AMD CPU speculation... and expert conjecture

[juanrga]

Spoiler

That is funny because I am the only who has posted both absolute values of FP and percentages. Also I already explained you that intermediate calculations depend of the media used to analyze the slide.

did you post them? could you re-post them, please? i'll address your 'explanation' after you post all of your findings.

Or you can spend your precious time searching the post where I wrote the numbers in a 2x3 'table' instead asking others to search for you.

i did search, and could not find any. i could say you never posted them but i know about tom's forum search function, it tends to skip relevant results some times.
and i did not ask anyone else to search, i asked You. You, who posted them in the first place, and in possession of the method of calculations as well as the results. please post those.

You insulted others calling them liars.

don't make this about others. i accused you of lying.
and.. to make this crystal clear, i didn't call anyone else liar. at least, to my knowledge, not without reason. like i said, don't make this about others. oh, you still haven't posted the data. i hope you stop avoiding.

Yes. The FX-6350 is ~2% faster and the FX-8350 is ~4% faster in that slide.

see, you post only the end results, not the calculations or intermediate results. those would be really helpful.

^^^ Answered before and again and again and again.

edit:
i was searching the site (that juanrga linked) for more kaveri info and came across their a10 7800k table. it's a summary for their findings(rumors):

APU [A10-6800K] [A10-7800K (speculated model number)]
Production Process [32nm PD-SOI] [BULK 28nm Gate First]
Core [Richland] [Steamroller B]
Frequency (Turbo) [4.1 GHz (4.4 GHz)] [2.9 GHz (3.2 GHz)]
iGPU [VLIW4] [GCN 2.0]
Stream Processor [384] [512]
Frequency iGPU [ 844 MHz ] [1 - 1.2 GHz]

from what i know about substrates, the choice of silicon and clockrates don't seem very encouraging..... hopefully the real deal would be better. i mean.. intel's cpus are made on bulk and they can hit up to 3.9 ghz turbo at stock. it'll also depend on how improved kaveri's (in turn steamroller's) turbo controller and in-built thermal sensors turn out to be. especially in laptops where cooling would be restricted. the igpu clockrate looks good. imo, the igpu looks very powerful even from the speculated specs. 512 gcn 2.0 cores at 1 ghz base clock is.... monsterous for an igpu. as long as amd feeds this monster well, it'll deliver.

one more thing, hawaii is not gcn 2.0. kaveri is gearing up to jump ahead of amd's existing gpu lineup by a generation. may be temporarily though, since 20nm gpus are very likely to be gcn 2.0.

The frequencies don't make many sense. The same author already admit that all the second wave ES are clocked at 3.5GHz. I continue expecting the final silicon to achieve 3.8GHz minimum (I hear rumors Kaveri samples achieving 4.5GHz with adequate cooling).

 

[tracker45]

Could they squeeze a 7790 on an APU ? Or is that asking for too much ?

 

[kansasboy001]

I think that AMD replacing the fx line with apus is great, If you think about all of the LGA 1155/1150 i7 processors are APUs, now IF amd can make an apu that can compete in this market segment, as the fx 8xxx series was originally intended for, then that would be great. In order to do so they need an apu with either 8 cores or 4 cores and 8 threads. I don't think that this is too far out. In fact I have thought ever sence I joined this thread that the steamroller for the enthusiaest market would be a A-10 on steroids.

Not only it is great the idea of AMD replacing FX line with APUs(*), but matches with what AMD is doing in servers where the Opteron 3330 CPUs (4 and 8 PD cores) are replaced by 4 SR core Berlin APU (*).

I would like to see 8-core APU, but there is none in the near horizon.

(*) And the CPU variant.

An 8 core APU is probably not possible at 28nm right now, but AMD still could have a hyper threading technology for the enthusiast APUs.
I think that the new branding for the FX line will be as follows:

4xxx series: Athlon
6xxx series: A-10
8xxx series: Phenom
A-4 will still remain low power
A-8 will be between Athlon and A-10

Remember the Phenom IV rumor, hence the Phenom Branding

Amd recycled the Athlon name for the low end, so I would only wake sense to have the Phenom back as the high end.

I still think that there will be an 8 core or 4 with some kind of hyper threading for the A-12 or Phenom line.

Remember Slide 13 said MORE performance with MORE cores.

An 8 core APU in a single die seems difficult. It would have a minimum size of about 480 mm2 @ 28nm and that is a big die. But maybe a MCM version by combining two 4 core dies could be made. The problem here is that would be probably too expensive considering the demand.

I think there are two realistic scenarios: Either (i) AMD replaces the entire FX line with APUs or (ii) replaces the FX-4000 with APUs and releases a refresh of the FX-6000/8000/9000 series using the same PD modules used in Warsaw CPU.

If Kaveri is a 20% faster than Piledriver. I think the option (ii) rules until 2015 when Carrizo (excavator APU) is released and then the FX line is abandoned definitively.

If Kaveri is a 45% faster than Piledriver. I think the option (i) rules. The chinese site that I mentioned above says that Kaveri will compete with Haswell i7-4770k.

I agree, unless they made the igpu really tiny and weak, only then could they have an 8 core apu.

According to what AMD told OEMs there will be no more FX branded processors, the 9590 was the last. There will be a new line of enthusiast processors released tomorrow.
Remember the Phenom IV Beca rumor. Companies sometimes leak partly true pieces of information. Return of the Phenom name along with Athlon, makes scene to me.

 

[m32]

How big will an 8-core APU (28nm) be with no iGPU & no L3?

I guess we could call it a CPU if I had my way

 

[gman97005]

Any info on the Crosshair VI Formula?

None so far, my friend just got done ordering his CHFV-Z and 8320 for his new build. You need to get a 290X my friend 😛

why would I need a 290x, my 2 6970s out perform one 290x.

Proof

For the moment, AMD's Kaveri mixed with a lowly R7 260x will outperform both of them..

 

[kansasboy001]

How big will an 8-core APU (28nm) be with no iGPU & no L3?

I guess we could call it a CPU if I had my way

I am talking about a less powerfull iGPU, taking up 25% of the chip, not 50% to allow for more cores. Yes the graphics would be crappy like intel sandy bridge, but it would still qualify as an APU.

 

[palladin9479]

How big will an 8-core APU (28nm) be with no iGPU & no L3?

I guess we could call it a CPU if I had my way

Actually it would be very small. Modern CPU's use about 50% of the die space for L3, sometimes more. AMD replaced the L3 with an iGPU for it's current crop of APU's while making them smaller then the PD CPU's.

It's entirely feasible to couple a 3~4M CPU with a cut down iGPU, something like 4~8CU and some L3.

 

[de5_Roy]

^^^ Answered before and again and again and again.

you did not answer anything at all. you keep on avoiding posting the calculations, the results and the method you obtained them. you posted zero amount of hard data that you claim to possess. a truthful and honest person would stand by his/her own claims and will properly present his/her side with hard data. s/he sure as heck would not dodge for multiple days. that makes me think that something is really amiss with your claims. please post those data a.s.a.p.

Spoiler

What Steamroller should be :

A cpu as good as an i5 2500k.
A cpu as good as an i5 3570k

An APU with weak GPU
An APU with Medium GPU
An APU with High end GPU


That's all we need from AMD.

you should add the cost of r&d, marketing and end price. if amd has a >core i5 2500k class cpu, they will sell it at $220, no less than that. at best, they'd launch it at $260-270 and drop it to $190-200 mid cycle. they thought zambezi was core i5-class and they had low early yields (danke glofo!). they positioned those cpus between lga 1155 core i5 and core i7 cpus.

@8 core apus and igpu: amd already has two apus for the consoles that have eight x86 cores and igpu. amd's main goal (and intel's as well) is to build low power socs. apus are like a few components short of an soc... but they can be called socs as well. apu is just a marketing term imo. here's an image from the ps4 teardown:
http://www.xbitlabs.com/news/graphics/display/20131110143543_PS4_Opens_Wide_Shut_Most_Awaited_Game_Console_Opens_Up_Its_Internals.html
at the middle-right side of the green pcb, see that big bga chip surrounded by 8 smaller chips? that's the apu. the pcb has 8 more gddr5 chips soldered on the underside of the pcb. the whole thing is cooled by a blower type fan.
the ps4 igpu(!) has 18 gcn compute units(1152 cores), 512bit imc connected to 8GB gddr5 memory @176 GB/s - manufactured on tsmc mobile node(bulk).
amd could, in theory, shed half of the shaders, imc bus and use the rest of the die for either more jaguar cores, of possibly put SR cores (i dream!).
edit: if they want to put SR on that cut down speculated apu, they'd have to port the design to glofo's high performance node first... or build it from ground up.

 

[juanrga]

I think that AMD replacing the fx line with apus is great, If you think about all of the LGA 1155/1150 i7 processors are APUs, now IF amd can make an apu that can compete in this market segment, as the fx 8xxx series was originally intended for, then that would be great. In order to do so they need an apu with either 8 cores or 4 cores and 8 threads. I don't think that this is too far out. In fact I have thought ever sence I joined this thread that the steamroller for the enthusiaest market would be a A-10 on steroids.

Not only it is great the idea of AMD replacing FX line with APUs(*), but matches with what AMD is doing in servers where the Opteron 3330 CPUs (4 and 8 PD cores) are replaced by 4 SR core Berlin APU (*).

I would like to see 8-core APU, but there is none in the near horizon.

(*) And the CPU variant.

An 8 core APU is probably not possible at 28nm right now, but AMD still could have a hyper threading technology for the enthusiast APUs.
I think that the new branding for the FX line will be as follows:

4xxx series: Athlon
6xxx series: A-10
8xxx series: Phenom
A-4 will still remain low power
A-8 will be between Athlon and A-10

Remember the Phenom IV rumor, hence the Phenom Branding

Amd recycled the Athlon name for the low end, so I would only wake sense to have the Phenom back as the high end.

I still think that there will be an 8 core or 4 with some kind of hyper threading for the A-12 or Phenom line.

Remember Slide 13 said MORE performance with MORE cores.

An 8 core APU in a single die seems difficult. It would have a minimum size of about 480 mm2 @ 28nm and that is a big die. But maybe a MCM version by combining two 4 core dies could be made. The problem here is that would be probably too expensive considering the demand.

I think there are two realistic scenarios: Either (i) AMD replaces the entire FX line with APUs or (ii) replaces the FX-4000 with APUs and releases a refresh of the FX-6000/8000/9000 series using the same PD modules used in Warsaw CPU.

If Kaveri is a 20% faster than Piledriver. I think the option (ii) rules until 2015 when Carrizo (excavator APU) is released and then the FX line is abandoned definitively.

If Kaveri is a 45% faster than Piledriver. I think the option (i) rules. The chinese site that I mentioned above says that Kaveri will compete with Haswell i7-4770k.

I agree, unless they made the igpu really tiny and weak, only then could they have an 8 core apu.

According to what AMD told OEMs there will be no more FX branded processors, the 9590 was the last. There will be a new line of enthusiast processors released tomorrow.
Remember the Phenom IV Beca rumor. Companies sometimes leak partly true pieces of information. Return of the Phenom name along with Athlon, makes scene to me.

Yes a 8-core with small GPU could fit in acepatable die space, but that configuration makes no sense. First, it goes against the natural market tendency. Both Intel and AMD (adds Nvidia with Tegra SOCs as well) are devoting more and more die space to igpu. Look at Intel Haswell. And Broadwell promises about a 40% more iGPU than Haswell!

Second, if AMD reduce the iGPU, it would be favoring Intel!!! AMD current advantage over Intel is in the GPU side.

Regarding the scenario (ii), it is entirely possible that AMD abandons the FX brand. In the server space the Opteron 6000/4000 are replaced by new brand Warsaw. AMD could use a new name for desktop as well. It makes sense.

 

[juanrga]

^^^ Answered before and again and again and again.

you did not answer anything at all. you keep on avoiding posting the calculations, the results and the method you obtained them. you posted zero amount of hard data that you claim to possess. a truthful and honest person would stand by his/her own claims and will properly present his/her side with hard data. s/he sure as heck would not dodge for multiple days. that makes me think that something is really amiss with your claims. please post those data a.s.a.p.

All the relevant data was given.

You are a curious individual, you cannot do elementary tasks such as reading posts in the thread or obtain FPS from a bar chart given to you, but at the same time you spend on continuously insulting to the person who you solicit assistance. A very smart move from your part. Continue that way.

@8 core apus and igpu: amd already has two apus for the consoles that have eight x86 cores and igpu. amd's main goal (and intel's as well) is to build low power socs. apus are like a few components short of an soc... but they can be called socs as well. apu is just a marketing term imo. here's an image from the ps4 teardown:
http://www.xbitlabs.com/news/graphics/display/20131110143543_PS4_Opens_Wide_Shut_Most_Awaited_Game_Console_Opens_Up_Its_Internals.html
at the middle-right side of the green pcb, see that big bga chip surrounded by 8 smaller chips? that's the apu. the pcb has 8 more gddr5 chips soldered on the underside of the pcb. the whole thing is cooled by a blower type fan.
the ps4 igpu(!) has 18 gcn compute units(1152 cores), 512bit imc connected to 8GB gddr5 memory @176 GB/s - manufactured on tsmc mobile node(bulk).
amd could, in theory, shed half of the shaders, imc bus and use the rest of the die for either more jaguar cores, of possibly put SR cores (i dream!).
edit: if they want to put SR on that cut down speculated apu, they'd have to port the design to glofo's high performance node first... or build it from ground up.

We know that AMD makes 8-core APUs for consoles. This has been discussed lots of time here (do you read the thread?). We are discussing the possibility of a 8-core APU using big cores. I.e. we are discussing a 8-core Kaveri or Carrizo APU.

The console APUs are balanced. The 8-core config. was chosen by game developers. Adding more than 8-core jaguars and cutting the gpu makes little sense.

Cutting half the shaders and replacing the tiny jaguar cores by big SR cores is what Kaveri is, basically.

 

[de5_Roy]

All the relevant data was given.

no. you only said you have the data, but never presented them nor how you obtained them in detail. if you had presented them, why not post it again? why so avoidant?

You are a curious individual, you cannot do elementary tasks such as reading posts in the thread or obtain FPS from a bar chart given to you, but at the same time you spend on continuously insulting to the person who you solicit assistance. A very smart move from your part. Continue that way.

i thought we were over my inability of reading graphs and other elementary tasks. that's why requested for Your data. moreover, me calculating is not enough, i have to verify and compare with yours. so.. either way, your calculations, methods and results are very important.
i am sorry that you feel insulted, but my accusations are based only on your actions. unfortunately, you have not presented anything to refute me or prove your claims.

Cutting half the shaders and replacing the tiny jaguar cores by big SR cores is what Kaveri is, basically.

hehe, not exactly. half of ps4 soc's memory bus would be 256-bit, while kavery has 128-bit dual channel bus. the apu i speculate would have 2x as much bw, thus feeding the 576 core igpu(more than kaveri's) very well-fed.

 

[juanrga]

I speculated before about why AMD would be moving to FINFETs. An 'enthusiast' tried to convince me of the supposed advantages of FD-SOI and why AMD would be moving to SOI. This is the same person that said that Kaveri was being delayed because was being made in SOI at Glofo. Well, those are the words of AMD vicepresident:

We are typically at the leading edge across the technology nodes. We are fully top-top-bottom in 28nm now across all of our products, and we are transitioning to both 20nm and to FinFETs over the next couple of quarters in terms of designs. So we will continue to do that across our foundry partners. […] We will do 20nm first and then we will go to FinFETs

No mention of SOI.

I think that first 20nm bulk products will be the next gen dGPUs (Pirate Islands series?).

FinFETs could be used for Carrizo APU, maybe.

 

[gamerk316]

So, we coming around to my prediction of 15% improvements per clock yet?

Regardless, I'm sure Juanrga will cling to one GPU-assisted benchmark that shows APU > FX. So we're just going to continue to go around in circles.

 

[Rum]

Today is the day that we have all been waiting for (APU summit starts), hopefully all of our questions will be answered about the APU and HEDT side of things!

 

[tracker45]

My predictions.



What Steamroller should be :

A cpu as good as an i5 2500k.
A cpu as good as an i5 3570k

An APU with weak GPU
An APU with Medium GPU
An APU with High end GPU


That's all we need from AMD.

 

[juanrga]

Spoiler

All the relevant data was given.

no. you only said you have the data, but never presented them nor how you obtained them in detail. if you had presented them, why not post it again? why so avoidant?

You are a curious individual, you cannot do elementary tasks such as reading posts in the thread or obtain FPS from a bar chart given to you, but at the same time you spend on continuously insulting to the person who you solicit assistance. A very smart move from your part. Continue that way.

i thought we were over my inability of reading graphs and other elementary tasks. that's why requested for Your data. moreover, me calculating is not enough, i have to verify and compare with yours. so.. either way, your calculations, methods and results are very important.
i am sorry that you feel insulted, but my accusations are based only on your actions. unfortunately, you have not presented anything to refute me or prove your claims.

You would be, because when you decided to start insulting me I decided not to help you

I wrote before both the absolute FPs and percentages. You cannot obtain them by yourself? Well, I am so sorry.

Cutting half the shaders and replacing the tiny jaguar cores by big SR cores is what Kaveri is, basically.

hehe, not exactly. half of ps4 soc's memory bus would be 256-bit, while kavery has 128-bit dual channel bus. the apu i speculate would have 2x as much bw, thus feeding the 576 core igpu(more than kaveri's) very well-fed.

basically != exactly. Of course there are differences between the PS4 APU and Kaveri APU

 

[tracker45]

Is the steamroller/kaveri conference being broadcast online ?

 

[de5_Roy]

You would be, because when you decided to start insulting me I decided not to help you

the only one who's not being helped is you. you not providing counter-argument with solid evidence only hurts your own cause. on top of that, you are being so avoidant for such a long duration (for multiple days) days that it makes me doubt your argument's validity even further. i can't really help you with your personal feelings.

I wrote before both the absolute FPs and percentages. You cannot obtain them by yourself? Well, I am so sorry.

no you did not. i can say for sure that you never posted those data after you brought your claims up in a post replied to one of mine. the percentage values are worthless without the procedure and intermediate results. on their own, they're just arbitrary numbers you seemingly made up.

Is the steamroller/kaveri conference being broadcast online ?

this seems to be the website
http://developer.amd.com/apu/
and amd's youtube channel
https://www.youtube.com/amd
keynotes to be presented:
http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-eventDetails&EventId=5051218

 

[juanrga]

So, we coming around to my prediction of 15% improvements per clock yet?

Regardless, I'm sure Juanrga will cling to one GPU-assisted benchmark that shows APU > FX. So we're just going to continue to go around in circles.

So you missed the leaked benchmarks showing >30% improvement per clock?

There is no doubt that APU will be faster than some FX chips in a pure CPU benchmark, but of course, the real strength of a APU is in heterogeneous compute. Why would I use an archaic piece of software that uses only a 20% of the performance of the APU?

And another of your predictions has been smashed. I mean your "nobody will use MANTLE". After Frostbyte and Activision engine, now it will be confirmed at APU13 that Unreal engine 3 will use MANTLE.

For several years now we have been working with AMD on exciting game titles like Deus Ex, Hitman, Tomb Raider, and we are currently working on the latest version of Thief, to bring the best possible experience to AMD hardware. In this presentation we will discuss some great AMD technologies that we are using in Thief. Most importantly, we will talk about Mantle and why it is something we have been waiting for for a long time. We are currently implementing Mantle inside of Unreal Engine 3 and will share some of our experiences. We will also cover additional AMD technologies we are supporting in Thief, like AMD TrueAudio, 4K Gaming, as well how are exploiting the power of the compute hardware in the GCN architecture. While we will primarily talk about our high-level view on these technologies and why we are excited about them, we will also share some helpful implementation tips from our programmers.

Other talks about MANTLE:

The small batch problem has been a serious challenge for PC game developers for over a decade. Even with major CPU performance improvements over that time, the number of draws that can be processed per frame hasn't substantially increased. Coupled with the inability to control low-level aspects of the PC graphics subsystem, the small batch problem has hamstrung the quality of games and the overall gaming experience; until now. Welcome Mantle. Mantle is the next step in the evolution of 3D APIs, enabling the development of highly efficient and content rich 3D games and applications. Imagine the possibilities if many conventional API restrictions are lifted & you're in charge of the GPU! In this talk we will present our design methodology and key features of the Mantle API. Learn how you can be a part of the 3D graphics innovation and unlock the power of your GPUs.

Mantle Unleashed: How Mantle changes the fundamentals of what is possible on a PC. Over the last 5 years, GPUs have become so fast that it has become increasingly difficult for the CPU to utilize them. Developers expend considerable effort reducing CPU overhead and often are forced to make compromises to fully utilize the GPU. This talk will discuss real-world results on how Mantle enables game engines to fully and efficiently utilize all the cores on the CPU, and how it's efficient architecture can eliminate the problems of being CPU bound once and for all.

Crytek will be also giving a talk about the new Cryengine 4. No sure if they will announce MANTLE support.

 

[blackkstar]

As for this marketing slide comparing APU to CPU, the point is irrelevant.

1. It's a marketing slide to promote APUs, of course they're going to be using a situation that favors APUs for their benchmark

2. If 4 SR cores are fast enough to compete with FX 6000 PD models then it's foolish to think that that means the APUs are going to replace CPUs. A CPU only SR chip would also increase in performance by similar ratios that the APU did using SR.

I just find it very hard to swallow that AMD is going to become a $650 GPU company and a $150 CPU company.

But the whole point of Mantle is to further destroy Nvidia from the bottom up, as it'll make lower end APUs comparable to higher end Nvidia parts. Realistically I would think that high end APU with Mantle would be competing with GTX 660 Ti and lower at least, but that's just a random guess.

Are they purposely putting weak GPUs in APUs to prevent low GPU sales ?

No, you have it backwards. You used to be able to buy dGPUs like 5450 and Nvidia had GPUs like GT 420.

Also, AMD wants to push APUs as hard as possible. Someone buying an APU on FM2(+) platform is buying an HSA compatible system. Someone buying an FX 4000 series with a (if it existed) 7450 would not be buying an HSA system.

All Nvidia has left right now are high end GPUs and workstation cards.

Tegra is a disaster that doesn't turn profits and doesn't get design wins. Things are so bad for Nvidia that they have to release their own hardware. This is like AMD releasing Temash, absolutely no one using it, and then AMD making their own tablets and laptops.

The low end is obliterated by Intel AND AMD offerings and Nvidia no longer stands a chance there because they feel like they should charge more just because it's Nvidia.

Nvidia knows this and they are quite scared. It's why suddenly there's a rumor that Nvidia paid off OriginPC to ditch AMD and then OriginPC gets new GPUs first, and then OriginPC gets a new GPU first.

Now, you can see sites like Anandtech write off GTX 480 (which was louder than 290x, mind you) as a great card while the quieter 290x is too loud and not worth the price premium. That's because people are getting paid that shouldn't be paid. Anandtech is probably the worst, I recall people claiming that Anandtech ran 290x in quiet mode the entire time just so it looked slower.

I'm waiting to hear some sort of news at APU13 about dGPU HSA. If it's even remotely hinted at that it's coming then we know almost for sure that there's going to be a dCPU and dGPU HSA platform and that it'll be coming soon.

By not having an HEDT platform that is HSA compatible, AMD is turning these GPU sales into just GPU sales instead of possible sales of HSA systems. At the least they should get something out and give people the option of going AMD CPU + AMD GPU for an HSA system.

I do think it's one of AMD's largest problems right now. If they go APU only and APUs can only serve until mid-range gaming machines, it's going to really hurt adoption of HSA as power users are more than likely going to be the ones buying HEDT and they're usually the power users who try new software and jump on things like CUDA.

Basically, it's like this. AMD going APU only relegates HSA to a type of "fast budget" segment and it will still leave Intel to the high end.

I really, really don't see that happening. I still feel that AMD is pushing APU hard right now because that's the easiest way to get a good HSA capable system install base, but leaving HEDT out of HSA completely seems like a massive mistake.

Let me ask you this, companies like Adobe, Autodesk, etc. Do you see them spending money on making HSA versions of their applications so only people on low end APU systems can use them? Or do you see them only wanting to add these types of features if the users are on a sort of 16 core Opteron system with FirePros?

How many companies do you think are going to invest in HSA software if it means isolating any sort of professional workstation?

And yes, I know there are FirePro APUs, but guess what? A 2m/4c APU is going to be horrible for traditional CPU workloads compared to a high end CPU from Intel or even AMD. Imagine you are working at a game studio, you have to render your model and the physics in the scene are HSA accelerated but the renderer is not (it hasn't shifted to HSA yet). So you wait on the CPU.

Is that jump in physics performance worth it? No, you'd just spend the money on a big traditional CPU and not bother about maintaining all this new fancy HSA software.

That's my two cents on the AMD APU only thing. It'd stop a TON of software from ever getting HSA support as adding HSA support to anything beyond games (and even games are a good example, who is going to add HSA features to games only for mid-range users?).

I want to hear from juranga (and others who agree with him) how AMD would plan to drive HSA adoption and to promote software developers of professional applications as well as games when the best they offer is a mid-range part.

 

[tracker45]

If an APU had equivalent performance of a gtx 660, that would damage sales for Nvidia AND Amd.

Basically the mid range GPU market would be reduced.

 

[gamerk316]

There is no doubt that APU will be faster than some FX chips in a pure CPU benchmark, but of course, the real strength of a APU is in heterogeneous compute. Why would I use an archaic piece of software that uses only a 20% of the performance of the APU?

Because a single compute unit is slow, and if the software doesn't scale close to linearly, you could easily gain -500% performance this way. And yes, I've seen it happen.

What you are trying to do is compare a CPU to a CPU+GPU. The fact both are on the same die, from a benchmarking perspective, is irrelevant. Give the 8350 even a cheap $50 GPU, and the 8350 pulls ahead again. Farther, that argument means you also need to compare with a i7-3770k and its iGPU, which while not powerful, will pull it ahead when compute tasks are considered.

So please, stop it. For purely CPU workloads, the APU is going to be sub-i3 performance.

 

[gamerk316]

If an APU had equivalent performance of a gtx 660, that would damage sales for Nvidia AND Amd.

Basically the mid range GPU market would be reduced.

Simple solution for that problem: Spin off ATI.

I'm not joking either; spin them off, collect a few hundred million, then run them out of business with cheap APU's, gaining a majority of the GPU market.

Right now, AMD's stuck with two separate units which are starting to conflict with eachother. One of them needs to go away for the company to be successful, otherwise all they are doing is competing with themselves.

 

[tracker45]

Mantle may increase the APU performance.

And we still don't know what GPUs are going into the new Steamroller APUs.

If a gtx 660 performance could be gotten out of an APU, that would be incredible.

 

[de5_Roy]

If an APU had equivalent performance of a gtx 660, that would damage sales for Nvidia AND Amd.
Basically the mid range GPU market would be reduced.

an apu will have gtx660 level perf when that sort of perf is available at entry level price. amd will have to empower their apus before that. until then, it won't happen.

So please, stop it. For purely CPU workloads, the APU is going to be sub-i3 performance.

only if 2M apus are considered. i think by the time excavator launches, amd will have enough die area to add more of cpus while keeping igpu area-budget intact.

Simple solution for that problem: Spin off ATI.

I'm not joking either; spin them off, collect a few hundred million, then run them out of business with cheap APU's, gaining a majority of the GPU market.

Right now, AMD's stuck with two separate units which are starting to conflict with eachother. One of them needs to go away for the company to be successful, otherwise all they are doing is competing with themselves.

spinning off ati will insta-kill amd. their gpus are what's been keeping them afloat. console deals, hsa - all dependent on the gpu. amd cannibalizing their entry level gpus with apus only help amd. they can convert cpu+gpu owners to apu (s.o.c.) owners for less cost as well as sell a full pc ecosystem.