How a R3k AMD CPU can keep up with a R8k Intel CPU :)

[Progenix Oj101]

It isn't exclusive period. It is called a custom order. R&D costs go into making the technology and that was done for their current line of APUs. It's a matter of piecing it all together from existing desktop and laptop technology. The margins would not be slim and profit made would actually help subsidise the cost of R&D.

The SoC used in the PS4 and Xbox One has a higher spec than any other Jaguar processor. That doesn't happen by changing a line in software or something to that effect, it is a whole new SKU while needed to be developed. There most certainly WERE R&D costs involved. Regardless, AMD does not get paid a bucketload of money for each SoC they sell in the PS4 and Xbox One, the margins are pretty slim.

I didn't mention anything of the kind. I mentioned a performance increase implying it to be over the regular desktop line.

Sorry, my misunderstanding. If that's the case, you're comparing a regular desktop setup to a HEDT setup, and using gaming as the benchmark for such is not fair. You don't buy a Core i7-5960X for gaming when an i7-4790K will do you just as well if not better. Unless, of courese, you're planning to run four top end graphics cards and need the available PCIe lanes to get every last fraction of a percentage point worth of performance.

If you're doing media editing or anything else that is heavily multithreaded, HEDT is a platform to consider. As I've said before, complaining about poor value for money with HEDT as a gaming rig is like complaining about poor value for money with a Quadro graphics card in a gaming rig.

It's the new units that set the market price and says a lot about what each company is capable of. Besides you are not being fair. You are comparing second hand Intel prices to new AMD prices. Sorry for being crass here but that's really scraping the bottom of the barrel in order to force Intel into a good light.

I was being kind, not unfair. I was showing an older, cheaper yet still faster Intel CPU against the FX-8350. My question stands - why spend more on a slower AMD processor than what you can get from Intel by looking at something three generations old? If you want me to compare price point to price point, the link I included previously (http://www.bit-tech.net/hardware/2014/07/03/intel-core-i5-4690k-review/5) will show you just how badly the i5-4690K walks all over the FX-8350 as soon as the CPU comes heavily into the equation during gaming. That i5-4690K is around R 3,000 - the same ballpark as the FX-8350 - yet is SIGNIFICANTLY faster. You also don't need a motherboard costing thousands for a gaming rig - what's wrong with the sub R 1,200 ASRock H97 Pro4? Four RAM slots, six SATA ports, what more do you need? If you're running 12+ hard drives, I doubt that the price of a motherboard is a major concern as you would have spent several times that on drives (I hope you're not running a dozen 250 GB drives - you're not, ARE YOU? ). Overclocking? Not really needed. Not when the the performance gap shown between the FX-8350 and i5-4590K is so large. You're not going to get a 100% overclock on the AMD, so you're not going to catch up to the stock i5-4690K. If you do want to overclock, the ASRock Z97 Pro4 is about R 500 more and has similar specs.

In fact, a little Core i3-4160 is faster in CPU intensive games and it's half the price of the AMD CPU

What you're still ignoring is that it doesn't do so at the same price point. The AMD cpu is less expensive than the Intel and so are the AMD motherboards. If there's any platform where you can actually afford a more expensive gfx card it's AMD and then it would beat the Intel one.

Are you saying that if the CPUs cost the same (which they do - around R 2,900 a piece) and the motherboards are R 200 apart (working on the R 1,200 board I mentioned vs the R 1,000 you mentioned) you should go for the SUBSTANTIALLY slower CPU? R 200 is hardly the price of a DVD writer, it isn't going to get you a graphics card exponentially faster than what you could have bought without that R 200.

Coming back to the second hand platform I mentioned, that works out R 2,000 cheaper than an FX-8350 and motherboard. R 2,000 WILL get you a much faster graphics card at almost any price point.

The AM3 socket was introduced in 2009. The AM3+ was introduced in 2011. There are other reasons for the performance bottleneck we are currently seeing like the switch to full computer design where previous chips were manually tweaked. This will allow more complex designs in future though. At some point Intel will experience the same bottleneck when AMD has moved past it. We can already see this happening. The Bulldozer architecture didn't offer the performance people were expecting where some of their earlier processors performed better in some regards. It's successors however did show the performance increases we expect to see if AMD was producing better chips.

As I said (twice), I was trolling when I said old socket = old performance.

And as we've seen it's hard to justify buying a more expensive Intel cpu and motherboard when the money can be spent on a gfx card that will possibly own the Intel rig of the same price.

See above.

Here's a few things you either don't know or have forgotten.

AMD brought 64-bit desktop computing to the market by developing their own architecture. Intel didn't have any interest in bringing their IA64 architecture to the desktop until they saw the potential of losing a lot of money to AMD. They went to Microsoft to get them to develop a consumer version of Windows that would run on their IA64 platform. MS said it was a no go. They were already developing for the AMD64 architecture which was cutting edge and there wasn't a need to include the Intel architecture. If Intel released a desktop IA64 it would be unsupported. Seems AMD isn't the only one making bad decisions.

Yes, they also brought the IMC controller to the desktop market (first introduced by DEC and HP in the early to mid 1990s), HyperTransport (not adopted by as such by Intel, but analogous to the QPI link) and the first true quad core desktop processor (the Kentsfield microarchitecture in the original Core 2 Quad was a MCM design).

I never said Intel didn't make mistakes. They did. Netburst was probably their biggest mistake, even though they turned it into a success it was a technological failure with far poorer IPC than P6 which was seen in the Pentium 3 before it. A step backwards, and a step AMD also made with Bulldozer. Pursuing IA64 was a bad decision indeed when compared to AMD64, and a mistake AMD also made by pursuing 3DNow! instead of MMX. They also missed out on scoring a deal with the iPhone, although that is more a lack of foresight (nobody could have known how big the iPhone would be). There was an FPU bug in the original Pentium, another mistake AMD repeated with the errata bug in the original Phenom. The list goes on. AMD's list is longer, their biggest mistakes were a lack of advertising and sitting on their haunches when they had a huge lead over Intel - a lead they could potentially have kept had they focussed on a K8 replacement sooner.

The full potential of AMD's Athlon 64's were never realised as computers at the time still didn't have a lot of memory. It's only recently that 4GB memory started to become the norm. 64-bit processes are actually slower in general on systems with less than 4GB memory so people that switched to 64-bit windows too early saw a performance decrease. If you have a computer with 4GB or more memory, and everyone will in time, then the performance increase you are experiencing is thanks to AMD.

That isn't the only tech AMD introduced. Their module architecture is way ahead of our time. By combining certain units so they are shared by two cores will allow more cores on a single die. Processes must use the resources more efficiently though like not running two threads that both require the coupled FPU on the same module. This would require OS support as well which is currently non-existent but when it's there you can have a performance increase by running two processes that can use an uncoupled FPU on the same module.

AMD's current CMT (module) design is a step up from HyperThreading, but a step backwards from an additional full core. Not everything can be coded to use the design efficiently, and this is an issue that effects SMT as well. Both SMT and CMT will lag behind a true multicore design in some instances. AMD has plenty of experience making desktop 6-core processors, I'm not sure why they adopted the current module design but, as I have shown, it is significantly slower on both a thread-for-thread and clock-for-clock basis compared to the older Thuban core.

Same with their new APU chips and the ARM Cortex-A5 co-processor that can be used much more effectively to run an OS for example but it's RISC architecture can also be used more effectively for some other tasks.

RISC is actually still present in desktop CPUs. SSE instruction sets are very close to RISC.

Breaking this post cause it is too long (more than 10,000 characters).

 

[Progenix Oj101]

Their Heterogeneous System Architecture is another unseen first from AMD. Instead of memory needing to be copied between CPU and GPU memory for respective execution on one or the other the memory is shared and accessible by both. Instructions working on memory can be executed on whichever processor is best suited for them without the need to transfer data between two memory pools.

AMD is definitely leading and inventing the industry but as always the market will not recognise what's way ahead of it. When these technologies are used by Intel to make your systems perform better though I'll still be supporting the company that are responsible for them.

Why support the company with the slower hardware just because they invented something? That's like driving a Cadillac because they were the first car to use the conventional layout was the Cadillac Type 53. I put my money where the performance is, for the first half of the 2000s that was AMD but they haven't caught up again since.

 

[Chevron]

Why support the company with the slower hardware just because they invented something? That's like driving a Cadillac because they were the first car to use the conventional layout was the Cadillac Type 53. I put my money where the performance is, for the first half of the 2000s that was AMD but they haven't caught up again since.

Because during the P4 days, and after Intel has proved themselves to be anti-competitive. They paid suppliers to not stick AMD. And this was proven in the courts.

I refuse to support a company like that.

And if AMD dies it will only get worse. Expect P4 prices on super slow hardware all over again.

 

[Progenix Oj101]

Because during the P4 days, and after Intel has proved themselves to be anti-competitive. They paid suppliers to not stick AMD. And this was proven in the courts.

I refuse to support a company like that.

And if AMD dies it will only get worse. Expect P4 prices on super slow hardware all over again.

You know they're not the only company to use unscrupulous tactics, right? Regardless, having a personal reason not to use them is not a valid argument for saying that AMD is a better purchase.

 

[Chevron]

You know they're not the only company to use unscrupulous tactics, right? Regardless, having a personal reason not to use them is not a valid argument for saying that AMD is a better purchase.

Value for money AMD is the better purchase. It's win-win for everyone.

 

[Progenix Oj101]

Value for money AMD is the better purchase. It's win-win for everyone.

Proof? I provided proof to the contrary, showing that in CPU intensive games the cheaper Intel CPUs can be faster in games. For a larger reference, have a look here http://www.anandtech.com/show/7963/the-intel-haswell-refresh-review-core-i7-4790-i5-4690-and-i3-4360-tested/9

No, they don't provide benchmarks for the FX-8350, I can't find a review that does offhand, but you can see the Core i5-4690K (same price range as the FX-8350) is on par with the i7-4790K in everything other than Bioshock.

This whole thread is a bit of a joke considering the first post where a HEDT system was used as a benchmark for gaming but anyway

 

[CataclysmZA]

Not to even mention all the money AMD is making from Microsoft, Sony and Nintendo. Intel is really loosing out on all the consoles sold.

Not really. During Q3 2014, Intel pissed away gobs of money into their mobile division, roughly the same amount that AMD recorded as their revenue for the quarter. Neither Intel nor Nvidia are feeling the pinch of not being in the console market. In fact, Nvidia is still reaping rewards from the licensing deal they struck with Sony for the PS3, which was a ripoff that Kuturagi could have easily avoided had he not been so hell-bent on making the world's most powerful gaming console.

It's designed for video/audio and image processing and even has slightly different features than a normal i7, which makes it significantly faster for the tasks it was designed for, but sacrificing it's performance in other areas like gaming.

I'd like to point out that at no point is Intel "sacrificing gaming performance." Intel's HEDT chips have large power draws and high TDPs and they frequently sit near those levels when being run through benchmarks or taxing workloads. "Gaming performance" usually comes from high-clocked cores that stay close to their rated speeds.

A new socket for every processor design isn't something that puts you ahead of your competition.

In Intel's case it actually does, as they offer better performance and higher efficiency with each new generation of products on a new socket (giving their board partners incentive to design better boards or differentiate on useful features). AMD, on the other hand, won't have anything new for the next two years for desktop users as their focus is now on APUs. The only reason why so many board vendors are jumping on AM1 is because it holds great promise as a low-cost computer and socketed SoC system that can see yearly upgrades and performance improvements just by replacing the APU.

AMD is actually expected to consolidate their socket designs.

And it's a move that directly benefits them primarily, not the consumer.

That's why you budget for an i5 instead and throw everything else into the rest of the system.

This doesn't make any sense to me. Systems based on a Core i5-4690K and a mid-range Z97 board is similarly priced as a FX-8350 and MSI's 970 Gaming.

 

[Chevron]

And it's a move that directly benefits them primarily, not the consumer.

And now you're talking out your ass.

The fact a 2/3/4year old AMD mobo can run the latest CPUs is fantastic for the consumer.

 

[HideInLight]

This doesn't make any sense to me. Systems based on a Core i5-4690K and a mid-range Z97 board is similarly priced as a FX-8350 and MSI's 970 Gaming.

That's the whole point, it was in context of getting an i5 over an i7 because that's where the price comparison and performance really falls. That's also an overclockers i5 setup.
But the fact is you can go a lot cheaper if you stick to stock, you put a R2300 i5 4th gen in a R700 Motherboard if you want to.

But I prefer secondhand, got my 3rd gen i5, a motherboard and 4 gig random for a total of R2400...

 

[CataclysmZA]

The six year old Core i7-920 is still a very capable CPU when it comes to gaming, why would you upgrade to a 2600K, 3770K, 4790K... etc for gaming?

Wouldn't higher base and boost clocks and better IPC be a valid argument? There's already a good case where moving from a i7-920 to a i7-2600K produces a noticeable boost in performance (frame-time wise) for CPU-limited games like Batman: AC, Crysis 3 and ARMAII. IIRC, comparing Nehalem to Haswell is already a 40-50% boost in general performance.

The actually used laptop parts for the consoles.

No. No they did not. Please don't start spreading the message that they are.

AMD was in a position where every cent they made counted; while the margins aren't public knowledge, they are slim.

Had AMD stuck with Thuban, added SSE 4.2, added an extra two cores and bumped up the clock speed a bit it would have had a winner - maybe not enough to topple Intel's chips, but it would have been at least 30 % faster. Bulldozer and onwards are failures compared to what AMD had.

The margins are actually public knowledge. I'll find the links from the articles as soon as I remember where I saw them (most likely Digitimes or Forbes), but AMD's making about $15 from each unit sold to Microsoft (margins from Sony will be similar) and Lisa Su was quoted as saying they were looking forward to extending that to around $20 by 2h 2015 thanks to cost-cutting and possible die shrinks, extending to $30 by 2017.

As for Thuban, its definitely possible that there could have been improvements left in the pipeline, but they were already hitting TDPs there as well. A die shrink wouldn't have solved their problems much.

The fact a 2/3/4year old AMD mobo can run the latest CPUs is fantastic for the consumer.

It only really works for two upgrade cycles at the most now. If anyone bought a 990FX board it would have supported Bulldozer and Piledriver with a downgrade to Phenom II, but that's about the extent of things now. FM1 was a dead end, as was FM2. FM2+ is a little better with support, but it to will go the way of the dodo two years after its launch (and its support for Bulldozer and Piledriver was a by-product of AMD merely iterating on the original Bulldozer design).

Intel is right - people only do entire platform upgrades now, relatively few will actually buy a board and shove several processors in there.

That's the whole point, it was in context of getting an i5 over an i7 because that's where the price comparison and performance really falls.

Ah, now I see.

 

[Swa]

The SoC used in the PS4 and Xbox One has a higher spec than any other Jaguar processor. That doesn't happen by changing a line in software or something to that effect, it is a whole new SKU while needed to be developed. There most certainly WERE R&D costs involved. Regardless, AMD does not get paid a bucketload of money for each SoC they sell in the PS4 and Xbox One, the margins are pretty slim.

Let's examine that claim. It seems you have a misunderstanding of what custom order is or in this case semi-custom order. With a custom order you can order basically anything that the manufacturer is capable of making. There's still not much R&D involved though as you have to provide the design and the manufacturer won't accept responsibility if it fails. Otherwise if they are doing the R&D you pay for it even before you get a single chip. They don't recover their costs only when they sell you the chip.

With semi-custom you have to stay within the limits of what they are already making or developed. So let's look at what the two processors have. The Jaguar is capable of 2GHz. The XBoxOne reportedly bumped the clock to 1.75GHz recently from it's it's original 1.6GHz. The PS4 is similar so there's no higher spec here. MS and Sony actually kept the spec low to keep the cost down for them. Both have 8 cores split among 2 compute units with 2x2MB L2 cache. This is essentially two processors on one die. Nothing new really as dual processor computers have existed for decades and AMD has a number of 4 core processors. The two have 768 and 1152 GPU cores respectively. Here they are differently speced but it doesn't take R&D to change GPU core numbers. The tech supports any number of cores that you want and can fit on the die. XBox uses DDR3 while the PS uses GDDR5, the controllers for both are now in existence for years already so still no R&D that I can see. The XBox has 32MB embedded eSRAM. I'm not sure if this is on-die or just closely linked but in either case it isn't anything we haven't seen before. The XBox360 had embedded memory and so did the Pentium Pro if I'm not mistaken.

So if there was R&D involved show us where it was as I can't find anything that they would have had to develop for either chip. This seems to be your main motivator for the low margins that you claim but if they did have to do R&D it would be recovered from Sony and Microsoft separately or push up the price of the chip. They wouldn't get a free ride from AMD.

Sorry, my misunderstanding. If that's the case, you're comparing a regular desktop setup to a HEDT setup, and using gaming as the benchmark for such is not fair. You don't buy a Core i7-5960X for gaming when an i7-4790K will do you just as well if not better. Unless, of courese, you're planning to run four top end graphics cards and need the available PCIe lanes to get every last fraction of a percentage point worth of performance.

If you're doing media editing or anything else that is heavily multithreaded, HEDT is a platform to consider. As I've said before, complaining about poor value for money with HEDT as a gaming rig is like complaining about poor value for money with a Quadro graphics card in a gaming rig.

And as I said you are sacrificing performance in one area for more performance in another. Performance sacrifice isn't something you usually pay extra for let alone R5k extra.

I was being kind, not unfair. I was showing an older, cheaper yet still faster Intel CPU against the FX-8350. My question stands - why spend more on a slower AMD processor than what you can get from Intel by looking at something three generations old? If you want me to compare price point to price point, the link I included previously (http://www.bit-tech.net/hardware/2014/07/03/intel-core-i5-4690k-review/5) will show you just how badly the i5-4690K walks all over the FX-8350 as soon as the CPU comes heavily into the equation during gaming. That i5-4690K is around R 3,000 - the same ballpark as the FX-8350 - yet is SIGNIFICANTLY faster. You also don't need a motherboard costing thousands for a gaming rig - what's wrong with the sub R 1,200 ASRock H97 Pro4? Four RAM slots, six SATA ports, what more do you need? If you're running 12+ hard drives, I doubt that the price of a motherboard is a major concern as you would have spent several times that on drives (I hope you're not running a dozen 250 GB drives - you're not, ARE YOU? ). Overclocking? Not really needed. Not when the the performance gap shown between the FX-8350 and i5-4590K is so large. You're not going to get a 100% overclock on the AMD, so you're not going to catch up to the stock i5-4690K. If you do want to overclock, the ASRock Z97 Pro4 is about R 500 more and has similar specs.

In fact, a little Core i3-4160 is faster in CPU intensive games and it's half the price of the AMD CPU

No you were using the price of SECOND HAND Intel compared to NEW AMD. That completely invalidates your question as AMD is less than Intel. The benchmarks you provided don't show the performance of an 8350 if the saving is spent on a better gfx card. If you're going to skimp on the motherboard just to make up for the price difference we might as well do that for AMD and go R600-R700 range. There's also nothing wrong with these AMD boards just as with your R1200 Intel board. Funny you're not considering a R600 Intel board.

Are you saying that if the CPUs cost the same (which they do - around R 2,900 a piece) and the motherboards are R 200 apart (working on the R 1,200 board I mentioned vs the R 1,000 you mentioned) you should go for the SUBSTANTIALLY slower CPU? R 200 is hardly the price of a DVD writer, it isn't going to get you a graphics card exponentially faster than what you could have bought without that R 200.

Coming back to the second hand platform I mentioned, that works out R 2,000 cheaper than an FX-8350 and motherboard. R 2,000 WILL get you a much faster graphics card at almost any price point.

The CPUs aren't the same. AMD is roughly R300 cheaper. Add that to the R700 price difference between similar quality AMD and Intel boards and you can get a substantially better gfx card.

Second hand holds no argument. If you're willing to go second hand Intel with a "special" you happen to have stumbled upon I might as well go second hand AMD. So you lose the advantage that second hand Intel had over new AMD and AMD still wins.

Yes, they also brought the IMC controller to the desktop market (first introduced by DEC and HP in the early to mid 1990s), HyperTransport (not adopted by as such by Intel, but analogous to the QPI link) and the first true quad core desktop processor (the Kentsfield microarchitecture in the original Core 2 Quad was a MCM design).

I never said Intel didn't make mistakes. They did. Netburst was probably their biggest mistake, even though they turned it into a success it was a technological failure with far poorer IPC than P6 which was seen in the Pentium 3 before it. A step backwards, and a step AMD also made with Bulldozer. Pursuing IA64 was a bad decision indeed when compared to AMD64, and a mistake AMD also made by pursuing 3DNow! instead of MMX. They also missed out on scoring a deal with the iPhone, although that is more a lack of foresight (nobody could have known how big the iPhone would be). There was an FPU bug in the original Pentium, another mistake AMD repeated with the errata bug in the original Phenom. The list goes on. AMD's list is longer, their biggest mistakes were a lack of advertising and sitting on their haunches when they had a huge lead over Intel - a lead they could potentially have kept had they focussed on a K8 replacement sooner.

This seems to be another belief not founded in reality. Intel's mistakes have been of a much bigger magnitude. Add to Netburst the monumental and expensive flop that was Rambus. 3DNow was a good idea but unfortunately the market didn't adopt it. Still it doesn't compare to IA64, one is a few additional instructions and the other is a whole architecture. 3DNow is actually superior to MMX allowing parallel floating point operations where MMX only allowed parallel integer operations. AMD's biggest mistake here was to not extend it and allowing Intel to take the lead with SSE.

AMD's second biggest mistake was probably repeating Intel's mistakes like Slot 1. Intel's successes as well weren't always their own. If I'm correct their temp control tech comes from Cyrix. Cyrix processors were always super overclockable where Pentiums struggled with heat issues.

It seems you still don't understand why the Bulldozer lacked performance. AMD introduced a completely computerised design system. No more manual tweaks but at the same time you can do more complex designs. This is something that had to occur at some point.

 

[Swa]

AMD's current CMT (module) design is a step up from HyperThreading, but a step backwards from an additional full core. Not everything can be coded to use the design efficiently, and this is an issue that effects SMT as well. Both SMT and CMT will lag behind a true multicore design in some instances. AMD has plenty of experience making desktop 6-core processors, I'm not sure why they adopted the current module design but, as I have shown, it is significantly slower on both a thread-for-thread and clock-for-clock basis compared to the older Thuban core.

The strength of the module design is that an additional core only takes up 50% more space and not 100%. In essence a module takes up 75% of the space of 2 full cores. The bulldozer was badly designed with the shared instruction pipeline being a bottleneck and causing a 5% performance loss. This was rectified with the Piledriver update. The floating point units have been extended to match Intel's 256-bit AVX units. However instead of having a 256-bit for every core it has 2x128-bits for every module that can be coupled to provide 256-bit only when needed.

RISC is actually still present in desktop CPUs. SSE instruction sets are very close to RISC.

Nope, if it has CISC instructions it's no longer RISC. SSE has complex instructions so it can't be termed RISC and even if it didn't it still part of the CISC instruction set and not a separate instruction set. A separate core with a RISC instruction set can be optimised for those instructions.

Why support the company with the slower hardware just because they invented something? That's like driving a Cadillac because they were the first car to use the conventional layout was the Cadillac Type 53. I put my money where the performance is, for the first half of the 2000s that was AMD but they haven't caught up again since.

Because of principles for one but you haven't actually shown that Intel gives you more bang for your back. I guess you also don't see the irony that it's because of our support for AMD that you have the privilege of paying Intel more for something that's faster.

Proof? I provided proof to the contrary, showing that in CPU intensive games the cheaper Intel CPUs can be faster in games. For a larger reference, have a look here http://www.anandtech.com/show/7963/the-intel-haswell-refresh-review-core-i7-4790-i5-4690-and-i3-4360-tested/9

No, they don't provide benchmarks for the FX-8350, I can't find a review that does offhand, but you can see the Core i5-4690K (same price range as the FX-8350) is on par with the i7-4790K in everything other than Bioshock.

This whole thread is a bit of a joke considering the first post where a HEDT system was used as a benchmark for gaming but anyway

No you didn't prove the contrary. Your comparisons are based on CPUs only but as we've shown you don't buy just a processor but a whole computer and money wise an AMD rig offers more value than an Intel one.

 

[HideInLight]

Actually going a low cost motherboard for a FX-8350 is really dangerous.
Those things can't handle the TDP.

Your kinda forced to go with the R1400+ range.

I think there's a reason AMD abandoned the FX range.

 

[Chevron]

No. No they did not. Please don't start spreading the message that they are.

Um yes.

AMD will be building two APUs based on Jaguar: Kabini and Temash. Kabini is AMD’s mainstream APU, which you can expect to see in ultra-thin affordable notebooks.

In both our Xbox One and PS4 articles I referred to the SoCs as using two Jaguar compute units - now you can understand why. Both designs incorporate two quad-core Jaguar modules, each with their own shared 2MB L2 cache.

http://www.anandtech.com/show/6976/...wering-xbox-one-playstation-4-kabini-temash/4

Do you get paid to be wrong or something?

 

[Swa]

It only really works for two upgrade cycles at the most now. If anyone bought a 990FX board it would have supported Bulldozer and Piledriver with a downgrade to Phenom II, but that's about the extent of things now. FM1 was a dead end, as was FM2. FM2+ is a little better with support, but it to will go the way of the dodo two years after its launch (and its support for Bulldozer and Piledriver was a by-product of AMD merely iterating on the original Bulldozer design).

Intel is right - people only do entire platform upgrades now, relatively few will actually buy a board and shove several processors in there.

Actually every processor from the AM3 generation onwards is compatible with AM3+ so that would be Athlon II, Phenom II, Bulldozer, Piledriver and we're still expecting Steamroller and Excavator before AMD releases their new Zen architecture. What's more AMD's Bulldozer chips which don't feature the extra pin that was added to AM3+ can work on some AM3 motherboards with a BIOS update. When has Intel ever accomplished such compatibility? Instead they have a different motherboard for every type of current generation processor that confuses consumers and limits upgradeability. Yes, Intel is right, if you're using them your only option is an entire platform upgrade.

If AMD consolidates their FMx and AMx for the new Zen processors it would be a huge benefit to consumers. Two different architectures on the same board so you can upgrade from integrated graphics to a more powerful processor and discrete graphics.

 

[CataclysmZA]

Um yes.

http://www.tomshardware.com/reviews/microsoft-xbox-one-console-review,3681-3.html

Do you get paid to be wrong or something?

That link 404's, so which laptop parts are you talking about? Certainly not the APUs and memory subsystems, that's for sure.

What's more AMD's Bulldozer chips which don't feature the extra pin that was added to AM3+ can work on some AM3 motherboards with a BIOS update. When has Intel ever accomplished such compatibility?

You can run socket LGA771 Xeon processors on LGA775 motherboards using a paper shim to cover up some pins, but that's neither here nor there. Neither Intel nor AMD officially supported these methods of getting chips to work on sockets they weren't really designed for.

Also, accept that AM3+ is dead already. AMD has said as much in their roadmaps. Steamroller won't come to it, Excavator is certainly not going to arrive on it. Its dead, let it be.

 

[Progenix Oj101]

Wouldn't higher base and boost clocks and better IPC be a valid argument? There's already a good case where moving from a i7-920 to a i7-2600K produces a noticeable boost in performance (frame-time wise) for CPU-limited games like Batman: AC, Crysis 3 and ARMAII. IIRC, comparing Nehalem to Haswell is already a 40-50% boost in general performance.

Enough to warrant spending thousands on a new CPU and motherboard? I think not. Crysis 3 between a Core i7-920 and Core i7-3770K is about a 15 % difference, the difference is as small as 10 % in ARMA III. Batman AC shows a larger difference at close to 20 %, but that was compared to the i7-4770K and even then the i7-920 was pushing 140+ FPS. Techspot.com has tested all of the games you mentioned and more on a large variety of CPUs, and for what it's worth the i7-920 and FX-8350 were normally on par - often the exact same FPS (eg Batman 141 vs 144 to the FX-8350, ARMA III 48 vs 48 - that's with the i7-920 at 2.66 GHz and the FX-8350 at 4.00 GHz, stock speeds for both. The i7-920 has headroom for about a 50-60 % overclock, the FX-8350 has headroom for about a 20 % overclock).

The margins are actually public knowledge. I'll find the links from the articles as soon as I remember where I saw them (most likely Digitimes or Forbes), but AMD's making about $15 from each unit sold to Microsoft (margins from Sony will be similar) and Lisa Su was quoted as saying they were looking forward to extending that to around $20 by 2h 2015 thanks to cost-cutting and possible die shrinks, extending to $30 by 2017.

To give some perspective, margins on CPUs and GPUs can be in excess of $500.

As for Thuban, its definitely possible that there could have been improvements left in the pipeline, but they were already hitting TDPs there as well. A die shrink wouldn't have solved their problems much.

Die shrinks result in a lower TDP. They've taken a step backwards as I showed several posts back, not just in IPC but in core-for-core performance as well. Thuban with two extra cores and a 200 MHz speed bump would have been a 45 % increase in multithreaded performance, instead we saw (in certain cases) Thuban beating out the newer CPUs with both a core AND frequency deficit. A shrink to 32nm would have allowed the speed and core increase without an increase in TDP, while offering a faster CPU.

 

[Chevron]

That link 404's, so which laptop parts are you talking about? Certainly not the APUs and memory subsystems, that's for sure.

The link had an extra forward slash, but I edited the post with an Anandtech link so it would be easier for you to understand.

 

[Chevron]

Die shrinks result in a lower TDP. They've taken a step backwards as I showed several posts back, not just in IPC but in core-for-core performance as well. Thuban with two extra cores and a 200 MHz speed bump would have been a 45 % increase in multithreaded performance, instead we saw (in certain cases) Thuban beating out the newer CPUs with both a core AND frequency deficit. A shrink to 32nm would have allowed the speed and core increase without an increase in TDP, while offering a faster CPU.

Funny thing is I'm still running fine with my 3.2Ghz Phenom II.

Paired with a 7950 there's no game I can't run at 1920*1200.

 

[Chevron]

Also, accept that AM3+ is dead already. AMD has said as much in their roadmaps. Steamroller won't come to it, Excavator is certainly not going to arrive on it. Its dead, let it be.

I doubt Steamroller will come to AM3+ as it's too late already, but I haven't seen anything regarding excavator though and I've been looking. All I find are forum posts of people asking the question if/when excavator will be coming to AM3+.