End Of The Line For AMD FX Processors. (UPDATE)

[TechFan@ic]

fyi AMD bought ATI in 2003 so you're wrong there they simply changed the branding in 2007

The acquisition was completed at the end of 2006.

http://www.amd.com/us/press-releases/Pages/Press_Release_113741.aspx

[Kuzma]

The acquisition was completed at the end of 2006.

http://www.amd.com/us/press-releases/Pages/Press_Release_113741.aspx

I stand corected (although the OP is still wrong just not AS wrong)

[LAwLz]

This is where you "heard" it .

http://linustechtips.com/main/topic/48571-intel-amd-architectural-discussion-how-far-ahead-intel-really-is/

There is a ton of bias in that post, sadly. Take it with a big grain of salt people. And like I said before, APUs might be the future, but they are not the present. Float-points are not the only area Intel is far ahead of AMD either, another example would be branch prediction. They also have different approaches when it comes to decoding (the most obvious difference being that Intel uses a µop decoder queue, one of the reasons why Intel CPUs uses so much less power, and it also gives a performance increase). If you then take into consideration the OoO differences, the execution engine, the different instruction sets etc, the differences becomes way bigger than just "AMD is good at integer and Intel is good at FP".

CPUs are far more complex than that post makes them out to be.

[TechFan@ic]

There is a ton of bias in that post, sadly. Take it with a big grain of salt people. And like I said before, APUs might be the future, but they are not the present. Float-points are not the only area Intel is far ahead of AMD either, another example would be branch prediction. They also have different approaches when it comes to decoding (the most obvious difference being that Intel uses a µop decoder queue, one of the reasons why Intel CPUs uses so much less power, and it also gives a performance increase). If you then take into consideration the OoO differences, the execution engine, the different instruction sets etc, the differences becomes way bigger than just "AMD is good at integer and Intel is good at FP".

CPUs are far more complex than that post makes them out to be.

My post isn't biased, well not according to the people that have been commenting on it.

And I do believe you are referring to the microOP cache that cuts the misprediction penalty by 3 cycles.

AMD's & Intel's microOPs are quite different, they're not comparable in fact : http://www.realworldtech.com/bulldozer/5/ .

The main reason Intel CPUs consume less power is because of the smaller die size as a result of the smaller manufacturing process that is a fact.

Certainly cutting the branch misprediction penalty does improve efficiency by improving performance but does not result in a lower power consumption on its own.

You also need to realize that intel's AVX (& AVX2) instruction set is a flaoting point instruction set, and in fact the upcoming AVX-512 instruction set is going to be used in Intel's Knight's Landing architecture a sucessor to the current line of many core Xeon Phi, essentially a GPU.

So Intel is actually off-loading its floating point work to the GPU which is exactly what AMD has already done with Kaveri.

The APU and heterogeneous architectures are the future, AMD is leading it and Intel is saliently admitting it.

[BookerDeWitt]

This as been AMD's plan all a long, get the ball rolling with Llano and Trinity then release HSA with Kaveri and Carrizo.

[Dylan522p]

AMD are still making APUs, Much like Intel DON'T make CPUs anymore, they make APUs (GPU+CPU on chip) so not much has changed apart from the fact they'll all have a GPU onboard from now on, most likely we will see an APU from them soon that matches the performance of the 8350

Not possible. They cannot make a FX8350 and a decent iGPU onto the same die because that would produce too much heat and be too big of a die. They will not have FX8350 in CPU from the APU line until atleast 2016.

[Dylan522p]

You are aware that Intel has stopped making non-apu processors three years ago, aren't you?

LGA2011 i7s and Xeons disagrees with you

 

AMD will still be making CPUs, They go by the name: Opteron 

Which have worse price to performance than Intels Xeons which is ridiculous and they guzzle power compared to Xeons.

 

I know less about opteron than Xeon Btw, there are xeons with IGPU. Do those that dont have igpu, have disabled one or???

Don't have GPUs. All the space is used for CPU performance which is why the FX line was abel to keep up with Intel's midrange offerings (FX used all the die space and transistor budget for CPU while Intel's midrange did not.)

 

What? I see new CPU's for every year until 2015...if it's an APU it doesn't necessarily mean it's low-end, maybe their new Steamroller cores perform a lot better than Bulldozer so you get much more performance, or there can be APU's with more cores than 4 (altough that wouldn't match the TDP). They would be fools if they didn't secure the high-end segment, and since they only make APU's now, it must mean that new A10 chips will be real powerhouses with an iGPU.

Steamroller is about 15-25% faster but 15% clockspeed lower. APU's are not getting more than 4 cores for another year and half atleast.

 

im really happy they are moving to high end APUs

high end and APU do not beliong in the same scentece until HSA has been fully implemented and evolved.

 

So, 8 core APU's then?

 

Damnit AMD...

Not possible until true 14nm for non-Intel which is 2016.

[matin94]

this sucks fucking ass. APU are good but dont think they're good for high end gaming. now there is no price to performance anymore. theres just gonna be Price Price Price to little changes

[EChondo]

Not possible until true 14nm for non-Intel which is 2016.

How about 6-core?

 

This is absolute BS. And to think my next build was going to be AMD parts, I'm not sacrificing for a damn APU. Like GoodBytes said, Intel i5 and i7 available for $500 and $700 respectively!

[amdart11]

This really sucks, as and Intel user, I actually really like amd products, especially the fx series. Without them, Intel will stop innovating to stay on top of the market. All we can do is wait and see what happens after. 

[TopWargamer]

[Dylan522p]

How about 6-core?

 

This is absolute BS. And to think my next build was going to be AMD parts, I'm not sacrificing for a damn APU. Like GoodBytes said, Intel i5 and i7 available for $500 and $700 respectively!

Honestly, I doubt they even will go 8 core like I said when the non-intel fabs get to true 14nm. AMD is going for the play the game where they keep an adequate quadcore, but a decent iGPU and they are going to increase that each shrink they can. Problem is Intel is doing the same thing except their iGPU is more efficent and better in every sku except the A10 on desktop. Even laptop A10's are garbage compared to intels offering. Intel won't increase the prices, at this point they will just keep them the same, while not increasing performance for another year and half. That's right, we are getting NO CPU performance increases FROM ANYONE in the consumer segment until Skylake. Broadwell completely reworks GPU but no IPC improvement and Haswell E is high end workstation and server stuff.

[Dylan522p]

This really sucks, as and Intel user, I actually really like amd products, especially the fx series. Without them, Intel will stop innovating to stay on top of the market. All we can do is wait and see what happens after. 

Intel has and will never stopped innovating at the high end of the market. They already have been only competing with themselves there for years. Where they will stop is the enthusiasts level where noone cares for iGPU or power consumption. Every other market cares about that.

[TechFan@ic]

Honestly, I doubt they even will go 8 core like I said when the non-intel fabs get to true 14nm. AMD is going for the play the game where they keep an adequate quadcore, but a decent iGPU and they are going to increase that each shrink they can. Problem is Intel is doing the same thing except their iGPU is more efficent and better in every sku except the A10 on desktop. Even laptop A10's are garbage compared to intels offering. Intel won't increase the prices, at this point they will just keep them the same, while not increasing performance for another year and half. That's right, we are getting NO CPU performance increases FROM ANYONE in the consumer segment until Skylake. Broadwell completely reworks GPU but no IPC improvement and Haswell E is high end workstation and server stuff.

Actually intel's GPU architecture requires 60% more die space to match AMD's GCN architecture in performance so they're significantly less area efficient.

So even with a larger node AMD can maintain the GPU performance lead, you need to understand that Intel's Iris Pro (GT3) GPU powered chips require a massive die, one that's twice as large as the 4770K or AMD's A10 APUs and far more expensive to manufacture.

http://cdn3.wccftech.com/wp-content/uploads/2013/09/6.jpg

It's the one on the far left and please note the eDRAM on top of the CPU/GPU die as well

 

I'm willing to bet that Intel's broadwell CPUs will have exactly the same die layout with 4 cores, 6MB L3 Cache and GT3 graphics but without the embedded DRAM with the jump from 22nm to 14nm Intel will bring the die size down and reduce manufacturing costs in the process as well as total power consumption.

 

[nicehat]

LGA2011 i7s and Xeons disagrees with you

 

Which have worse price to performance than Intels Xeons which is ridiculous and they guzzle power compared to Xeons.

 

high end and APU do not beliong in the same scentece until HSA has been fully implemented and evolved.

 

 

Opterons are basically a bunch of Piledriver cores taped together onto a die. Keeping this in mind, Its obvious the Xeon is more efficient. All you have to do is go back and look at the TDP of Haswell vs Piledriver. But when you start talking about price:performance being worse, I cant understand how or why you are saying that. If you even bother to look at the price of an Opteron and compare it to a similar Xeon you would not make that statement with a straight face

 

Opterons have many advantages that really are overlooked. For starters, they are sold by core count and clock speed, everything else (feature set, cache size, RAM speed, bus speed) is identical between processors in the same model line. In addition, there is no such thing as a "crippled" Opteron CPU. Not so with Xeons, only the really expensive ones are "fully functional" with full-speed buses, full-speed memory controllers, all of the bells and whistle features and extras enabled, (the full amount of cache, for instance). Opterons are also supported for more than 1-2 generations which allows the end user to extend their investments which translates into additional savings over time. Not to mention that boards are far less expensive than their Xeon counterparts.

 

Wait what were we talking about again? Oh yea price to performance being better for Xeons... Id like to disagree 

 

Id like to add that there are no high end APUs in the market. All of the current offerings are buildups and stop gap solutions to compete with Intel integrated GPU solutions. The true high end APU flagship/battleship will feature full out excavator cores with a unified FPU process instead of a shared one. 

[Dylan522p]

Actually intel's GPU architecture requires 60% more die space to match AMD's GCN architecture so they're significantly less area efficient.

So even with a larger node AMD can maintain the GPU performance lead, you need to understand that the Intel's Iris Pro GPU uses a massive die, one that's twice as large as the 4770K or AMD's A10 APUs and far more expensive to manufacture.

It's the one on the far left and please note the eDRAM on top of the CPU/GPU die :

http://cdn3.wccftech.com/wp-content/uploads/2013/09/6.jpg

 

I'm willing to bet that Intel's broadwell CPUs will have exactly the same die layout with 4 cores and GT3 graphics but without the embedded DRAM.

 

Nope, you are looking at the percentage compared to die of each CPU alone not to each other, when you need to compare transistor count and power usage. Intel's Iris Pro is not that big at all. The eDram also isn't on the same die. It is the same size as the desktop A10's, but with significantly less power consumption and way more CPU power. (4960HQ is the specific processor I am referring to which has full Iris Pro. 264mm2 vs 246 of the A10. Also, Broadwell is not a small change to iGPU. It is the biggest one they have had since 2005ish. 

 

 

Opterons are basically a bunch of Piledriver cores taped together onto a die. Keeping this in mind, Its obvious the Xeon is more efficient. All you have to do is go back and look at the TDP of Haswell vs Piledriver. But when you start talking about price:performance being worse, I cant understand how or why you are saying that. If you even bother to look at the price of an Opteron and compare it to a similar Xeon you would not make that statement with a straight face

 

Opterons have many advantages that really are overlooked. For starters, they are sold by core count and clock speed, everything else (feature set, cache size, RAM speed, bus speed) is identical between processors in the same model line. In addition, there is no such thing as a "crippled" Opteron CPU. Not so with Xeons, only the really expensive ones are "fully functional" with full-speed buses, full-speed memory controllers, all of the bells and whistle features and extras enabled, (the full amount of cache, for instance). Opterons are also supported for more than 1-2 generations which allows the end user to extend their investments which translates into additional savings over time. Not to mention that boards are far less expensive than their Xeon counterparts.

 

Wait what were we talking about again? Oh yea price to performance being better for Xeons... Id like to disagree 

 

Id like to add that there are no high end APUs in the market. All of the current offerings are buildups and stop gap solutions to compete with Intel integrated GPU solutions. The true high end APU flagship/battleship will feature full out excavator cores with a unified FPU process instead of a shared one. 

No, you must not have paid attention to the Ivy Xeons. Before I would have agreed the Opterons were better price to performance if you hosted your own racks. With the Ivy update, you can go with the 2660v2 which offers the same performance as the highest end Opterons but with the same price and you save tons in the long run, especially if you don't host your own racks. Also your mention of Opterons lasting more gens is false. Opteron doesn't get an update till excavator in which they will use a different socket.

[TechFan@ic]

Nope, you are looking at the percentage compared to die of each CPU alone not to each other, when you need to compare transistor count and power usage. Intel's Iris Pro is not that big at all. The eDram also isn't on the same die. It is the same size as the desktop A10's, but with significantly less power consumption and way more CPU power. (4960HQ is the specific processor I am referring to which has full Iris Pro. 264mm2 vs 246 of the A10. Also, Broadwell is not a small change to iGPU. It is the biggest one they have had since 2005ish. 

264mm at 22nm equals 384mm at 32nm without the eDRAM which is the node for the A10, that's 56% larger without the eDRAM which if you add puts the Intel chip at well over 450nm, also remember that Richland uses the old VLIW4 GPU architecture, while Kaveri will use GCN, AMD states a 30% GPU performance improvement from Richland to Kaveri.

This puts the the 7850K at the same performance level as the i7 4950HQ at 55W but with almost half the number of transistors i.e half the die size at the same process node, you can extrapolate the die size of Kaveri from this picture to be roughly at 237mm.

Intel will very likely use a die shrink of the i7 4950HQ but without the embedded DRAM for Broadwell since it increases the GPU unit count from 10 on the i7 4770K to 16 which should give Intel their 40% GPU improvement they've been talking about under lower power consumption.

You also need to remember that Kaveri doesn't have to go against Broadwell, since Broadwell's coming at the tail end of 2014 just one quarter away from Carrizo.

[Dylan522p]

264mm at 22nm equals 384mm at 32nm without the eDRAM which is the node for the A10, that's 56% larger without the eDRAM which if you add puts the Intel chip at well over 450nm, also remember that Richland uses the old VLIW4 GPU architecture, while Kaveri will use GCN, AMD states a 30% GPU performance improvement from Richland to Kaveri.

This puts the the 7850K at the same performance level as the i7 4950HQ at 55W but with almost half the number of transistors i.e half the die size at the same process node, you can extrapolate the die size of Kaveri from this picture to be roughly at 237mm.

You forget that the 4 cores are also larger in number of transistors and Intel's additional L3 cache. You can knock off 216mm² of the die which is just purely CPU related. 184mm² is still a huge die for a GPU for that performance but you have to remember that it is running as a much lower clock speed than it can for power consumption. There have been a few people who have gotten the GPU to run at 2ghz almost with some really low level tweaking and OCing, not sure how though with a locked multiplier and voltage.

[TechFan@ic]

You forget that the 4 cores are also larger in number of transistors and Intel's additional L3 cache. You can knock off 216mm² of the die which is just purely CPU related. 184mm² is still a huge die for a GPU for that performance but you have to remember that it is running as a much lower clock speed than it can for power consumption. There have been a few people who have gotten the GPU to run at 2ghz almost with some really low level tweaking and OCing, not sure how though with a locked multiplier and voltage.

The future is exciting is all what we really need to say.

AMD leading in graphics & Intel leading in x86 I think will remain to be the case for the foreseeable future, but both are closing the gap on each other in either area.

[EpicGeekonFire]

Guess Intel will be giving us even smaller increases in CPU now because they'll basically own the market now...

AMD must have given up with Intel... just screw them and move on...

[nicehat]

No, you must not have paid attention to the Ivy Xeons. Before I would have agreed the Opterons were better price to performance if you hosted your own racks. With the Ivy update, you can go with the 2660v2 which offers the same performance as the highest end Opterons but with the same price and you save tons in the long run, especially if you don't host your own racks. Also your mention of Opterons lasting more gens is false. Opteron doesn't get an update till excavator in which they will use a different socket.

 

So if you are an IT company, you are willing to throw out last gens boards, and CPUs for the latest Ivy update and throw some more cash at a new 1155 board (assuming you are greater than a generation old)? That doesnt sound kind to your bottom line, or any argument for Intel having a better price:performance. The fact is that AMD has continued to support the G34 platform since 2010. I cant say that Intel can boast about the same support for their sockets. That is what I meant when I wrote about extended board support. Also, a quick google search indicates that you can get a 6380 based server for the price of a Xeon E5-2640 based server. The Opteron machines need more energy to do their job, but once again you get better performance per dollar than Intel's midrange offerings

[Dylan522p]

The future is exciting is all what we really need to say.

AMD leading in graphics & Intel leading in x86 I think will remain to be the case for the foreseeable future, but both are closing the gap on each other in either area.

Yup. Will be interesting to see who catches the other first.

 

 

So if you are an IT company, you are willing to throw out last gens boards, and CPUs for the latest Ivy update and throw some more cash at a new 1155 board (assuming you are greater than a generation old)? That doesnt sound kind to your bottom line, or any argument for Intel having a better price:performance. The fact is that AMD has continued to support the G34 platform since 2010. I cant say that Intel can boast about the same support for their sockets. That is what I meant when I wrote about extended board support. Also, a quick google search indicates that you can get a 6380 based server for the price of a Xeon E5-2640 based server. The Opteron machines need more energy to do their job, but once again you get better performance per dollar than Intel's midrange offerings

1. No servers big enough to actually upgrade every 2-3 years or less would never use 1155.

 

2. LGA2011 will have a lifetime of 2 compared to AMD's 2 years (they have only had 2 years in which they released CPUs. The other year was pretty much nothing.) The lifetime argument does not stand.

 

3. E5 2660v2 and Opteron 6386SE both cost $1400 and are Intels upper midrange lower high end Sku and AMDs highest end Sku.

[Giggitygoebbels]

Guess Intel will be giving us even smaller increases in CPU now because they'll basically own the market now...

Which also means,all the computers in my house excluding my iPad and Surface,is going to be more future-proof?

Yes, but to sacrifice 40% die for igpu... and you have to pay it... and you dont use it because you are gamer and have a GPU

Intel does make cpus, they go by the name : Xeon

AMD will still be making CPUs, They go by the name: Opteron

You 2 forgot the LGA2011 i7s.

Now Intel can do the pricing how they want.

They could sell now every CPU for $100 more and there is nothing we can do about it :(

I must be lucky that I bought my 4930k early...or maybe not.

And yet, when Microsoft tries to open doors with ARM processors support with Windows RT, people bash Microsoft for it's effort.

Well I guess now we are stuck with a Intel monopoly. I can't wait to spend 500-700$ for a mid range Intel CPU.

Congrats people.

So,my 4930K is a mid range Intel CPU?No It is low end from Intel's perspective.8 x Xeon E7-8890v2 is high end.Therefore to push you into the Xeon E5 and E7 market,Core i series and Xeon E3 series will have a price raise.Definitely.

How about 6-core?

This is absolute BS. And to think my next build was going to be AMD parts, I'm not sacrificing for a damn APU. Like GoodBytes said, Intel i5 and i7 available for $500 and $700 respectively!

In BGA format,soldered on high end motherboards only,and the total costs 1000-1500$ at least.

AMD must have given up with Intel... just screw them and move on...

No,this is not a DFI and Lanparty story...AMD is not going to fade away like that...no way...

[orangecat]

I'm not trying to sound like an intel fanboy but good. AMD has been suffering when it comes to the high end CPUs for ages and their APUs destroy anything from Intel. its hard to say that a APUs graphics are anything good especially for gaming but they are 100x faster then HD graphics from Intel. I don't like AMDs CPUs anyways they always ran a bit hot, consumed a bunch of power and where always slower than the intel equivalent.

[Giggitygoebbels]

So if you are an IT company, you are willing to throw out last gens boards, and CPUs for the latest Ivy update and throw some more cash at a new 1155 board (assuming you are greater than a generation old)? That doesnt sound kind to your bottom line, or any argument for Intel having a better price:performance. The fact is that AMD has continued to support the G34 platform since 2010. I cant say that Intel can boast about the same support for their sockets. That is what I meant when I wrote about extended board support. Also, a quick google search indicates that you can get a 6380 based server for the price of a Xeon E5-2640 based server. The Opteron machines need more energy to do their job, but once again you get better performance per dollar than Intel's midrange offerings

 

E5 2640 or 2640v2?The difference is,a 2640 is a higher clock speed 6 core 12 thread,while the v2 is a lower clock speed 8 core 16 thread.It does make a fair bit of difference.

 

Yup. Will be interesting to see who catches the other first.

 

 

1. No servers big enough to actually upgrade every 2-3 years or less would never use 1155.

 

2. LGA2011 will have a lifetime of 2 compared to AMD's 2 years (they have only had 2 years in which they released CPUs. The other year was pretty much nothing.) The lifetime argument does not stand.

 

3. E5 2660v2 and Opteron 6386SE both cost $1400 and are Intels upper midrange lower high end Sku and AMDs highest end Sku.

1.Agree

2.Well so Opterons are no good Value.

3.The 2660v2 obviously wins easily.