[TechPowerUp Article] Why The Litecoin Craze Hurts More Than Helps Brand AMD Radeon

http://www.techpowerup.com/196320/why-the-litecoin-craze-hurts-more-than-helps-brand-amd-radeon.html

I figure it's alright to link to another Tek user's blog entry about this subject, here: https://teksyndicate.com/users/supermilleno/blog/2013/12/25/how-war-between-amd-and-nividia-going-my-opinions

Have fun with this one, guys. It's a great read. And it may also help shed some light on the subject of where AMD might head towards next. Maybe AMD Radeon might be one direction they're headed for gamers, but if CryptoCurrency is such a high-demand, high-paying market, AMD may find themselves selling highly-parallellized compute cards meant specifically for hash compute, much like how Intel Xeon Phi and nVidia Kepler cards are designed for supercomputer and compute tasks.

If AMD can design a fairly low-cost, high-performance card for people to buy, we may end up seeing people buying a lot of these cards, making the supply of AMD Radeon cards for gaming cheaper, while also having higher price/performance than AMD Radeon cards do. There's a lot of advantages to this approach, and I think AMD can make an entirely new segment. We've got FirePro and Radeon. Maybe we could have AMD Dash, meant for speedy hash calculations using a discrete add-on PCIe card (read: dual-slot with fans). Much like how nVidia started Kepler for supercomputers, and Quadro is for video/photo editing, perhaps AMD could segment their cards in a similar way?

My idea:

AMD Radeon: Gamers, viewing videos (very light load), moderate parallelized compute. Best price/performance for gaming. Has TrueAudio and some other features, but included audio isn't "studio-grade" and it doesn't output marketing-grade "audiophile" throwaway terms using the DisplayPort or HDMI ports (ahem: 96 Khz, 192Khz, 24-bit, etc).

AMD FirePro: Meant for content creators. Supports output for "audiophile" marketing-grade throwaway numbers. Is meant for video H.264 and H.265 rendering, audio rendering, and supports TrueAudio for movie makers, game developers, and more. It also supports 12-bit color space for those who want those features.

AMD Dash: Meant for CryptoCurrency calculations. Not meant for anything else. Has high price/performance ratio here, and can output some video, but that isn't it's focus. Most components removed. (Side idea: could also be used for supercompute applications, in order to further segment each card from it's adjacent card. This may help increase performance, and help AMD analyze the sales data to determine which areas are most profitable for them to continue to pour engineering and research money into.)

And we've also got their CPU/APU area. Well, that's a whole other subject. But AMD could just launch a true 8-core APU, leaving the StreamProcessors to be used for compute operations like Physics in order to get better performance in games. That might help alleviate PhysX performance issues, assuming game developers and others take advantage of such and allow it to be highly parallelized by an AMD APU. AMD might also instead go for a more Intel Xeon Phi approach, and make several lower-clocked x86 cores which are more familiar for developers. However, due to heat and single-threaded performance issues, and the lack of enough bandwidth to make effective use of this, I find it highly unlikely.

AMD might be able to fix this with 20nm fabrication process and a new architecture, since they've now been hearing this issue being yelled at them through the megaphone known as Facebook and the entire internet. The same also applies to their FX-line of CPUs. Maybe we'll see them fix this. Maybe not. Let's hope they listen. And maybe they'll even give us some extra single-threaded performance, and maybe even an epicly full-featured chipset to go along with it - but I wouldn't hold my breath for that.

Very interesting.

an 8 core apu with 20nm architechure doesn't sound so realistic, since it has to be priced at a reasonable level, at least compareable to the i7 4770k, and for people to deem it as a "true" 8 core they would need A LOT of L1, L2, and L3 cache for each cpu. Each cpu has to actually be very strong, or else it will become another fx 81xx failure, and result in horrible single core proformance. All this power from the cpu, but what makes an apu an apu? the igpu. If they want something with this kind of power, it would be stupid to pair it with an igpu that's weaker than the ones intel has. This means they need something the power of a r7 250 or so. then the final issue: power. This is gonna need a lot of power, and if you just look at the a10 6800k, it already uses a lot for a 4 core with its igpu. how much do you think you might need for this one? and that also means bigger heatsinks, which is even more costly. too many problems for a "true" 8 core apu that needs technicians to design as well. why do you think AMD turned the console chips into "jaguar" apus? these come from the server chips amd makes, and these had to be custom made. but whats different is that the cpus on the apus are very weak, so it uses less power and space. Then look at the fx 8xx0. That series was recognized only as a 4 core with 8 threads, since it shared cache between 2 cores, even though the cores were physical.

What amd really needs is something that can target Nvidia's gpus in different areas. I'm still thinking that they need the small form factor pcs. Many htpc cases can only support a low profile card. Nvidia is slowly doing that, by shrinking their high end cards and giving them different coolers. look at the titan. it has a blower style cooler, which might not be the coolest for the pcb, but it does give ssf pc cases better airflow. And look at other high end cards that they make, like the 770, or 780, or even the 780 ti. (leaving the 760 out because we already have WAYYY too many oems making super short and tiny ones, its stupid) The power that they use is usually lower than amd cards, and they also have smaller cards as well. amd's new flagship, the r9 290(x), sure is powerful, strong, solves the crossfire connector problem with digital connecters, but it adds on the cons of the previous cards: Noise, Heat, and Size. The card is much bigger than the previous flagship card, the 7970 ghz. It's longer by almost an inch, its fatter, it uses pretty much the same cooler, and it really just is huge. The noise is about the same (when non reference cards are more popular), which was something many people didnt like about the 7970 ghz, and the heat is about the same, maybe a little lower this time though. Now this might not affect some people who have huge cases (*cough haf stacker 922) and plenty of other things in their cases that make more noise and heat. ( insane water cooling unit that makes pulping noises every few minutes) This just doesn't seem to be the right direction. Sure its faster than a 780 and goes against the 780 ti at higher resoultions, but it looks akward compared to other cards. it takes in almost 400 watts on the card! imagine crossfiring on this in a matx case. I think amd has to wake up.

"They have technicians there, making noise. No one, is a gamer. They make bad cards because they don't listen."

really wish i picked up a few more R9's to re-sell at 100 / card profit.

I got a 7850 back then for 130 bucks... might sell it for 150 :p

Very good. However, look at the difference in heat output between Sandy Bridge and Ivy Bridge. Or the graphics performance and decreased energy consumption between 40nm GPUs (like GTX 480, GTX 460) and 28nm GPUs (like GTX 680, GTX 660 Ti), The graphics horsepower increased per watt a lot. And the same goes for CPU performance.

Sure, AMD may have be a bit behind on the whole CPU area, but remember, they have the right idea. Although they can't improve their single-threaded performance, they are working on other areas.

Their graphics cards right now suffer from a lot of issues. But this will be fixed over time, and they can really, *really* get the money to hire more engineers, gamers, technicians, designers and developers/programmers to help get their drivers and hardware to be on par with Intel's stuff. They just need the money first, which is where my idea for the "AMD Dash" discrete boards comes in. It allows those who want to buy such boards to have them for very low cost, and it gives AMD a much needed cash injection.

Why put more memory on a board that doesn't need it? One less component to spend money on unnecessarily. Why make a GPU that has loads of ROPs, TMUs and other things which might be irrelevant for hash calculations? If AMD can make a GPU tailored for hash calculations, they can make it less noisy, output less heat, consume less energy, and deliver it at a lower cost. This could be where AMD's "custom CPU design" idea could be best put into practice!

Also, consider this; AMD knows developers need to start coding their games and programs to utilize more cores. AMD realized developers are lazy and don't want to do that, so AMD went ahead and made an API do take care of that for them (cough, Mantle, cough). If AMD can design a programming language, DLLs, and a compiler that also does something similar for CPU design (maybe incorporating Heterogeneous System Architecture optimizations) automatically (without the need of the developer to think about it), it would make AMD systems much more powerful. It could give AMD the boost from their extra cores, and by using instruction sets that are optimized for AMD architectures (but not so much for Intel's instruction sets), this could allow AMD to gain an extra bit of foothold in the performance market.

Better instruction sets are one of the things in a CPU that can give it extra single-threaded performance. If AMD uses programming languages and software to give their hardware an advantage, it might be a big deal for their sales. And if it makes Intel's performance go up, but in terms of raw percentage AMD's performance goes up more, it means AMD will be more competitive versus Intel. That's where they can take a big advantage. And I hope they do. =)