AMD Starts Linux Enablement On Next-Gen "Zen" Architecture

[undersuit]

TBM, FMA4, XOP, or LWP ISAs

All apparently Bulldozer specific instructions... looks like they really are trimming the fat for their new architectures. Are the FMA3 vs FMA4 wars still on going?

[chuckula]

Almost no other company bothered to use them, including ones with very sufficient funding, simply because the results aren't good enough.

Let me fix that for you: Every single other company that is even remotely interested in high-performance and power-efficient computing is desparately trying to use finfets to copy Intel's lead. Samsung being example #1 in finally getting a cellphone chip out in 2015 that includes finfets.

The 20nm process basically died simply because it DIDN'T have finfets and all the "14nm" and "16nm" processes we have heard about from companies other than Intel are actually just 20nm + finfets, which is what 20nm would have been in the first place if it had been done right.

Don't think for one second that other companies aren't interested in Finfets: There's a major difference between not being interested in a technology and being very interested in a technology that's unattainable.

[grndzro]

Not to mentain how all AMD GPUs i've ever had turned out faulty.

Sorry calling bullshit on this one. I have had 10 AMD/ATI gpu's starting from ATI Rage3d/Mach64 and never had a single one fail. That includes 2 1950xtx's that ran at 100c for 2 years. I know it's cool to bash AMD but cmon man don't lie to us.

Nvidia's bumpgate bit me for 3 cards. I don't go on about how Nvidia is garbage. I hate NV but they have good products.

[aaronage]

If they can't catch up to Pascal, Nvidia's gonna leave them in the dust. I'm hoping AMD surpasses Nvidia because believe me; I want to jump ship to team red, but AMD is always lagging too far behind the competition (on both the GPU and CPU side of things) for me to want their hardware over Intel's and Nvidia's. (Not to mentain how all AMD GPUs i've ever had turned out faulty, whereas not a single Nvidia GPU I've ever had has been faulty; and the AMD Phenom really felt like a step back from the Intel Core 2 I had before it, even if it had much better specs on paper...)

Phenom/Phenom II were a good series of CPUs.

Phenom (Agena) was a little hampered by low clocks (caused by the relatively poor 65nm process), but even with that it was still a better design on a better platform to (e.g. a single die with L3 cache vs. MCM communicating over FSB, Hyper-Transport, IMC and 7xx series chipsets).

Phenom II X4 (Deneb) held up against Core 2 Quad in performance, efficiency and overclockability. It didn't beat Core 2 Quad but it was in the same ballpark.

Phenom II X6 (Thuban) was a brilliant chip. Realistically, it was better than Nehalem when you consider the benefits of a hex-core K10 design vs. 4C/8T Nehalem. It was more efficient, didn't have the unpredictable scaling of hyper-threading with Nehalem and overclocked just as well as Deneb. They managed to maintain the same clocks and power draw characteristics of Deneb despite two extra cores. It shows how good AMD's 45nm process was.

Things really went south with Llano and Zambezi - they just couldn't match the huge leap forward that Sandy Bridge brought in terms of perf/watt and IPC. Ivy and Haswell have only extended the gap and AMD are now a few generations behind...

The x86 situation today really shows why we need ARM to take off - but that's a topic for another day lol

[eydee]

So cut-down math blocks and removal of lightweight profiling?
Nice job, AMD. Time to kick all your fans in the balls.

Are you an assembly programmer? It's not a common thing these days. Fans don't even have an idea what these things are and they don't have to know. People use programs, not processor instructions. Let those who write compilers worry about it.

[Jabberwocky]

Sorry calling bullshit on this one. I have had 10 AMD/ATI gpu's starting from ATI Rage3d/Mach64 and never had a single one fail. That includes 2 1950xtx's that ran at 100c for 2 years. I know it's cool to bash AMD but cmon man don't lie to us.

Nvidia's bumpgate bit me for 3 cards. I don't go on about how Nvidia is garbage. I hate NV but they have good products.

Yeah, I second bullshit.

I've owned various Nvidia (dating back to diamond edge) and ATI/AMD (dating back to rage). I've had some shitty cards along the way, but mostly due to bad vendors. I learned the hard way by buying the XFX 7900GT. My first high-end GPU I bought with my own well deserved cash. It had editors award in every PC magazine I used to read. The card broke after 13 months of usage, never overclocked it, same thing happened to 2 other buddies of mine all who bought the same XFX cards. XFX laughed all the way to the bank after selling those cards.

I rejoined Nvidia with a MSI 9600GT, loved that card. Gave it to my young bother for his birthday last year (2014).

I sometimes still take a chance with cheap manufactures like powercolor due to "emotional/personal experience" reasons, but mostly stick to the more expensive stuff. I gained a lot of respect for powercolor after volt modding (soldering) my friend's Radeon 9800 SE and unlocking the extra "pipelines". First just with software, ran it for a couple of weeks with various stress tests and then flashed with a Pro BIOS. It's 2015 and the card still works, it runs Call of Duty 4: Modern Warfare and Command & Conquer 3: Tiberium Wars. Had to replace some chaps on my last working AGP motherboard

PS: When I say cheap I actually mean my country (ZA) makes deals with China, I got my powercolor R9 280X for less than 50% of the Sapphire R9 280X's cost. I'm not exactly sure what the price difference is abroad.

[gens]

Are you an assembly programmer? It's not a common thing these days. Fans don't even have an idea what these things are and they don't have to know. People use programs, not processor instructions. Let those who write compilers worry about it.

FMA4 and XOP are great instruction sets
if compilers could write good code for them, i'm sure many programs would run a lot faster
(pixman could benefit from xop and there were talks about x264 xop optimizations)

[nightmarex]

FMA4 and XOP are great instruction sets
if compilers could write good code for them, i'm sure many programs would run a lot faster
(pixman could benefit from xop and there were talks about x264 xop optimizations)

fma4 is in there actually.. xop is gone you have to read the contents in the tarball.

+ { "CPU_ZNVER1_FLAGS",
+ "Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuSYSC ALL|CpuRdtscp|Cpu387|Cpu687|CpuFISTTP|CpuNop|CpuMM X|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a|CpuABM|CpuLM|Cpu FMA|CpuFMA4|CpuBMI|CpuF16C|CpuCX16|CpuClflush|CpuS SSE3|CpuSVME|CpuSSE4_1|CpuSSE4_2|CpuAES|CpuAVX|Cpu PCLMUL|CpuLZCNT|CpuPRFCHW|CpuXsave|CpuXsaveopt|Cpu FSGSBase|CpuAVX2|CpuMovbe|CpuBMI2|CpuRdRnd|CpuADX| CpuRdSeed|CpuSMAP|CpuSHA|CpuXSAVEC|CpuXSAVES|CpuCl flushOpt|CpuCLZERO" },

[carewolf]

FMA4 and XOP are great instruction sets
if compilers could write good code for them, i'm sure many programs would run a lot faster
(pixman could benefit from xop and there were talks about x264 xop optimizations)

Yep, and there are still things in XOP that can't be done with normal AVX, and won't be available again until 3rd gen AVX-512.

[carewolf]

Let me fix that for you: Every single other company that is even remotely interested in high-performance and power-efficient computing is desparately trying to use finfets to copy Intel's lead. Samsung being example #1 in finally getting a cellphone chip out in 2015 that includes finfets.

The 20nm process basically died simply because it DIDN'T have finfets and all the "14nm" and "16nm" processes we have heard about from companies other than Intel are actually just 20nm + finfets, which is what 20nm would have been in the first place if it had been done right.

Intel's 14nm is the same as Intel's 20nm + finfets. That is pretty much the definition of the technology. The size numbers doesn't mean anything anymore.