Discussion: AMD's last hope for survival lies in the Zen CPU architecture

Page 12 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.

You got that right.....competition is good, if AMD falls you could be paying big bucks for just a mid level cpu from intel
 


You got that right! I still can find a 69 cent taco at Del Taco to compete with Taco Bell's $1.39 taco and they're every bit the same o' same o'. But Del Taco's limited to western states so the rest of the country loses out. In the same sense, I don't want to see a sole chip developer at the top of the heap.
Paying an inflated prices for tech which was once available at affordable price will hurt a lot of wallets, mine included. I've always owned Intel chips but I wouldn't mind a closely comparable if not better AMD chip.
 

Intel still has models under $100 for people on a tighter budget. There are even NUCs with $30 embedded chips.

15 years ago, $300 bought you barely adequate processing power for office use. Today, $200 buys you enough processing power for most high-end gaming and a fair amount of professional applications.

People may lament the lack of periodic price cuts but processing power is still cheaper than it has ever been before.
 


And it will probably continue to be with or without aggressive competition. Hell Intel has been without aggressive competition since they released the Core i series. I would say Core 2 but AMD was still highly competitive in the server market during those times.

Software has yet to catch up to current CPU power and I doubt it will since the current goal of most software seems to be efficiency in utilization for low power systems since mobile is the current dominator of technology. You need a program to be as light as possible to allow for longer battery life.
 


I agree when has Intel ever brought out anything good when it comes to graphics. Intel's processors are great but without AMD Graphic's cards you do not have a good gaming rig! so for that reason alone I cannot see the end of AMD. Intel needs AMD if only for that reason unless Intel is willing to get int graphics which has yet to be seen.
 


Look at the broadwell integrated graphics, they are pretty damn good, way better than any of AMDs integrated graphics.

AMDs graphics cards are only good because they bought ATI, and now its branded AMD.
 

Intel does not need AMD's GPUs since Nvidia still exists and based on the Steam survey where AMD's GPU market share keeps falling to the benefit of Intel IGPs and Nvidia, most gamers who want discrete graphics prefer Nvidia GPUs anyway.
 


I personally loved ATI. Always had a decent chip. I decided not to buy a nVidia GPU from the time I got my 9700Pro on. That changed recently. My HD7970GHz was getting old and I felt like replacing it. So I looked at the Fury/FuryX and the GTX 980Ti (I always buy to keep for years the HD7970 was about 4-5 years old). I went with the 980Ti (Asus Strix). Mainly because the Fury X was the same cost and was hard as hell to find in any stores.

I was shocked that I even considered a nVidia card. But the Fury X was not compelling enough and I really didn't feel like having a CLC GPU in my system but wanted a GPU that will perform decently for the next 2-3 years.

Maybe Greenland or the one after will convince me to buy it when I am ready for my next GPU.

Problem is that for the past two generations, Radeon has not been performing well. The HD7900 was great but that lasted only till the GTX 680 and even then the only thing keeping it afloat was the crytop-currency. The R9 290 series was not so great unless you were mining. They were throttling and performance was OK. After market GPUs made it better but then nVidis launched the 900 series and it has been a uphill battle.

Fury is not bad but it is not great either. High power draw along with limited clock speeds hurts it.

 


Take a close look at Fury. They didn't scale most of the resources in the GPU other than core count, implying they mostly wanted to test HBM (especially since they didn't even wait for HBM to have higher max capacity than 4GB). Another big thing they did with Fury was adding more sensors for powertune. Then comes Nano with Powertune finalized and look at what it did. Nano is completely competitive with Maxwell in power efficiency and performance.

Fiji was just a test run with an old name to hype it up. AMD's in a very good position for next year in graphics now that they've got their crap together on it. Exactly how that will pan out, well, that's not obvious. We know they should at the very least beat Maxwell in efficiency thanks to the fixed Powertune and the new process, but we don't know by how much nor by how much Nvidia will improve with Pascal since there is little information beyond this.

I'm more worried about Zen since we have even less confirmed information to go on with it.
 


Nano is slightly better than a GTX 980 in performance but is vastly limited by Powertune and the limited power draw (limited to 175W TDP, only one 8pin connector). Then there is the price. For me Nano is over priced for the performance. The Fury X is better but I didn't want a CLC and Fury is limited to two options : Sapphire and Asus. All the reviews I have seen are showing they are very limited in overclocking and their performance is not as good unless I am gaming at 4K, which I don't plan to do until I can buy a single GPU and get maxed out 60FPS.

I am not saying Fiji is bad but it honestly is nothing amazing even with HBM. AMD didn't need to test it, no matter what HBM is the future of GPUs.

But they are having inventory issues that is for sure. The top end GPU from AMD you can buy at Frys Electronics is a 390X. They have not had a Fury of any kind in stock ever since launch. Yet they have at least 3 options for the 980Ti and a ton for the 980.

Again it was a personal decision. One I never thought I would have made as I have had a Radeon in my system for 13 years
 

There are no signs of the Fury hurting in any way, shape or form from having "only 4GB" precisely because Fury is bottlenecking elsewhere that did not get scaled adequately first. There is no point in having more than 4GB when the architecture is already bottlenecking before the whole 4GB gets used.
 


You're looking at a product instead of the technology. The importance of the Powertune implementation in Nano isn't the limits, but the granularity of it. Those limits are easy to change(well, for AMD, not so much for the buyers). The granularity of it allows much greater power efficiency by not wasting power on components that don't need it every cycle (almost all of the GPU is idle at some point, even if not every part is idling simultaneously). The limit on power draw on Nano is artificial and not the point. AMD could give Nano more headroom with another six or eight pin power connector and let the limits be more open to overclocking software, but at that point, you're completely neglecting the purpose of the Nano. For the record, much of Maxwell's efficiency lead against AMD's other cards is from Maxwell having the same type of technology.

As for HBM, yes, of course AMD needed to test it. Just because they did a test run now means they can do a better job next generation when it matters the most. Testing it now means AMD can design more robust circuitry in the GPU for the HBM memory next generation. Testing Powertune now also means they can improve further on it or leave it as it is and focus on something else, like fixing GCN's problems with scaling up to Fiji's size. Furthermore, competitors to HBM with modified GDDR5 memory are in the works, so it isn't even certain that it is the future of GPU memory. Also, Fury was AMD's first big GPU (around 600mm^2), so it's a test on that as well. If the poor availability is linked to this, then it goes on to prove that AMD did need a test run of big GPU dies if they're going to make more of them in the future in order to improve yields. However, I'd also point out that there are other sites with available Fury X models, so that might just be Fry's not trying to carry them. Did you look at Newegg? They have fifteen new, three open box Fiji models in stock right now.

Yes, the prices aren't there, but that is almost completely irrelevant when we're talking about the technology. It doesn't matter if Nano is overpriced for the performance beyond that meaning you aren't buying one. Nano proves that AMD can still compete with Nvidia in power efficiency and Fury X proves that AMD can compete with Nvidia in the top-end performance. That is what matters when we're looking at what happens next. We don't need to guess whether or not AMD is going to keep falling behind in graphics like they did in CPUs because Fiji proved that AMD can keep pace.

None of this relates to whether or not I'd buy one. I wouldn't buy any Fiji card even if I had $650 to throw away right now.
 


There is evidence of bottlenecking in a few niche gaming situations due to memory capacity, but that is not my point. I was referring to the GPU resources that were not scaled up alongside core count, presumably what you're referring to by saying the architecture is bottlenecking.
 
Hey guys. Why is everyone so disturbed that Jim Keller left or fired? Who cares maybe it was a good thing. One more thing. I'm thinking there's alot more to the picture than just AMD vs Intel. I think there's closed door meetings that go on between the two and they swap a little info. Not all but some. Don't you find it funny that back a little while Intel wanted to tap into some of that Mantle action? I think Microsoft, Intel, Nvidia and AMD have collaborated on the development of DX 12. Also don't you find it a little funny that this new Zen architecture will have Hyper threading like technology?

The failure of Zen will not mean Doom. It'll hurt and they may shrink a little. Seems the house of cards is being shuffled as we speak.
 


I do find it funny that AMD is going with a design with SMT after they criticized Intel for it. In fact they also criticized Intel for having a MCM design for their first quad core yet AMD has also utilized that.

But if Zen fails they will be in trouble. They need a good product that can sell.
 


Intel pretty much acknowledged both criticisms as true. Intel stopped using Hyperthreading for years until Nehalem came out and it was working better with better OS and software support. In the Netburst days, almost everyone criticized Hyper-Threading because it was a significant detriment more often than it was beneficial, even in threaded work.

Intel's MCM usage with Core and Core 2 was a considerable bottleneck on the FSB. That's why Intel stopped using it except for server chips and that's why AMD only uses it for server chips where the bottleneck of the interconnects is much less significant because the workloads are suited for that anyway. AMD had native quad core designs at the time, so of course they tried to capitalize on the marketing of this.

The thing I find funny was AMD thinking they could do Netburst better than Intel did. Yeah, that didn't work out too well with Bulldozer.
 

With SMT yielding a 20-30% performance increase for almost no extra die area or power cost and modern CPUs being more than fast enough for most people to be more interested in power-efficient/mobile computing than high performance, AMD did not have much other choice. Can you name even one modern mature architecture other than AMD's that does not use SMT? (No, ARM does not count as mature yet since they are still catching up on classic architectural tweaks. Once ARM's scheduler grows wide enough to frequently have under-used issue ports, SMT will become a low-hanging fruit.)
 


Oh I understand why they are using it. It is beneficial. I am just seeing the irony in it after people laughed at it. I always had a feeling that SMT would become a norm.
 

It worked quite well for me - the differences in general system responsiveness while multi-tasking on my P4 was quite obvious between HT on vs off. But just like today, most reviewers, benchmarks and people focus on single-task performance where HT was indeed less than optimal most of the time.
 


It wasn't just single threaded tasks that performed worse. Even some applications that could scale well on the Pentium Ds would run worse. Undoubtedly, that was partially XP's fault with its inferior multithread handling, but still. Part of it was probably cache and register issues with keeping more data from two threads on at once since Intel added more for Hyper-Threading in Nehalem.
 


Pentium Ds were dual cores not HT. HT lasted for the Pentium 4 then they released Pentium D which had two actual cores (in a MCM configuration).

As I said I am just seeing the irony in the fact that anything AMD has said was a bad idea they have utilized. They trashed MCM, even on the server side. Yet when it came time to throw more cores they used it. And of course there is the excuse of FSB vs IMC but point is, AMD trashed SMT and MCM designs and are now using them because they are beneficial in getting performance where you need it.
 
?

Your analogy is completely backwards.

Once the Big 3 (McDonald's, Wendy's, Burger King) had a $1 burger on the menu, a number of things happened.

First, it turned out to be impossible to get away from. All 3 of them tried various means to get the $1 burger off their menus and simply could not do it, because customers just went elsewhere to get the $1 burger and sales plummeted.

Second, quite a few of the more expensive burgers stopped selling and were discontinued. Even after, the Big 3 tried new versions of more expensive burgers and they have all failed.

Third, once the other fast food contenders, like Taco Bell, realized what was happening, they all started offering $1 menu items too, and the competition became even more pointed.

Bottom line is that once you go down in price for a specific product or type of product, you can never, ever, go back up. Marketing 101.

The only way to get higher prices is with a game changer product, which is what AMD is hoping Zen is. I hope Zen delivers because it will directly affect Intel pricing, which is good for all consumers.

 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom