AMD CPU speculation... and expert conjecture

[szatkus]

^^
The truth comes out on the hatred for AMD.

Really 4x the transistor count? Isnt that part of the initial problem with a longer pipeline=more transisntors= more lag? So in your mind the latency should be increased 4x... lol.

2nd if amd fixes the idiotic 104 clock cycles for l3 cache latency will change rather fast, compared to intel's 21? clock cycles.

There is more than one aspect to cpu performance, throughput isnt the only factor.

More transistors don't imply longer pipeline.

 

[de5_Roy]

More transistors don't imply longer pipeline.

"4x transistors" or "complexities" doesn't really mean anything. it's a blanket statement and quite possibly a red herring. it shoulda been specified as pipelines or decode or cache or something else.

 

[juanrga]

During Core conference Keller praised the ARM ISA and mentioned that K12 core will have a bigger engine that its x86 sister because the ARM ISA allows to spend more transistors on compute.

I have asked about performance gap and someone who knows some details about the new architectures said me that he is guessing the K12 core will be about 10% faster than its x86 sister. I will try to get more details.

 

[genz]

Well AMD right now won't have anything competitive for the next 10 years without serious financial help in addition to their existing flotation, so it makes sense that instead of trying to regain ground to Intel on single threaded applications they position themselves as a SMT powerhouse and get themselves a large compute lead in the niche markets.

We all know what the bitcoin craze did for them and we all know that Nvidia actually sells a decent amount of Tesla units, which fit the description of APU if I ever head one, as do the Xeon Phi's from Intel. With the right Mantle implementation, and Mantle code in OpenGL and DX12, as well as standalone Mantle to support the DX9/10/11 based systems out there, we could be looking at a lot of units being moved for the low end of scientific compute, as well as a consolidation of the processing capabilities of the iGPU and CPU on AMD's APUs.

I wonder how much additional FLOPS could be pulled out of an AMD CPU if the FPUs on it's iGPU were also processing the data. I also wonder (knowing that AMD Integer units are usually reasonably strong on it's GPUs) how much additional Integer power is usable.

Would it be enough to compete with Intel?

 

[juanrga]

^^
The truth comes out on the hatred for AMD.

Really 4x the transistor count? Isnt that part of the initial problem with a longer pipeline=more transisntors= more lag? So in your mind the latency should be increased 4x... lol.

2nd if amd fixes the idiotic 104 clock cycles for l3 cache latency will change rather fast, compared to intel's 21? clock cycles.

There is more than one aspect to cpu performance, throughput isnt the only factor.

More transistors don't imply longer pipeline.

Evidently.

 

[noob2222]

^^
The truth comes out on the hatred for AMD.

Really 4x the transistor count? Isnt that part of the initial problem with a longer pipeline=more transisntors= more lag? So in your mind the latency should be increased 4x... lol.

2nd if amd fixes the idiotic 104 clock cycles for l3 cache latency will change rather fast, compared to intel's 21? clock cycles.

There is more than one aspect to cpu performance, throughput isnt the only factor.

More transistors don't imply longer pipeline.

going to critique a technicality? what should I have used, >? or >=

longer pipeline equals more transistors is what I said. I didn't say the reverse. the arrow >= would have meant that the equality is one way only. adding more transistors isn't the only option to speed something up. shorten the pipeline >= less transistors, not 4x more. remove transistors = remove lag, both on the pipeline and the cache. adding transistors = adding heat.

let me put it another way. Haswell quad core with an on board graphics has 1.4b transistors, remove the graphics and its somewhere ~1B, PD which is apparently 65% slower uses 1.2b transistors, so according to Juan AMD needs 4.8B transistors to compete ...

No, AMD needs to remove transistors to bring it closer to what Intel has, one way is to shorten the pipeline, another way is to remove the automated design process,

Automated Design = 20% Bigger, 20% Slower

or maybe your just defending juan's hate on AMD.

 

[palladin9479]

Ok guys it's no where near as easy as your talking about, I would go as far as to say most people here have no clue what they are asking for.

The automated design process was only used on the original BD design, since then it's been all by hand modification. This was done as a cost savings method since AMD doesn't have nearly the amount of engineers that Intel does and thus simply can't design CPU's fast enough by hand. To design a new uArch is a monumental task, it's labour intensive and extremely expensive done the traditional way. That is the reason most companies only design a new uArch every five to seven years and just use tweaking and modifications in between major overhauls.

Now for the moduler design, what most of you are crying about, again that was done as a cost savings. This enables AMD to design one set of components and reuse them for consecutive future design's without needing much labor intensive manual reworking. They can take an x86 chip and turn it into an ARM chip by replacing a few pieces without needing to redesign the chip. They can add, replace or alter components without needing major reworking. They can even use a single die to mass produce an entire line of chips vs a different die. This is a huge costs savings, something they desperately need since they predominately target the budget crowd. The FX8350 is $180 USD, the FX6300 is $120 USD and the A10-6800K is $140 USD. Those are stupidly cheap prices for what those parts provide and they won the design's for the new consoles because their chips were moduler and thus cheaper to implement then designing a new custom CPU for each console.

Now I know people are screaming and setting themselves on fire about ST performance, and I'm going to tell you right now they won't ever get anywhere close to Intel. Doing integer math is easy, everyone does it the exact same way and there is no way to do it faster. The only way to increase the speed of integer math is to do multiple operations simultaneously and x86 is not designed to do that. Only VLIW and SIMD languages can do that natively, and those bring their own problems. So the real speed determine is now about how fast you can guess which instructions to load in which order and having the required data loaded locally before the instruction needs it. That is all about branch prediction, instruction order optimization and cache accuracy, things that Intel stomps everyone else on the planet in. AMD's actually about on par with everyone else (Oracle / IBM / ARM / ect..) with their technology, it's that Intel is so damn far ahead of everyone at inline hardware level code optimization that trying to compete with them on that level is suicide. Another hit that AMD takes, due to the modular approach, is that each individual component inside their chips isn't as tightly coupled to the others. Typically during your design process you hand optimize a bunch of connectors and buffers such that the timing is extremely precise, this is expensive and renders the chip unable to be modified easily without redoing the optimization. PD was mostly a hand optimized version of BD, AMD went through each component individually and optimized it further and tightened up the timings and thus we saw a performance increase. They did this again with SR, there is a sizable performance increase but due to their process technology and the thermal limits of that bolted on iGPU, they went with a lower clock rate. SR is basically a more efficient version of PD on a different process, which is why I'm very interested to see what the 860K can do. The 750K was already a very good steal for budget systems that didn't need SFF / LP. Also wish they would design a three module version of SR without the iGPU, would make a good chip for cheaper but not bargin bin level systems, call it the 870K or something. Probably won't happen because it would require another die design which just increase's production costs.

 

[juanrga]

AMD ha no choice but to Drop prices

Yes, with a new arch coming only in 2016 or 2017. AMD only can compete now in price with FX 6000/8000/9000 series.

 

[juanrga]

Well AMD right now won't have anything competitive for the next 10 years without serious financial help in addition to their existing flotation, so it makes sense that instead of trying to regain ground to Intel on single threaded applications they position themselves as a SMT powerhouse and get themselves a large compute lead in the niche markets.

We all know what the bitcoin craze did for them and we all know that Nvidia actually sells a decent amount of Tesla units, which fit the description of APU if I ever head one, as do the Xeon Phi's from Intel. With the right Mantle implementation, and Mantle code in OpenGL and DX12, as well as standalone Mantle to support the DX9/10/11 based systems out there, we could be looking at a lot of units being moved for the low end of scientific compute, as well as a consolidation of the processing capabilities of the iGPU and CPU on AMD's APUs.

I wonder how much additional FLOPS could be pulled out of an AMD CPU if the FPUs on it's iGPU were also processing the data. I also wonder (knowing that AMD Integer units are usually reasonably strong on it's GPUs) how much additional Integer power is usable.

Would it be enough to compete with Intel?

Yes, AMD is pivoting away from the PC market towards what they call emerging markets. Apparently they consider that a small and efficient core will be better placed for their semicustom business than a big core (Haswell-like). Moreover Keller suggested that both cat and Bulldozer lines will be fused in future; this point to some kind of intermediate size core. Big compared to jaguar but small compared to Haswell/Broadwell/Skylake.

An APU is the integration of both a LCU and a TCU. LCUs are optimized for serial work; TCUs are optimized for parallel work. In ordinary APUs, the role of the TCU is executed by a GPU, but it could be another compute unit such as a DSP (HSA specification consider this possibility).

Nvidia Tesla and Xeon Phi are TCUs and don't fit the description of APU. Neither Tesla nor Phi are good at serial work. This is why Intel will be relying on ordinary Xeon CPU for serial phases of applications, whereas Nvidia will be using ARM CPUs from Applied Micro and Power CPUs from IBM.

MANTLE will help AMD to remain competitive with Intel in the gaming PC space despite having a single thread deficit, but will be of no help on server/workstation/HPC and other markets.

About the compute power by using the iGPU, check the hot chips 26 AMD slide. I gave the link above. Therein you can find one benchmark of CPU compared to non-HSA APU, and to HSA APU: using the iGPU under HSA framework increases the performance a lot of compared to traditional approaches.

 

[blackkstar]

At 4GHz Haswell is about 68% faster than 4GHz Piledriver. However, I expect a clock disadvantage for next AMD arch: around 3.0--3.5GHz due to process node. This will increase the gap to about 90% faster. Broadwell will be about 5% faster than Haswell. This increases the gap to 95%. Skylake performance is unknown. But it seems a mayor bump in performance (e.g. L1 and L2 caches are doubled in size compared to Haswell). Let us be conservative and assume another 5% on top of Broadwell. The final gap is of about 100% over Piledriver.

To achieve something faster than Skylake, AMD would increase the complexity of its big cores by about 4x. Yes 4x more transistors! AMD would invest lots of money, and use lots of engineers for a project so complex like that.

However, AMD is firing engineers and the finances are in red numbers again. Papermaster mentioned AMD is not giving to Keller all the resources needed for the next arch. Thus Keller will do the best he can with the limited resources that AMD has.

Moreover, above percentages are for ordinary integer x86 software. If we consider FP/extensions then things look poor for AMD. Piledriver is a 8 FLOP/core architecture. Skylake will be a 64 FLOP/core architecture. To beat Skylake in this aspect, AMD would improve its floating point units by a factor of more than 8x!

Nobody believes that the next arch. will be "faster than Skylake". Specially when the people who is spreading such strong claims in this thread is the same people who said us in the past that Steamroller was going to crush Haswell in performance... cof cof.

Your 30% more single thread performance with lower clocks implies about 50% IPC gain over Piledriver. Well, Piledriver is so low baseline that that figure looks feasible.

I expect AMD next arch. to be behind Haswell in both integer and floating point. But then AMD will use MANTLE to close the gap with Intel (i.e. AMD CPUS will be probably good enough for MANTLE/DX12 games) and will rely on moar cores once again. E.g. A 8-core 'Zen' FX chip against a quad-core Skylake chip.

I have asked about performance gap and someone who knows some details about the new architectures said me that he is guessing the K12 core will be about 10% faster than its x86 sister. I will try to get more details.

Nice, so K12 doesn't have a chance of coming close to haswell either. That's really odd considering Apple A7 is supposed to be really close to Haswell performance. But if K12 is only 10% faster than K12 x86 sister core and K12 x86 sister core is not going to be close to Haswell performance, yet Apple A7 is already there, K12 is going to SUCK! Congratulations Juan you disproved AMD x86 so hard that you disproved K12 ARM in the process!

A7 vs Pentium G3220
http://browser.primatelabs.com/geekbench3/compare/713102?baseline=713418

Pentium G is at 3ghz and A7 is at 1.4ghz, by the way. If A7 does that well against Haswell already and AMD is going to be that far behind in 2 years, well wow. The whole thing is a waste of time. I am glad you disproved AMD entirely.

AMD are on track to catch up on high performance cores

http://www.rage3d.com/articles/hardware/amd_worldcast/

Wow, mind = blown. Not only did Juan disprove AMD, but he disproved Jim Keller too! Truly an incredible feat! Thank goodness we have you to help ensure we get an Intel only future for HEDT!

All we need is for you to disprove JPR now! http://jonpeddie.com/publications/pc_gaming_hardware_market_report/

Can you prove to me how this is wrong and HEDT is the niche part of desktops? Clearly he made some mistakes because he says HEDT is growing, HEDT has higher market share than other desktops, etc.

 

[Reynod]

Lets keep it civil and not start picking on individuals ... be grateful people go out there and surf and bring items here to discuss ... and are brave enough to make some predictions. Sadly they are often not quite accurate, but I have to respect someone willing to try. If your a real Intel fan and don't like AMD then don't post here ... because your negativity is not needed. Go post in the Intel thread and wallow in the glory ...

 

[jdwii]

Intel is boring Its already to amazing we need some Amd!!! ^^^

 

[sapperastro]

Well AMD right now won't have anything competitive for the next 10 years without serious financial help in addition to their existing flotation, so it makes sense that instead of trying to regain ground to Intel on single threaded applications they position themselves as a SMT powerhouse and get themselves a large compute lead in the niche markets.

We all know what the bitcoin craze did for them and we all know that Nvidia actually sells a decent amount of Tesla units, which fit the description of APU if I ever head one, as do the Xeon Phi's from Intel. With the right Mantle implementation, and Mantle code in OpenGL and DX12, as well as standalone Mantle to support the DX9/10/11 based systems out there, we could be looking at a lot of units being moved for the low end of scientific compute, as well as a consolidation of the processing capabilities of the iGPU and CPU on AMD's APUs.

I wonder how much additional FLOPS could be pulled out of an AMD CPU if the FPUs on it's iGPU were also processing the data. I also wonder (knowing that AMD Integer units are usually reasonably strong on it's GPUs) how much additional Integer power is usable.

Would it be enough to compete with Intel?

PC Powerplay, an Australian magazine, did just such tests on the A10 7850k vs the i7 4770k. The overall compute tests still had the intel chip in the lead, though only by the barest of margins.

If HSA was the norm today, the A10 7850k would be one hell of a powerhouse for the money. Unfortunately, HSA isn't the norm today...

I will try to dig up the magazine and post the test used and the marks.

 

[8350rocks]

Ok, so I am going to refute a few things here...

1. SMT does not preclude CMT from the discussion. Everyone associates SMT with hyperthreading, however, hyperthreading is a software trick to accomplish SMT. Meanwhile CMT is a hardware optimization to accomplish the same thing, just in a different way.

2. Juan has no clue what is going on with the new uarch. That much is clear. AMD is not 100% behind Intel in anything, nor will they be. I have it on good authority, the design goal for the new uarch is to be equal performance to skylake within 5% MoE.

3. AMD has basically given Jim Keller the keys to the kingdom. They are giving him pretty much whatever he wants. They are also paying a lot of attention to benchmarks, and becoming far more proactive in that space.

4. Optimization for their hardware, through software, is a big priority. When I went to AMD, they wanted to know how the drivers were doing for GPUs. They want to be number 1 in discrete GPUs.

5. HEDT is their highest volume high margin part, and it is good for OEMs as well. Their margins on HEDT parts are still very good, and they will not be abandoning that sector any time soon.

 

[de5_Roy]

what is "zen"? is it the codename for the new x86 uarch? i am tired of typing "k12's sister x86 uarch" every time. i'll even take "prairie dog town fork red".

 

[8350rocks]

what is "zen"? is it the codename for the new x86 uarch? i am tired of typing "k12's sister x86 uarch" every time. i'll even take "prairie dog town fork red".

So, if I were in the know...and knew that K12 was the ARM uarch, knowing that Zen is also a new uarch, I would suspect that Zen would be the unannounced codename for the x86 uarch...

*hopefully that was not too cryptic*

 

[blackkstar]

Ok, so I am going to refute a few things here...

1. SMT does not preclude CMT from the discussion. Everyone associates SMT with hyperthreading, however, hyperthreading is a software trick to accomplish SMT. Meanwhile CMT is a hardware optimization to accomplish the same thing, just in a different way.

2. Juan has no clue what is going on with the new uarch. That much is clear. AMD is not 100% behind Intel in anything, nor will they be. I have it on good authority, the design goal for the new uarch is to be equal performance to skylake within 5% MoE.

3. AMD has basically given Jim Keller the keys to the kingdom. They are giving him pretty much whatever he wants. They are also paying a lot of attention to benchmarks, and becoming far more proactive in that space.

4. Optimization for their hardware, through software, is a big priority. When I went to AMD, they wanted to know how the drivers were doing for GPUs. They want to be number 1 in discrete GPUs.

5. HEDT is their highest volume high margin part, and it is good for OEMs as well. Their margins on HEDT parts are still very good, and they will not be abandoning that sector any time soon.

I probably should have toned it down against him but I get a little frustrated when I see people operating on logic that doesn't even make sense yet talking so confidently. We basically saw a post that contradicted everything he's said in the past while contradicting reality in nearly every bullet point. If someone is going to spread such misinformation that blatantly contradicts reality while going "muh inside sources", someone needs to call him out on it before people take it seriously. The math he posted doesn't even work out properly such that K12 and Zen would be remotely competitive parts with existing ARM parts, let alone ARM and x86 parts a year or two from now. And he made such a statement only through his own posting.

I am really happy to see 4 and 5. AMD has been basically bullied by Intel and Nvidia in terms of software. It's so bad I compile my entire OS from scratch just to avoid it, and every AMD chip I own is a lot happier getting optimized GCC code.

I hope that we get to see AMD GPU OpenCL drivers improve so that we can get Blender kernel to compile. The day that happens I will more than likely pick up a high end AMD graphics card, probably two of them at least. Baking and rendering on something like Hawaii is a dream come true.

@de5_roy, Zen is the rumored name for the K12 x86 sister core but it's not official yet. It is an interesting name. Zen is a Buddhist term that states you should be focused on what your own nature while excluding other things to seek your own enlightenment and higher state of being. It focuses on personal expression and daily life for the benefit of others. Zen is a Southern Chinese thing and K12 mountain is located close to southwestern China in India, which is also a strong emerging market.

K12 mountain is very difficult to climb because of military presence and politics. Consider the hostile environment of software and marketing toward AMD and you can see similarities.

So the two code names are a bold statement about how AMD has a tough mountain to climb that most people won't be able to climb, and the way to do it is through focusing on what AMD is good at while seeking a higher state of being that benefits others and ignoring fighting Intel and company directly as they should be improving themselves and growing on what they are good at instead of trying to be good at what others (Intel) is good at.

I guess you could say the K12 and Zen code names are about AMD climbing a difficult mountain with a huge military and corrupt political environment, and that they will climb it not by fighting that military and political system by becoming a stronger military and political system, but by focusing on what AMD does well using their strongest points to become something new and stronger that exceeds what they were in the past. And they believe that by doing this and focusing on making things better for others (game developers with Mantle and many cores, good consumer choices, etc) that they can pull it off.

 

[etayorius]

what is "zen"? is it the codename for the new x86 uarch? i am tired of typing "k12's sister x86 uarch" every time. i'll even take "prairie dog town fork red".

So, if I were in the know...and knew that K12 was the ARM uarch, knowing that Zen is also a new uarch, I would suspect that Zen would be the unannounced codename for the x86 uarch...

*hopefully that was not too cryptic*

Man, i really hope you`re right... but your aim has not been the best in the last couple of years... i also don`t think Keller will do a miracle on AMD with Red numbers, i could see AMD new Arch get on par with Haswell level of performance which is almost 50% faster than Piledriver, but by the time Skylake arrives Intel could leap by another 20-30% at a minimum. AMD would probably have to increase performance by almost 100%, and to be honest i doubt that AMD or anyone else knows the exact level of performance Skylake may have, even less to give the claim they will be on par with Skylake, i am not dissing your post though, just seems very unlikely that AMD can do this sort of leap on their new Arch with so much money leaking from everywhere, i can see Raja getting all the budget he wants since the GPU division is keeping AMD afloat, but i don´t think Keller have the same level of resources... i even saw him jokingly asking for more resources on two different videos, which in both he gets rejected.

 

[bmacsys]

Gaming PC's are still climbing its been stated by Newegg and tiger and well everyone in game dev's indie gaming is really huge now more so on PC's. PC's are dying out in general but PC gaming is climbing really people and claim otherwise but yet again facts are not on their side. The truth is most people already have a good enough pc for their little facebook games. Steam has 75 million users i wouldn't call PC gaming a niche market.

Edit off topic 100% just mad! I hate pulling data from the freaking database at this school so darn slow. Horrible database admins. Its also so unsecured its sad i actually am filing a complaint right now.

75 million Steam pc gamers are just that, a niche. A veritable drop in the bucket. Now you take that 75 million users and divide that by how many years they go between upgrades. The number then gets even smaller. Cellphones and tablets are literally selling in the billions.

 

[jdwii]

what is "zen"? is it the codename for the new x86 uarch? i am tired of typing "k12's sister x86 uarch" every time. i'll even take "prairie dog town fork red".

So, if I were in the know...and knew that K12 was the ARM uarch, knowing that Zen is also a new uarch, I would suspect that Zen would be the unannounced codename for the x86 uarch...

*hopefully that was not too cryptic*

Man, i really hope you`re right... but your aim has not been the best in the last couple of years... i also don`t think Keller will do a miracle on AMD with Red numbers, i could see AMD new Arch get on par with Haswell level of performance which is almost 50% faster than Piledriver, but by the time Skylake arrives Intel could leap by another 20-30% at a minimum. AMD would probably have to increase performance by almost 100%, and to be honest i doubt that AMD or anyone else knows the exact level of performance Skylake may have, even less to give the claim they will be on par with Skylake, i am not dissing your post though, just seems very unlikely that AMD can do this sort of leap on their new Arch with so much money leaking from everywhere, i can see Raja getting all the budget he wants since the GPU division is keeping AMD afloat, but i don´t think Keller have the same level of resources... i even saw him jokingly asking for more resources on two different videos, which in both he gets rejected.

Lets be honest here if Intel leaped another 20-30% which is HUGE for 1 gen i would be surprised. The last couple of tocks intel had was around 5-7% improvements and i bet they will be focusing more on the igpu then the CPU anyways. Amd on the other hand had Piledriver out for 2 years now. I don't have faith in Amd when it comes to CPU's since they haven't been competitive with Intel in over 8 years, but Jim keller has the experience and he isn't making any wild claims like Amd did in the past with the head guys. I just hope it won't be some small core design.

 

[jdwii]

Gaming PC's are still climbing its been stated by Newegg and tiger and well everyone in game dev's indie gaming is really huge now more so on PC's. PC's are dying out in general but PC gaming is climbing really people and claim otherwise but yet again facts are not on their side. The truth is most people already have a good enough pc for their little facebook games. Steam has 75 million users i wouldn't call PC gaming a niche market.

Edit off topic 100% just mad! I hate pulling data from the freaking database at this school so darn slow. Horrible database admins. Its also so unsecured its sad i actually am filing a complaint right now.

75 million Steam pc gamers are just that, a niche. A veritable drop in the bucket. Now you take that 75 million users and divide that by how many years they go between upgrades. The number then gets even smaller. Cellphones and tablets are literally selling in the billions.

For one how many of those same people also bought a toaster oven or microwave its so sad that people in this industry still think one for all. Its never going to happen that way its been claimed since the 80's that people would only have 1 device but its not like that at all.
People want lots of options they want 4 HD tv's in their house with at least 1 blu-ray player one console and 2+ PC's or laptops, maybe even a netflix setup box. Plus the kids and wife might want some ipads as well. Then through in some stereos.
I think we should all copy and paste this -casual gamers play on tablets and smartphones they don't play on PC's or consoles anymore or at least not as often-
Casual gamers are not the same as hardcore gamers. Look at the PS4 sales and now compare that to the Wii U a casual system its market is done those casual gamers now play on tablets and smartphones, while the PS4 is hitting record sells in the console space.

Its getting to sound like a broken record "PC gaming is dying" but its just that a broken record that needs to be thrown away.

Spoiler

http://techreport.com/news/25136/ea-financials-show-growing-pc-game-revenue
" EA's PC revenue was higher than its Xbox 360 revenue ($256 million) or its PlayStation 3 revenue ($238 million)."
http://arstechnica.com/gaming/2014/04/analyst-pc-gaming-now-brings-in-more-money-than-console-gaming/
"DFC Intelligence analyst David Cole that PC gaming is far from dying—and it's actually outperforming the console sector overall these days."
http://www.pcgamer.com/2014/07/11/pc-gaming-hardware-market-worth-more-than-21-5-billion-globally-new-report-says/
"We continue to see a shift in casual console customers moving to mobile," JPR Senior Analyst Ted Pollak said. "While this is also occurring in the lower-end PC gaming world, more money is being directed to mid- and high-range builds and upgrades by gamers."
http://techreport.com/news/24575/pc-gaming-market-grew-8-in-2012
"The PC Gaming Alliance claims the market was worth $20 billion in 2012, an increase of 8% over the previous year and a whopping 90% over the past five years."
http://www.forbes.com/sites/jasonevangelho/2014/04/28/as-global-pc-game-revenue-surpasses-consoles-how-long-should-console-makers-keep-fighting/
"The PC isn’t dying, it’s merely redefining its form factor"
This statement i like and i might be switching to a small build in the next few years instead of a big case

So based on this information i really do not see how people keep getting so confused about all of this. Its like we won the war and people are claiming we are dying off, instead its just getting started.
Actually instead of lying to are self's i would actually like to find more evidence to why hardcore PC gaming is growing and why people are buying more higher-end hardware now. Why is EA seeing more money on the PC then even the 360.

 

[de5_Roy]

So, if I were in the know...and knew that K12 was the ARM uarch, knowing that Zen is also a new uarch, I would suspect that Zen would be the unannounced codename for the x86 uarch...

*hopefully that was not too cryptic*

i am just happy that i won't have to type a long codename. zen is perfect. i'd shorted "prairie dog town fork red" down to pdtfr, btw.

@blackkstar: the mountain is called K2, 2nd higest after mt. everest. i assume k12 is derived from previous k-series uarched like k6, k10 etc. but amd would be able to accurately confirm if it's correct.

@guys: should stop comparing zen to haswell. seriously. haswell is a 4/5/watever generation evolved, tuned core while zen is 1st gen. x86 uarch from amd. the comparison isn't fair, imo. same with k12 or a57 performance expectations. both are 1st interation cores from amd and arm resp. according to skylake rumors, carrizo or zen might be able to compete if intel's recent "converged core" hints turn out to be true. how much, is difficult to say. i hope cpus from either uarch will undercut intel's mainstream high end and offer value to customers. and lastly, don't buy into astroturfing and keep your expectations low.

 

[blackkstar]

Man, i really hope you`re right... but your aim has not been the best in the last couple of years... i also don`t think Keller will do a miracle on AMD with Red numbers, i could see AMD new Arch get on par with Haswell level of performance which is almost 50% faster than Piledriver, but by the time Skylake arrives Intel could leap by another 20-30% at a minimum. AMD would probably have to increase performance by almost 100%, and to be honest i doubt that AMD or anyone else knows the exact level of performance Skylake may have, even less to give the claim they will be on par with Skylake, i am not dissing your post though, just seems very unlikely that AMD can do this sort of leap on their new Arch with so much money leaking from everywhere, i can see Raja getting all the budget he wants since the GPU division is keeping AMD afloat, but i don´t think Keller have the same level of resources... i even saw him jokingly asking for more resources on two different videos, which in both he gets rejected.

20% or more is a pipe dream. Haswell was supposed to bring a lot of big architectural changes, and it did. And it didn't materialize into any sort of strong performance. Single thread is at a stand-still and it's very difficult to go beyond what we have. There was a lot of low hanging fruit back in the day when we did see 20%+ gains in performance. That is all gone now.

Not to mention Intel is more focused on efficiency for mobile and servers than they are for raw performance. Given how Intel has been behaving since Sandy Bridge, I would expect Intel to continue to make a design decision that increases efficiency over a design decision that increases performance.

75 million Steam pc gamers are just that, a niche. A veritable drop in the bucket. Now you take that 75 million users and divide that by how many years they go between upgrades. The number then gets even smaller. Cellphones and tablets are literally selling in the billions.

The fact that mobile is going to wipe out HEDT runs on a lot of assumptions. It's assuming that ARM will continue to grow and will not reach a point where most ARM users have a device that's good enough and doesn't need to be ugpraded constantly. 4 years ago a new ARM chip was a significant upgrade over the last model. Now we are reaching the point where we're getting 4 or 8 cores and massive resolution screens because ARM is already outgrowing iOS and Android software ecosystems. The pace of upgrades for ARM devices is going to wind down. ARM and ARM makers know this as well, it is why they are pushing ARM into new markets like low end laptops and servers. The ARM for mobile devices is not going to last forever at current growth rates.

Also, JPR I believe, found that these sort of ARM devices actually led to more HEDT sales, as it was getting users who weren't previously interested in gaming to pursue gaming even further on a stronger platform.

Android and iOS are also coming up with software solutions that require less demanding hardware, like ARTS and Metal. You are really going to see ARM slow down when performance reaches good enough, because power consumption and battery life will be the end goal, and the display is the most power hungry part of a mobile device.

However, if you assume 75 million users and .29% of market share with $160 average sale price, that's almost $35 million in revenue. That's not that bad. It's not enough to support an architecture designed just for HEDT, but if you can make it up in some other markets you are good. That also doesn't include any of the other parts that come from HEDT cores, like all non-cat core APUs that aren't Stars and lower FX models. Hex core has decent numbers and it's a safe bet to imagine AMD has solid numbers here, far better than Intel hex core.

If you assume FX 6000 series is on 1% of steam machines, that's 750,000 FX 6000 CPUs. Assuming $120 for an FX 6000 series part, that's $90 million in revenue.

I will admit I just pulled out 1% from thin air, but my point is that a small percentage on SHS can mean a whole lot of revenue. If AMD can release a solid CPU in FX 6000 series price range that beats Intel, they will have a lot of revenue on their hands. If they manage to spread the multi-core Mantle love to DX and OpenGL, they can skimp on spending a ton of money developing a single thread beast and instead make a really good budget gaming chip with alright single thread that comes in at a killer price point.

Given the trend JPR found of ARM gaming being a gateway to PC gaming, AMD can make an absolute killing in emerging markets where they only have mobile ARM devices.

And now that we are in the multi-core era, and the difference between a $80 Pentium and a $1000+ Extreme Edition is primarily the number of cores and threads, AMD can just throw a ton of cores at the problem and create a high end device without having to invest a ton optimizing single core performance.

They are playing for the long term. You also have to remember that they get to see a lot of market information we don't. Those JPR articles that state things like enthusiast desktop is expected to see growth is not the entire article or dataset. It's a small chunk designed to get your attention and to get you to buy the whole dataset and analysis.

 

[juanrga]

what is "zen"? is it the codename for the new x86 uarch? i am tired of typing "k12's sister x86 uarch" every time. i'll even take "prairie dog town fork red".

So, if I were in the know...and knew that K12 was the ARM uarch, knowing that Zen is also a new uarch, I would suspect that Zen would be the unannounced codename for the x86 uarch...

*hopefully that was not too cryptic*

Man, i really hope you`re right... but your aim has not been the best in the last couple of years... i also don`t think Keller will do a miracle on AMD with Red numbers, i could see AMD new Arch get on par with Haswell level of performance which is almost 50% faster than Piledriver, but by the time Skylake arrives Intel could leap by another 20-30% at a minimum. AMD would probably have to increase performance by almost 100%, and to be honest i doubt that AMD or anyone else knows the exact level of performance Skylake may have, even less to give the claim they will be on par with Skylake, i am not dissing your post though, just seems very unlikely that AMD can do this sort of leap on their new Arch with so much money leaking from everywhere, i can see Raja getting all the budget he wants since the GPU division is keeping AMD afloat, but i don´t think Keller have the same level of resources... i even saw him jokingly asking for more resources on two different videos, which in both he gets rejected.

Well, ~100% is also the number that I posted before, where I also considered the expected clock gap, but that number was for ordinary integer x86 code. For floating point code the gap is much bigger, Piledriver peaks at 8 FLOP/core, whereas Skylake will do 64 FLOP/core. Thus for floating point workloads, AMD would increase performance by a factor of 8x! to caught Skylake performance.

 

[juanrga]

I insisted on asking him about the 10% performance gap between K12 and Zen, but he couldn't disclose details of the new arch. I have to admit that 10% looks small for me. ARM is expected to be about 20--30% more powerful than x86 at those power consumption levels: 50--90W.

If his number is correct, I only can hypothesize that the ARM core is being capped for maintaining socket/pin compatibility with its sister x86 core

What he has said is that the new design abandons the CMT family approach (Bulldozer) entirely. He has not confirmed if the core will be BIG (Haswell-like size) or only a scaled up version of jaguar core (another source said me the cores will be small he said explicitly "Puma-like minicore").