HumanSmokeAlmost ALL that success was purchased.
AMD's K6? The original K6 project went tits up. AMD purchased NexGen and used their Nx686 as the basis for the K6
AMD Athlon (K7)? AMD licenced DEC's system bus and hired the DEC Alpha team (Dirk Meyer et al) to design the K7, along with DEC's Jim Keller who was instrumental in putting together HyperTransport
Athlon64 (K8) Also based on DEC's work (along with AMD's own design chief, Fred Weber of course) with David Cutler and Robert Short (both then working for Microsoft) instrumental in getting AMD64 up and running.

The difference back then was that there were a lot more processor IHV's in play - many failing ( Cyrix, NexGen, C&T, DEC), and IP and talent could be readily purchased if your own in-house projects met with failure. That isn't the case now with x86.

Yeah but AMD still got the initiative and strategy correctly done for those cases, purchased or not, they got the thing done, doing the first 32bit + 64bit processor. Back on these opteron 64 days, intel execs said AMD were crazy. But this changed Intel plans for their Itanium processor.

Their processors today are not the disaster people believe. Bad press is all they've got. They had the TLB bug on the Phenom and questionable design in their bulldozer architecture. But they hit well on APUs and GPUs (purchased or not, AMD has been doing a lot in the GPU area)

ThefumigatorYeah but AMD still got the initiative and strategy correctly done for those cases, purchased or not, they got the thing done, doing the first 32bit + 64bit processor.

Yes they did, but the point is that the opportunities to make those strides thanks to picking up IP and chip architects from other companies don't really exist in both numbers and at affordable prices in this era. Remember they basically purchased an IHV with a ready-to-go next generation architecture (NexGen) for $847m, and DEC's IP which laid the foundation for all AMD's subsequent processors for next to nothing thanks to the 1998 FTC Consent agreement.

ThefumigatorBack on these opteron 64 days, intel execs said AMD were crazy. But this changed Intel plans for their Itanium processor.

That is incorrect. Intel's x64 program actually was two distinct developments. The first was a 32/64-bit program (the original P7, later renamed Merced - Here's a magazine article outlining the original concept in 1995, so pre-dates AMD's K7). This original program was shelved because Intel realized that they were being marginalized by Microsoft as the industry prime mover. Concentrating on Itanium - a purely x64 architecture, was a gambit to both wrest control back (since MS's software was 16-bit/32-bit based), and to head off competition from x64-based RISC architectures, which Intel perceived as the long term threat to x86.
Where AMD won wasn't on landmark originality (both AMD and Intel were incorporating RISC architectural features into CISC processors), it was a willingness to partner with Microsoft and open source SUSE group to get the compiler and ISA (AMD64) up and running- and accepted as an industry standard, where Intel wanted full control of IA64 and to bend the industry to their way of thinking. Having MS and SUSE on board gave AMD64 instant respectability, instant and inbuilt adoption (thanks to MS), and a fast introduction, - so fast that Intel had to bow to AMD64.

ThefumigatorTheir processors today are not the disaster people believe. Bad press is all they've got. They had the TLB bug on the Phenom and questionable design in their bulldozer architecture.

Their biggest issue is slow execution and a lack of strategic planning, compounded by the huge debt burden they took on in buying ATI (rather than just licence the IP) which led to having to sell off the foundry business, and their mobile IP to Qualcomm - both of which came back to bite them in the ass, but both that were a necessity thanks to Ruiz paying twice the market value for ATI.

ThefumigatorYeah but AMD still got the initiative and strategy correctly done for those cases, purchased or not, they got the thing done, doing the first 32bit + 64bit processor. Back on these opteron 64 days, intel execs said AMD were crazy. But this changed Intel plans for their Itanium processor.

Their processors today are not the disaster people believe. Bad press is all they've got. They had the TLB bug on the Phenom and questionable design in their bulldozer architecture. But they hit well on APUs and GPUs (purchased or not, AMD has been doing a lot in the GPU area)

Their biggest boost was due to Intels failure to realize that the P4 model would never scale and the desire to have a hugely deep instruction pipeline, they failed and then used their money to buy off OEM's like Dell to keep using their chips and not give AMD a better foothold in the market.

AMD was saved by the IT communities who built and saw the performance of their bought architecture. But along with that the AMD was poor at many scientific calculations due to the same design that gave them great normal precision performance. The side effect of this is why a completely stable overclock in X86 would lead to errors in X64, but at the time few were using the X64 and it wasn't an issue.

It's funny to me, before Maxwell arrived on the scene the AMD people were spouting drivel about how Hawaii was superior across the board and could support DirectX 11.1/11.2, something NVIDIA has yet to do.

I'm not saying that R9 is better anymore, but AMD isn't really horrific at hardware design. Maxwell is just THAT GOOD at energy efficiency.

R-T-BAMD isn't really horrific at hardware design

They are great at what they do, and I'm betting that right now they are optimizing energy efficiency so they can cram more processing units to still fit the power envelope.

R-T-BMaxwell is just THAT GOOD at energy efficiency.

The problem is that even kepler was an efficiency winner over current offerings from the red team. Now, amd has to compete with two generations of energy efficiency optimizations from nvidia. They will need pull a similar kind of paradigm shift optimization nvidia managed to pull.
They could start by looking at why is Maxwell so much more efficient than Kepler.
Nvidia made a job for warp scheduler (dispatcher) in their GPU really really simple :

Before:

After:


No more shared crossbar. Each scheduler has same set of separated resources available (full width of one warp) so there is no need for complex logic to decide which resources are free to take up a new warp, scheduler decision logic gets simple and fast and ultimately low powered.

So if nvidia can shuffle around internals of their gpu and profit this much, amd should be able to do the same, right? The thing is, I don't see how. I would love for someone to convince me otherwise.



Each GCN compute unit has a single scheduler and single scalar processor and a single branch unit, and way too many many vector processing units. Not enough diverse resources to optimize scheduling around.

NVIDIA especially Maxwell, leads the market and beat the AMD with Consumption, more significantly. they surprise the AMD with too little power consumption for the high END GPU which competing with 290x which consumes half of your rig power. generates too much heat and noise. whereas, 970 only draw little power and less noisy and ultra high in performance. if we talk about the market strategy, they blown AMD away with incredibly low price.

Lower price..........Greater performance

so NVIDIA playing efficient game all the way and leading in no time. so with respect to performance, AMD have to figure out their marketing/selling and pricing strategy as well.

Well, r9 290 still offers the highest perf/dollar on the market. The lower price argument doesn't fly at all here. I can buy a new 290 for about 260 eur over here.

Maxwell may offer better value but that depends on what matters to you. The issue is that r9 290 clearly shows how necessary it is to push down the TDP limits at our current state of graphics processing units, and it begs the question how AMD will tackle engineering their 3xx high end without throttle.

It is worrying, because Tonga is not really a big step forward in terms of power budget, it is merely a way to cut costs for AMD really.

Vayra86Well, r9 290 still offers the highest perf/dollar on the market. The lower price argument doesn't fly at all here. I can buy a new 290 for about 260 eur over here.

Most of AMD's current lineup are solid options at performance-per-$, which is great for consumers, but just means that AMD's margins get cut to the bone. AMD, after all, is supplying rebates in order to move inventory (built up in the expectation that the GPU mining craze would continue). The fact that AMD's high end offerings are still being outsold despite this, at a rate of around 4-to-1 tend to point to customer preference rather than pricing.

Vayra86...and it begs the question how AMD will tackle engineering their 3xx high end without throttle.

If AMD don't take the hint and introduce dynamic boost, they may as well fold the tent. Power consumption at the high end might not be a deal breaker ( an enthusiast graphics card owner probably has a decent, if not overspecced PSU), but it is imperative in the mainstream and mobile markets, where customers are more sensitive to power usage/heat/noise, and OEMs battling for battery life.

Vayra86It is worrying, because Tonga is not really a big step forward in terms of power budget, it is merely a way to cut costs for AMD really.

It is actually a strange turn of events that Tonga, while being the latest incarnation of GCN,is also inferior in performance-per-watt to any other GCN architecture GPU. I'd exclude the 7970 GHz Ed., simply because that was a hasty kludge of heavy-handed voltage addition to ensure a quick bump in clock speed to counter the GTX 680/770

buildzoidThat's no longer an option because AMD always reverse engineered intel CPUs today that would take years and years to do because of how many transistors there are.

Yeah, specially the Athlon brand CPU... /sarcasm ;)

Dammit Raj!!!!

You messed up on hells kitchen and now AMD too?!!

Lol