AMD CPU speculation... and expert conjecture

[Ags1]

All this talk of market share seems to neglect the fact that expensive stuff gets made for small niches - like the R9 290X or the Titan. The market for 6 and 8 core steamroller is much larger.

Here in Holland AMDs 6 and 8 core products trade from 90 to 170 euros - these are not high end prices, these are mass-market prices, competing with Intel bread-and-butter i3s and i5s.

 

[tracker45]

What are the chances they make a mess of it and the CPUs are not very good ?

We can't have another Bulldozer.

 

[The Q6660 Inside]

I would of never buy a new GPU if i knew the new ones where coming out soon unless i needed one that instant

Then you'd never be buying anything as there is always something "coming out soon". For a long time my dual GTX 580 Hydro's served they're purpose. Recently they stopped being sufficient for what I do so I've decided to replace them. And since I'll be draining and redoing my loop now is a good time to upgrade the entire kit. These cards should last me another three or so years before being replaced. Sometime in the next year or so I will be swapping out my CPU, might even go Intel if AMD doesn't have a product to replace my FX8350.

I will buy that 800D for $100 😛 In 900D range, you might as well go caselabs, I can haz pics?
@tracker, you cannot possibly be worse, the FX line still makes good competition with Haswell, so they cannot possibly screw up and make it bad, even a 15% increase in single-threaded will give Intel a run for its money. Think of Bulldozer as Agena, and Piledriver as Deneb. Steamroller will be Thuban and hopefully, we won't make a mistake like Agena after that. Bulldozer was a good idea, it was just too ahead of its time...

 

[8350rocks]

Returning to the origin of this discussion: AMD is not releasing FX 8-core Steamroller because there is no enough market for that.

I seriously think this part is entirely rubbish. There is most certainly a market for it, and it is pretty substantial; however, they're losing a great deal of that business to the 4770k unless they act sooner than later.

I suppose you would argue there is no market for the 4770k then? 4c/8t chips are the same thing as 8c/8t chips in the grander scheme of things.

The 8 core AMD chips are aiming at the 4 core market...last time I checked, that market share had actually passed 50% of the PC market in the US. Considering every house has at least one PC, that means there are at minimum 175 million quad core PCs in the US alone. Think they couldn't take some of the 4 core i7 business with a hot 8 core that could keep up at single thread workloads? I know they would seriously compete...

You cannot translate market share from a company to another. A company that is economically healthy can afford niche markets, whereas a company who was in the red some months ago cannot.

i7-4770k and FX-8350 are not mainstream. Stop from pretending that 50% of the market would be interested in a FD-SOI 5GHz 8-core FX Steamroller.

Well, your issue is pretending that the 4770k and the 4770 and the 8 core FX series are priced the same, they are not.

You can buy that 8 core monster for less than a 4570, which happens to be a *VERY* mainstream part.

Your logic is failing in comparing rough equals in performance as being equals in price. Anyone looking at buying a quad core CPU from an OEM of any sort would see the 4570 on a list of CPUs that could be had in a *VERY* affordable PC. If you could get more raw computing power for less money, why would you not do it?

 

[8350rocks]

What are the chances they make a mess of it and the CPUs are not very good ?

We can't have another Bulldozer.

There will not be another bulldozer...not while Jim Keller is running the CPU show at AMD.

 

[GOM3RPLY3R]

Sorry to budge in, but I was looking for some GTX 770 DCuII references for my school project when I saw a R9-290X review.

Let's start with the information you want most, cost. The official AMD MSRP on the AMD Radeon R9 290X will be $549. This places it $100 below the cost of the GTX 780, which is $649.

Is this true? I'm trying to wrap my head around the fact that this card is better than the Titan, making the price to performance over twice as better than the Titan.

It just seems way too good to be true. I hate to admit it, and even though I HATE AMD drivers, I'm probably just going to get 3 of those.

 

[Cazalan]

What are the chances they make a mess of it and the CPUs are not very good ?

We can't have another Bulldozer.

Switching to the bulk process will likely not allow the high overclocks currently achieved by the SOI process. However it should be cheaper to make.

 

[Cazalan]

Is this true? I'm trying to wrap my head around the fact that this card is better than the Titan, making the price to performance over twice as better than the Titan.

It just seems way too good to be true. I hate to admit it, and even though I HATE AMD drivers, I'm probably just going to get 3 of those.

It is impressive performance for the price. Same as what the 7970 launched at. However I'm waiting for next years 20nm graphics cards.

 

[-Fran-]

It's not that simple. FX8 parts do have 8 cores, they're just not as strong as Intel cores because it's 2 ALU vs 3 ALU. That gives Intel cores an advantage.

In retrospect AMD could have made the "module" a single 4 ALU core. That would have been better for traditional PC workloads.

I am just messing with semantics (a.k.a semantic troll, haha) and since I don't want to go all over that discussion again, I won't argue about it "not being an 8 core".

Anyhow, there is a market, always.

The problem with AMD's "catch" (MOAR CORES) doesn't interest average joe for the simple reason of Software being simple.

I put somewhere (an article I think) that the real problem with PCs (lappies and desktop computers, on whatever uarch) is that there's really no hardcore software using anything under the hood, except for games.

I remember AMD trying to leverage their APUs not so long ago with "image processing", but never saw NOTHING come out of that.

Remember the "hello computer" Star Trek joke? I'm going that route with this point.

Y NO U HELLO COMPUTER?!

It makes me sad, really.

Cheers!

PS: http://www.youtube.com/watch?v=v9kTVZiJ3Uc

 

[jdwii]

Noise levels are a big problem for gaming rigs. Nothing ruins your experience like hearing the high whining sound of your GPU's. It's the entire reason I went WC, just got tired of hearing my GTX 285's while I was playing something. People who are used to it won't realize how much it's hurting them until they play on a nice rig that doesn't make sound.

Lets see what ASUS ROG does with the card.

Next Gen Graphics and Process Migration: 20 nm and Beyond
http://www.pcper.com/reviews/Editorial/Next-Gen-Graphics-and-Process-Migration-20-nm-and-Beyond

Sorry to budge in, but I was looking for some GTX 770 DCuII references for my school project when I saw a R9-290X review.

Let's start with the information you want most, cost. The official AMD MSRP on the AMD Radeon R9 290X will be $549. This places it $100 below the cost of the GTX 780, which is $649.

Is this true? I'm trying to wrap my head around the fact that this card is better than the Titan, making the price to performance over twice as better than the Titan.

It just seems way too good to be true. I hate to admit it, and even though I HATE AMD drivers, I'm probably just going to get 3 of those.

Yes its true it beats the titan more than it loses and its 550$ vs 1000$ but wait for aftermarket coolers.

 

[The Q6660 Inside]

Sorry to budge in, but I was looking for some GTX 770 DCuII references for my school project when I saw a R9-290X review.

Let's start with the information you want most, cost. The official AMD MSRP on the AMD Radeon R9 290X will be $549. This places it $100 below the cost of the GTX 780, which is $649.

Is this true? I'm trying to wrap my head around the fact that this card is better than the Titan, making the price to performance over twice as better than the Titan.

It just seems way too good to be true. I hate to admit it, and even though I HATE AMD drivers, I'm probably just going to get 3 of those.

I never had an issue with the 5870 and 4870 catalyst drivers, Crossfire blew back then though. Well, a Sapphire Reference 290X is $579 with BF4 on the egg and already out of stock. The 7970 and more importantly, the 7950 did the same thing, had amazing Performance Per Dollar, roasting the old GTX 580 in the process, though in this case it is not as much. ASUS managed to do jack sh*t this time around with the 280Xs, so when the CUII and Vapor-X, Toxic, etc OCd cards come along, we should be in business.

 

[jdwii]

Wasn't the main reason Amd didn't include 3 ALU's is because they could not feed them fast enough??? doesn't the phenom design have 3 ALU's? Pretty sure the major issue with Piledriver is its shared decoder, FPU, and its high latency cache(bigger issue than you think). To my knowledge Steamroller fixes the Shard decoder issue and i think tweaks the FPU a bit more besides that minor tweaks the probably add up to 5% more performance at most so i would say 30% boost in Performance per clock is not out of question the shared decoder was a huge issue by its self(20 to even 25% performance penalty).

But now since its been confirmed that BULK will be used instead of SOI the clock speeds are questionable they do still have Clock mesh technology but that only goes so far. For example if they clock this new CPU at 2.9Ghz the performance will most likely be the same as the A10 6800K(except better graphics).

Does anyone know when exactly these new chips will be out?

 

[The Q6660 Inside]

Wasn't the main reason Amd didn't include 3 ALU's is because they could not feed them fast enough??? doesn't the phenom design have 3 ALU's? Pretty sure the major issue with Piledriver is its shared decoder, FPU, and its high latency cache(bigger issue than you think). To my knowledge Steamroller fixes the Shard decoder issue and i think tweaks the FPU a bit more besides that minor tweaks the probably add up to 5% more performance at most so i would say 30% boost in Performance per clock is not out of question the shared decoder was a huge issue by its self(20 to even 25% performance penalty).

But now since its been confirmed that BULK will be used instead of SOI the clock speeds are questionable they do still have Clock mesh technology but that only goes so far. For example if they clock this new CPU at 2.9Ghz the performance will most likely be the same as the A10 6800K(except better graphics).

Does anyone know when exactly these new chips will be out?

The APUs will be out by Very Early Q1 2014 to the consumer, and Deployed to the OEMs and sellers in December IIRC. For all we know, it still has 2 ALUs per core and yes even Agena had 3 ALU/core. Considering the 100W TDP of the new A10, I assume we could be somewhere around 3.3 with 3.6 Turbo Boost in terms of clock speed.

 

[8350rocks]

Wasn't the main reason Amd didn't include 3 ALU's is because they could not feed them fast enough??? doesn't the phenom design have 3 ALU's? Pretty sure the major issue with Piledriver is its shared decoder, FPU, and its high latency cache(bigger issue than you think). To my knowledge Steamroller fixes the Shard decoder issue and i think tweaks the FPU a bit more besides that minor tweaks the probably add up to 5% more performance at most so i would say 30% boost in Performance per clock is not out of question the shared decoder was a huge issue by its self(20 to even 25% performance penalty).

But now since its been confirmed that BULK will be used instead of SOI the clock speeds are questionable they do still have Clock mesh technology but that only goes so far. For example if they clock this new CPU at 2.9Ghz the performance will most likely be the same as the A10 6800K(except better graphics).

Does anyone know when exactly these new chips will be out?

The APUs will be out by Very Early Q1 2014 to the consumer, and Deployed to the OEMs and sellers in December IIRC. For all we know, it still has 2 ALUs per core and yes even Agena had 3 ALU/core. Considering the 100W TDP of the new A10, I assume we could be somewhere around 3.3 with 3.6 Turbo Boost in terms of clock speed.

That's F*(@#$%^ Pathetic...gain 30% performance and lose it all back in clock speed...

I seriously hope they're not that foolish...maybe that's why they held it up to Q1 2014...perhaps...we'll get similar clocks. I doubt it on bulk...but even if the turbo core on the new ones matches the base clock of the 6800k and it's ilk, then maybe we will see some performance gains.

Oh, and kiss overclock headroom goodbye...

🙁

They have to be cooking up something special for the high performance parts...

 

[-Fran-]

http://www.phoronix.com/scan.php?page=article&item=arm_odroidxu_octa&num=1

Here juan, some new ammunition for you.

Cheers! 😛

 

[8350rocks]

http://www.phoronix.com/scan.php?page=article&item=arm_odroidxu_octa&num=1

Here juan, some new ammunition for you.

Cheers! 😛

So, what you're saying is...it takes 2 quad core ARM CPUs to defeat a dual core Intel Atom...

That doesn't at all bode well for ARM.

 

[palladin9479]

Now continue the argument, link with the bold part in above quote and you almost get it.

Consider the dies that didn't pass FX6 qualifications cost $100M. Take those $100M and add the cost to use those dies for FX4 chips; this includes: fabrication (including printing FX4xxx labels on the chip), additional quality testing, packaging, storing, shipping to distributors... The cost also includes the cost of maintaining the inventory, custom taxes for international shipping, web pages for the product, and so on. Imagine the whole cost are another $100M.

Now assume during an instant that nobody want to buy FX4 series chips, if you are fabricating them you are loosing $200M and it is better if you stop at the splitting point where some dies didn't pass FX6 qualifications and put in the trash those dies.

Umm no ... just ..... no ... stop it. Your math is horribly bad.

The QA binning process happens near the end of a production line not the start. You don't know ahead of time if the die is gonna be good enough for the top product. The production cost of $100M is already spent, it's done, finished. All you need to do is set the feature flags and bolt it into it's packaging. That cost is near zero.

AMD is making profit on the FX4xxx series chips, which are just defective FX8xxx series. That means they make even more profit on the FX8 and FX6 chips. The only time they don't make a profit is if they throw the chip into the trash.

 

[8350rocks]

Now continue the argument, link with the bold part in above quote and you almost get it.

Consider the dies that didn't pass FX6 qualifications cost $100M. Take those $100M and add the cost to use those dies for FX4 chips; this includes: fabrication (including printing FX4xxx labels on the chip), additional quality testing, packaging, storing, shipping to distributors... The cost also includes the cost of maintaining the inventory, custom taxes for international shipping, web pages for the product, and so on. Imagine the whole cost are another $100M.

Now assume during an instant that nobody want to buy FX4 series chips, if you are fabricating them you are loosing $200M and it is better if you stop at the splitting point where some dies didn't pass FX6 qualifications and put in the trash those dies.

Umm no ... just ..... no ... stop it. Your math is horribly bad.

The QA binning process happens near the end of a production line not the start. You don't know ahead of time if the die is gonna be good enough for the top product. The production cost of $100M is already spent, it's done, finished. All you need to do is set the feature flags and bolt it into it's packaging. That cost is near zero.

AMD is making profit on the FX4xxx series chips, which are just defective FX8xxx series. That means they make even more profit on the FX8 and FX6 chips. The only time they don't make a profit is if they throw the chip into the trash.

Yes...

This is further evidenced by the old Phenom II series having "core unlocker" motherboards. Where some chips that could have been, say a X4 955 were deactivated and sold as a X3 or X2 because of demand. People would then buy the significantly cheaper X2 or X3 and attempt to unlock the cores, in some instances they would get no cores, some times they would get 1 extra core, and other times they would get 2 extra cores for free by "unlocking" the extra cores.

Now, sometimes you got a correctly binned chip that was supposed to be a X2 or X3 and got nothing. However, about as often as not, you got at least 1 extra core out of the ordeal.

 

[palladin9479]

Wasn't the main reason Amd didn't include 3 ALU's is because they could not feed them fast enough??? doesn't the phenom design have 3 ALU's? Pretty sure the major issue with Piledriver is its shared decoder, FPU, and its high latency cache(bigger issue than you think). To my knowledge Steamroller fixes the Shard decoder issue and i think tweaks the FPU a bit more besides that minor tweaks the probably add up to 5% more performance at most so i would say 30% boost in Performance per clock is not out of question the shared decoder was a huge issue by its self(20 to even 25% performance penalty).

But now since its been confirmed that BULK will be used instead of SOI the clock speeds are questionable they do still have Clock mesh technology but that only goes so far. For example if they clock this new CPU at 2.9Ghz the performance will most likely be the same as the A10 6800K(except better graphics).

Does anyone know when exactly these new chips will be out?

Yes and no on the ALU part. By definition you can only do one operation at a time, it's how scalar works. Super scalar uarchs have prefetch engines and branch predictors that attempt to process instructions before they are needed. Also you are decoding the large x86 instructions into several miniature micro-instructions that are then distributed amongst the various internal CPU resources. Keeping three ALUs active all the time isn't easy so Intel introduced HT as a way to get more usage out of them. AMD went a different route and just created two cores with two ALU's each. You can get more theoretical performance out of the chip (total not single) but at the expense of a ~20% efficiency hit when both cores on a module are loaded.

 

[The Q6660 Inside]

Wasn't the main reason Amd didn't include 3 ALU's is because they could not feed them fast enough??? doesn't the phenom design have 3 ALU's? Pretty sure the major issue with Piledriver is its shared decoder, FPU, and its high latency cache(bigger issue than you think). To my knowledge Steamroller fixes the Shard decoder issue and i think tweaks the FPU a bit more besides that minor tweaks the probably add up to 5% more performance at most so i would say 30% boost in Performance per clock is not out of question the shared decoder was a huge issue by its self(20 to even 25% performance penalty).

But now since its been confirmed that BULK will be used instead of SOI the clock speeds are questionable they do still have Clock mesh technology but that only goes so far. For example if they clock this new CPU at 2.9Ghz the performance will most likely be the same as the A10 6800K(except better graphics).

Does anyone know when exactly these new chips will be out?

The APUs will be out by Very Early Q1 2014 to the consumer, and Deployed to the OEMs and sellers in December IIRC. For all we know, it still has 2 ALUs per core and yes even Agena had 3 ALU/core. Considering the 100W TDP of the new A10, I assume we could be somewhere around 3.3 with 3.6 Turbo Boost in terms of clock speed.

That's F*(@#$%^ Pathetic...gain 30% performance and lose it all back in clock speed...

I seriously hope they're not that foolish...maybe that's why they held it up to Q1 2014...perhaps...we'll get similar clocks. I doubt it on bulk...but even if the turbo core on the new ones matches the base clock of the 6800k and it's ilk, then maybe we will see some performance gains.

Oh, and kiss overclock headroom goodbye...

🙁

They have to be cooking up something special for the high performance parts...

Now continue the argument, link with the bold part in above quote and you almost get it.

Consider the dies that didn't pass FX6 qualifications cost $100M. Take those $100M and add the cost to use those dies for FX4 chips; this includes: fabrication (including printing FX4xxx labels on the chip), additional quality testing, packaging, storing, shipping to distributors... The cost also includes the cost of maintaining the inventory, custom taxes for international shipping, web pages for the product, and so on. Imagine the whole cost are another $100M.

Now assume during an instant that nobody want to buy FX4 series chips, if you are fabricating them you are loosing $200M and it is better if you stop at the splitting point where some dies didn't pass FX6 qualifications and put in the trash those dies.

Umm no ... just ..... no ... stop it. Your math is horribly bad.

The QA binning process happens near the end of a production line not the start. You don't know ahead of time if the die is gonna be good enough for the top product. The production cost of $100M is already spent, it's done, finished. All you need to do is set the feature flags and bolt it into it's packaging. That cost is near zero.

AMD is making profit on the FX4xxx series chips, which are just defective FX8xxx series. That means they make even more profit on the FX8 and FX6 chips. The only time they don't make a profit is if they throw the chip into the trash.

Yes...

This is further evidenced by the old Phenom II series having "core unlocker" motherboards. Where some chips that could have been, say a X4 955 were deactivated and sold as a X3 or X2 because of demand. People would then buy the significantly cheaper X2 or X3 and attempt to unlock the cores, in some instances they would get no cores, some times they would get 1 extra core, and other times they would get 2 extra cores for free by "unlocking" the extra cores.

Now, sometimes you got a correctly binned chip that was supposed to be a X2 or X3 and got nothing. However, about as often as not, you got at least 1 extra core out of the ordeal.

They tended to be unreliable when you OCd em.

 

[8350rocks]

Now continue the argument, link with the bold part in above quote and you almost get it.

Consider the dies that didn't pass FX6 qualifications cost $100M. Take those $100M and add the cost to use those dies for FX4 chips; this includes: fabrication (including printing FX4xxx labels on the chip), additional quality testing, packaging, storing, shipping to distributors... The cost also includes the cost of maintaining the inventory, custom taxes for international shipping, web pages for the product, and so on. Imagine the whole cost are another $100M.

Now assume during an instant that nobody want to buy FX4 series chips, if you are fabricating them you are loosing $200M and it is better if you stop at the splitting point where some dies didn't pass FX6 qualifications and put in the trash those dies.

Umm no ... just ..... no ... stop it. Your math is horribly bad.

The QA binning process happens near the end of a production line not the start. You don't know ahead of time if the die is gonna be good enough for the top product. The production cost of $100M is already spent, it's done, finished. All you need to do is set the feature flags and bolt it into it's packaging. That cost is near zero.

AMD is making profit on the FX4xxx series chips, which are just defective FX8xxx series. That means they make even more profit on the FX8 and FX6 chips. The only time they don't make a profit is if they throw the chip into the trash.

Yes...

This is further evidenced by the old Phenom II series having "core unlocker" motherboards. Where some chips that could have been, say a X4 955 were deactivated and sold as a X3 or X2 because of demand. People would then buy the significantly cheaper X2 or X3 and attempt to unlock the cores, in some instances they would get no cores, some times they would get 1 extra core, and other times they would get 2 extra cores for free by "unlocking" the extra cores.

Now, sometimes you got a correctly binned chip that was supposed to be a X2 or X3 and got nothing. However, about as often as not, you got at least 1 extra core out of the ordeal.

They tended to be unreliable when you OCd em.

That's predominantly why I shied away from mentioning overclocking there...I knew some that had a X3 turned into X4 that overclocked well and stayed that way for several years. I knew others who had bad luck and couldn't run stable at all much over stock clocks.

 

[blackkstar]

Returning to the origin of this discussion: AMD is not releasing FX 8-core Steamroller because there is no enough market for that.

I seriously think this part is entirely rubbish. There is most certainly a market for it, and it is pretty substantial; however, they're losing a great deal of that business to the 4770k unless they act sooner than later.

I suppose you would argue there is no market for the 4770k then? 4c/8t chips are the same thing as 8c/8t chips in the grander scheme of things.

The 8 core AMD chips are aiming at the 4 core market...last time I checked, that market share had actually passed 50% of the PC market in the US. Considering every house has at least one PC, that means there are at minimum 175 million quad core PCs in the US alone. Think they couldn't take some of the 4 core i7 business with a hot 8 core that could keep up at single thread workloads? I know they would seriously compete...

You cannot translate market share from a company to another. A company that is economically healthy can afford niche markets, whereas a company who was in the red some months ago cannot.

i7-4770k and FX-8350 are not mainstream. Stop from pretending that 50% of the market would be interested in a FD-SOI 5GHz 8-core FX Steamroller.

Please stop with this FUD. DT overall is declining but Gaming PCs are rapidly growing.

http://techreport.com/news/25197/as-overall-pc-market-declines-growth-expected-in-gaming-hardware

PC market is "dying" because the folks who had desktops to check on Facebook and social sites and not much else are replacing them with tablets, phones, etc.

The Gaming PC market is growing because people aren't happy with consoles anymore and the barrier to entry for a decent gaming rig is not high. For a little more than Xbone you can get a decent gaming rig, and if you have parts from previous builds (like monitors, keyboards, etc) you can really make out well.

gaming PC will continue to grow. AMD is doing everything it can do help it grow right now (making ports easier, providing top end hardware for PCs, providing console hardware, providing tools for console developers).

If/when AMD releases a HEDT cpu, it will look significantly different from bdver2 and bdver1. If you remember, AMD designed it as a server chip first and a PC chip second, meaning it's more focused for integer workloads.

AMD is pushing gaming really hard and it's the big growth sector in desktops. They have a chip that is not that strong at float, which games love, and is strong in integers.

I would not be surprised if SR for HEDT was completely cancelled and that crazy chip we saw with the beefed up FPUs was what was going to replace it. AMD would sell more CPUs if they were focused on gaming as opposed to servers.

So the theory I'm proposing is that AMD will let ARM + Jaguar take over servers and then release a dCPU that is more optimized for games.

The evidence I'm using for this argument is that:

The die shot leak has a beefed up FPU
AMD claims they are focusing on higher growth markets
AMDs dCPU right now is aimed at a low growth market
Gaming PC is a relatively high growth market.

Therefore, AMD might change their dCPU design such that it fits more in line with the needs of games as opposed to servers.

Existing SR design is a server chip with a fixed front end. AMD has control of game software now very well with their console wins. I expect them to capitalize on it rather heavily.

Take the 0.3% share adds a 6.5% grow predicted by JPR and the result is 0.3%: 0.3x1.065=0,3195 which rounds to 0.3. Now takes 91% and add the same growth you obtain 97%. Therefore it is natural that AMD is releasing products for ~97% of the market.

I see no evidence for what you claim. What I see is what follows.

1) AMD is migrating from CPUs to APUs. Nov. conference has been renamed as APU 3.
2) Server roadmaps suggest no SR FX 8-core. Warsaw uses PD.
3) Centurion labeling as FX-9000 series suggest there is no more FX.
4) Some docs seem to suggest that FX series is finished.
5) Contrary to common misconception MANTLE has been released with APUs in mind. Increasing 50% top dGPU is not so important like increasing 50% the performance of an APU
6) New W8.1 drivers released to increase APU performance during gaming
7) Slides such as this

Look to the first item in the "FROM".

FROM here must be taken as synonym for "yesterday AMD"

[/quotemsg][/quotemsg]

1) HSA is extremely important to AMD and APUs are the ONLY products they have that support HSA completely. That doesn't mean that APU is all AMD is going to go after. They could very well be pushing APU now to further HSA for now, and later on come out with an HSA enabled system. Considering AMD's flexible system with basically interchangeable SoC parts, I don't see that as impossible.

2) And I have been saying there is no SR FX 8 core on server because that version of SR has been cancelled for excavator or another version of Steamroller. Rumors of two versions of Steamroller have been around for YEARS and to see one version of Steamroller come out for one product segment doesn't mean that SR is dead.

3) Yes it does, but I've addressed this before that FX brand name has been tarnished beyond belief. It went from an image of beating Intel EE editions to bdver1 CPUs. If AMD released a good CPU, the last thing they would want to do is call it FX. FX is synonymous with "faildozer", "heatdozer", and all sorts of names. FX, to a lot of people who weren't even around in the P4EE days, means a hot, slow chip that was barely faster than the chip it replaced. That's a huge problem. Look at how AMD renamed their GPU products and look how well they are selling so far. Already completely sold out on Newegg...

4) See 3

5) I absolutely agree with you on this and Mantle is an extremely valuable asset to AMD's entire product stack, from mobile GPUs to low end dGPU to high end 290x setup with 4 way crossfire.

6) The Windows 8.1 improvements more than likely have a lot more to do with the fact that XBone runs windows 8.1 kernel (and probably more) and MS is just letting the XBone performance modifications trickle over to PC. They have no reason not to.

7) I still don't see that slide as proof that AMD is done with HEDT class CPUs.

"Designing/building high performance products with emphasis on cores"

That does not mean that AMD is no longer going to focus on high performance products. It means they are done with MOAR COARS, which they should be done, a 5m or 6m chip would be ridiculous and no gamers would want to touch it. AMD already has had enough of a problem getting people to go 8 core CPU by (previously, before the console wins) making it a point that you could stream games and play games better on 8 core than 4c or 4c/8t.

The node thing has always been a problem, and it's good they want to fix it.

"power-performance optimized cores" can mean anything. Even Intel 8 core could be considered "power-performance" type of design. You are skewing it to imply it's only APU.

"Agile and Flexible SoC Methodology" doesn't mean it's going SoC only, it could mean that SoC methodology is only part of the picture. I would even argue that this already happened with AMD going for the whole "we're going to offer 'building blocks' for our customers to create semi-custom solutions". It doesn't mean that AMD can not take those SoC components and make their own non-SoC products with them. Ergo, GPU is not an SoC yet AMD released dGPUs.

Rest of the points seem APU neutral.

However the point I am trying to make is that a lot of these points you are making are not concrete in any form. Do I 100% think that AMD will release a big CPU in the future? No. Do I think they might? Yes.

(sorry I screwed up my quote before, This forum has some sort of problem with trimming people's comments)

 

[8350rocks]

Anyone else miss something that is very telling about AMD's future HSA capabilities in the 290X review?

The new crossfire mechanism (XDMA) uses the PCIe bus to transfer data.

Think about that for a moment, no CF connectors to transfer data.

If they can transmit data across the PCIe bus well enough to crossfire 2 GPUs and have it be better than the old method that used the bridge connectors...then what's to stop them from using HSA on a GPU with a HSA enabled dedicated CPU as well?

Absolutely nothing. This reminds me a bit of hcl123's post about using HTX, which would be more efficient, to enable HSA through a PCIe slot that also used HTX. However, "small moves Ellie" (to borrow a line from Contact), this is a step in a direction that allows AMD to do many things that Intel and Nvidia cannot even fathom. It's also gone primarily uncommented and unnoticed so far, perhaps that's a good thing.

 

[The Q6660 Inside]

Anyone else miss something that is very telling about AMD's future HSA capabilities in the 290X review?

The new crossfire mechanism (XDMA) uses the PCIe bus to transfer data.

Think about that for a moment, no CF connectors to transfer data.

If they can transmit data across the PCIe bus well enough to crossfire 2 GPUs and have it be better than the old method that used the bridge connectors...then what's to stop them from using HSA on a GPU with a HSA enabled dedicated CPU as well?

Absolutely nothing. This reminds me a bit of hcl123's post about using HTX, which would be more efficient, to enable HSA through a PCIe slot that also used HTX. However, "small moves Ellie" (to borrow a line from Contact), this is a step in a direction that allows AMD to do many things that Intel and Nvidia cannot even fathom. It's also gone primarily uncommented and unnoticed so far, perhaps that's a good thing.

Well, isn't HTX half dead, aside from that one PCI-E hybrid combo slot patent or whatever it was? The 2900XT, whoops, 290X is a damn good card, we can all thank Nvidia for jacking the GK110 consumer card prices sky high or else this will have been another GTX 480, but the 470 (290) would be admired throughout history, or a few years before being a distant memory...

(man I like that shroud!)

 

[-Fran-]

So, what you're saying is...it takes 2 quad core ARM CPUs to defeat a dual core Intel Atom...

That doesn't at all bode well for ARM.

Take a closer look at the numbers and TDP of each. Also, didn't the Atom have HT? Or was it 1 core with HT? Uhm... I can't remember.

Anyway, it does show ARM in a positive light IMO. Put an i3 ULV in there and I'm sure they'll be awfully close. I think ARM can do better in higher TDPs and put a good show to AMD and Intel. I wonder how it would compare to Brazos or Temash... That would be interesting to see as well.

Cheers!