stock price. At this point, all AMD and it's shareholders care about is doing something to bring up the stock price. Have you ever read through SEC fillings like a 10-K or a 10-Q? It's really the best way to get a grasp of these things. Here's the company search tool on the SEC's website, which has full information for every public company. Search by typing out the company's full name - e.g. Advanced Micro Devices, not AMD: https://www.sec.gov/edgar/searchedgar/companysearch.html

This current release seems to have a shot at giving their stock the "shot in the arm" it needs. They will sell to AMD fans - given. But the tech may prove to be compelling enough to win over some enthusiasts this time around.
 


No, high end does have a higher "profit margin". But being a profitable market does not revolve around just having "high profitable" SKUs on sales. There's also volume.

A simple example would be:

Fury X: $A to manufacture, $B selling price -> $(B-A) profit and the margin I think was defined as (B-A) / B?
7870: $A' to manufacture, $B' selling price -> $(B' - A') profit, same idea for margin.

In any case, I'm saying that (B - A) > (B' - A'), but there are C and C' numbers such that (B' - A') * C' > (B - A) * C

What I'm trying to get at is that, you sell a bunch more in the 200-ish price bracket as opposed to the 600-ish bracket, so in terms of actual revenue, it makes more sense to have better products in the 200-ish price range and not in the 600-ish, except for bragging or marketing, since they're not the big revenue drivers.



I'd say we're not discussing different things. You just have different assumptions than me. In particular, in my mind the 200-ish bracket is the biggest selling part and not higher brackets. Lisa Su, Jen-Hsun Huang and Brian Krzanich (Intel's CEO) want to move customers to a higher bracket, but moving customers is hard to do. They depend on a LOT of factors to do it. In particular, something that it's called "replacement goods".

Cheers!
 


I mean, yes, sales volume is a multiplier. It just brings up a factual question for which you'll have to look at the SEC filings to find the answer: "What are the relevant sales volumes?"

There will be a break even point based on sales volume for overall profitability in each segment. Assuming there's $20 profit in low-end GPUs and $100 profit in high-end GPUs, you'd have to sell five-times as many low-end products just to break even with (i.e. match) high-end profit. It all comes down to answering that quesiton.

But even though AMD is smaller than its competitors, it certainly has the resources to hire super smart people to run the stats on this issue. I'm sure they have. It all falls under price modeling, and they've set prices for their GPUs. Without looking up sales volume figures, I'd be highly surprised to find that low-end actually drives overall profits, despite what CEOs from the main chip makers say. I mean that would begin to get at the notion of oligopic fraud throughout the market, and there's not enough to suggest that's what's going on.
 


Well... On that last point, there was already a class action lawsuit against ATI and nVidia at some point in time.

At least I can now see we've reached a consensus of what we don't really know and can put the argument to rest 😛

In any case, we might not know the "gross sales" numbers to identify each optimal target market for a product, but I do know that moving products to different (higher) tier price brackets is not an easy thing to do on any market for any product.

Like I asked a couple of pages back, how many of the readers in this thread have something more expensive than $250? How many of us break the 1200 USD overall PC price? How many of our non-casual friends do? How many of the people asking questions around the forums want to spend over 250 on any given component?

Those are fair questions to answer as well.

Cheers!
 


Yes, all fair and interesting questions. It seems like there's two perspective in our discussion. First is the AMD consumer, and second is the AMD investor. Those questions are good to know for consumers, and it may be premature to buy right away before seeing various test results. But for investors, the main question is what the company reports in the SEC filings, and now seems like a good time to buy stock (not GPUs). Maybe the perspective differences is what drove this. Who know? In any case, I'm excited to see what happens from both perspectives.
 


The problem with these calculations is this:

They are trying to use the past in order to calculate the future. What I mean is that yes, for games that have already been released, these calculations hold. But we are talking about a NEW generation of cards that allows for the NEW games to do MORE. And yes, that means more VRAM usage.

People who are buying the new cards aren't doing so in order to play Far Cry 3 at max settings. Or CS:GO. They are buying then so that they can play whatever is coming out around christmas 2015 at max settings. So they need a decent cushions on their specs, to make up for the improvements, which unerringly cost more resources, in the newer generation.
 


If you want to drive the discussion into "stock value", I'd say that's a murky discussion to had. Since stock price is basically 50% speculation, 30% facts and 20% lies, I'd really not like to go there, haha.



Not to be an ass, but they can't calculate things "for the future" when the "future" is not even here. You have to extrapolate information out of the investigation they conducted. For instance, they used Battlefield in one; we all know EA is using Frostbite in other games and will be used in newer games. The tendency for that game engine is to be well optimized and well rounded. Games that come out of that game engine have a smaller chance of crippling video card that *today* have 4GB down the road. I also made the point for GTA V. Whatever Rockstar is using scales weird in my opinion. In particular, it looks like they have not optimized the engine to make an efficient use of memory. Case in point, take a look at the Witcher test. Or even better, any MMO RPG that has *massive* environments with great graphics. That supports my take on blaming the game engines.

Now, no one knows what will happen down the road. If costs for memory go up too much and AMD with nVidia can't surpass 8GB (for example) for a good set of years, game engines will necessarily need to be optimized and so on. Like I said, until "the future" is not here, your guess is as good as mine (like they say).

Cheers!
 
The AMD investor vs the AMD consumer... funny, but what makes one happy might upset the other. The investor wants higher prices, while the consumer wants cheaper products.

Enter Nvidia who stepped in a few weeks ago and basically told AMD that their new, expensive to produce card could only sell for $650. With profit margins thus trimmed and market share at 1/3rd and declining, I wonder how investors are feeling.

The fans sure are pleased though. Again, it's funny but they should actually be wishing that these cards were more expensive. Because that would mean that the product is worth it, and the company has a strong position in the market relative to the competition..
 


Maybe, but the stock just hit a one-month high. :)
 


add AIB to the list. i heard rumor some pf AMD AIB did not happy with the low price and the barrage of rebrands. and one of them considering to drop AMD.
 

This is an fascinating interactive chart on AMDs stock prices. AMD stock gained 0.12 with the big E3 announcements yesterday, up to 2.47 from the day before.
http://

You can spot the longer term trends as well. The 290X was released on October 23, 2013. On that day, AMDs stock actually dropped 0.04 from the day before, but began a steady climb after that, from 3.14 up to the 4.47 range. (Remember that the XBox One and Playstation 4 came out around the same time.) So, it stayed bouncing around the 4.00+ range for a while, until Maxwell GTX 970/980 was released. That was not good for AMDs stock value as it took a drastic loss, from 4.2 to 2.6, where it has remained ever since, well below 3.0. It'll be interesting to check back after the Fury X reviews come out.

And a chart comparing AMD vs. Nvidia for reference.
http://
 
Quick question i'm not trying to bait a graphics war, but. I keep hearing about this "tegra" chip that nvidias gonna have to release is it a specific type of gpu they release to have competition for amd's new stuff? Like the 980 has the ti? That said i will be buying a fury x is the relative future if it is as good as people say it is when benchmarks eventually drop.
 

nvidia's tegra lineup is made of ARM CPU based socs (system-on-chip). that means tegra chips have CPU cores, integrated GPU, northbridge and southbridge, other hardware blocks all on the same die. those are for smartphone, tablet, media streamers and set top boxes etc. the latest two tegra socs have integrated a small, watered down block from nvidia's main gpu microarchitecture (kepler and maxwell) as the iGPU. tegra is for a whole different class of devices which serve very different purpose from desktop gfx cards in PCs. tegras won't interfere with comparison with amd's gfx cards. fury x competes with other desktop high performance gfx cards from nvidia such as gtx 980/980TI/titan X and such. tegra socs compete with other ARM socs like Qualcomm's Snapdragon series, Mediatek's socs, HTC Kirin socs.
 


tegra is SoC. you can say that is nvidia version of APU? although it is for mobile market and competes directly with qualcomm, mediatek and the likes. it has nothing to do with discrete gpu at all. i think AMD also tries to go after mobile but so far i haven't seen much success. Project Discovery never goes beyond prototype. but probably wise if AMD does not involve themselves into this market. even nvidia and intel themselves have to struggle in this segment. in recent earnings nvidia already mention about selling their soft modem division that they acquired back in 2011. tegra mostly will focus on automotive (been there since the very beginning) and small form compute device. and probably tablet if there is anyone interested to use their SoC. intel for their part have burn a lot of cash just to have a foothold in mobile. i think this move ultimately make it hard for AMD to enter mobile.
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom