28nm Trinity Successor Rumored To Debut in Q2 2013

[A Bad Day]

[citation][nom]otacon72[/nom]AMD is so far behind intel it's just sad at this point. APUs are a joke.[/citation]

Look pal, it's a lot more expensive for a lappy motherboard to have two sockets and a PCI-E interface for a dedicated GPU and CPU, than to have one socket and a combo APU. Not to mention the cooling system would be simpler and cheaper.

And if you're going to say how great Intel's IGPs are, you're insane. They might be sufficient for videos, but choke under gaming.

This isn't the 1990's and early 2000's era anymore, where games relied mostly on CPU.

 

[K2N hater]

[citation][nom]otacon72[/nom]AMD is so far behind intel it's just sad at this point. APUs are a joke.[/citation]
You could also say Intel CPUs are a joke because the integrated graphics don't play Crysis.

 

[blazorthon]

[citation][nom]urban legend[/nom]OK here's the deal being totally unbiased.AMD is good for APU and server environment right now.Desktop is over with (except for APU).so bottom line unless going APU, you go Intel.[/citation]

FX-6100 or FX-8120. Disable one core per module to get three or four single core modules, three for the 6100 and four for the 8120. You get a roughly 25% boost in performance at the same frequency as you previously had with all modules sharing their resources between two cores instead of a single core per module. That puts them a little ahead of Phenom II in performance per Hz while making them capable of significantly higher frequencies while using about the same or even somewhat lower amounts of power.

It also gives them even higher overclocking headroom to the point where they can easily compete with non-K edition i5s and i7s in gaming performance and BLCK/Turbo overclocking performance, although not in power efficiency (still, it's better than before). It's not perfect, but it's quite something. Vishera seems like it has a significant enough performance increase to let it challenge the K edition CPUs in the same method of usage.

Also, for highly threaded performance, this is not necessary and would likely be detrimental to performance. Considering that the highly threaded performance is already very high on BD CPUs, they don't need much of a boost, although a scheduling fix would likely help and software could be optimized for BD as well, although that's a little less reasonable.

 

[A Bad Day]

[citation][nom]K2N hater[/nom]You could also say Intel CPUs are a joke because the integrated graphics don't play Crysis.[/citation]

I'm going to be nit-picky, define, "play".

When I see "play", I see that the software won't crash.

I think a player can get away with a 320p resolution with all eyecandy off.

But who the hell wants to play it anyways at such settings?

 

[blazorthon]

[citation][nom]A Bad Day[/nom]I'm going to be nit-picky, define, "play".When I see "play", I see that the software won't crash.I think a player can get away with a 320p resolution with all eyecandy off.But who the hell wants to play it anyways at such settings?[/citation]

TBF, gaming on an HD 2000/3000 as graphics can cause crashes or simply not work with some DX applications such as some games, especially if you're using OEM drivers.

 

[verbalizer]

FX-6100 or FX-8120. Disable one core per module to get three or four single core modules, three for the 6100 and four for the 8120. You get a roughly 25% boost in performance at the same frequency as you previously had with all modules sharing their resources between two cores instead of a single core per module. That puts them a little ahead of Phenom II in performance per Hz while making them capable of significantly higher frequencies while using about the same or even somewhat lower amounts of power.

It also gives them even higher overclocking headroom to the point where they can easily compete with non-K edition i5s and i7s in gaming performance and BLCK/Turbo overclocking performance, although not in power efficiency (still, it's better than before). It's not perfect, but it's quite something. Vishera seems like it has a significant enough performance increase to let it challenge the K edition CPUs in the same method of usage.

Also, for highly threaded performance, this is not necessary and would likely be detrimental to performance. Considering that the highly threaded performance is already very high on BD CPUs, they don't need much of a boost, although a scheduling fix would likely help and software could be optimized for BD as well, although that's a little less reasonable.

I rather have a better CPU then to go through all the 'trickeration' you speak of.
it's a FAIL chip no matter what spin you attempt to put on it.

 

[A Bad Day]

[citation][nom]blazorthon[/nom]TBF, gaming on an HD 2000/3000 as graphics can cause crashes or simply not work with some DX applications such as some games, especially if you're using OEM drivers.[/citation]

Yeah, there are three kinds of drivers I hate:

1. OEM GPU drivers. Rarely updated while the Nivida/AMD drivers get all of the compatibility, stability and performance goodies.

2. Specific OS-only driver. Nope, that Vista driver for that printer will not work with Windows 7. And there won't be a W7 driver for the printer.

3. Bloatware driver. Requires 500 MB of useless shovelware (such as bloated customer support, automatic product sales update, etc) to install and use a 100 KB driver.

 

[markem]

respect to intel who are competing with AMD on integrated GPU & laying on the massive pressure, If AMD does
not achieve at-least 50/60% gpu performance, its good bye AMD.. Intel isn't playing around like AMD is, by not scavenging its desktop gpu sales, So minor updates from AMD wont work anymore with haswell.

AMD is at an disadvantage from 28nm vs 22nm but AMD still shouldn't be losing on the GPU front as intel wont on the CPU front... as soon as one takes GPU/CPU lead in both, its game over for the other

 

[palladin9479]

FX-6100 or FX-8120. Disable one core per module to get three or four single core modules, three for the 6100 and four for the 8120. You get a roughly 25% boost in performance at the same frequency as you previously had with all modules sharing their resources between two cores instead of a single core per module. That puts them a little ahead of Phenom II in performance per Hz while making them capable of significantly higher frequencies while using about the same or even somewhat lower amounts of power.

It also gives them even higher overclocking headroom to the point where they can easily compete with non-K edition i5s and i7s in gaming performance and BLCK/Turbo overclocking performance, although not in power efficiency (still, it's better than before). It's not perfect, but it's quite something. Vishera seems like it has a significant enough performance increase to let it challenge the K edition CPUs in the same method of usage.

Also, for highly threaded performance, this is not necessary and would likely be detrimental to performance. Considering that the highly threaded performance is already very high on BD CPUs, they don't need much of a boost, although a scheduling fix would likely help and software could be optimized for BD as well, although that's a little less reasonable.

There's another performance enhancement that you can do. There is a little known piece of software called PSCheck, it's AMD's tool to control the p-stats and voltages of each core / module on a CPU. Basically K10stat for Bulldozer uArch CPUs.

Now a little bit of info first, everything starting with the K10 arch and forward has multiple multipliers / voltages for each CPU core. That number in the BIOS only represents the initial speed for everything upon system boot, its the worst place to do overclocking (best place for voltages / enable cores). Using special software you can manipulate the exact multipliers / voltages of each of the p-stats on a CPU and then force different cores into different p-states.

The above means you can overclock just two cores while under-clocking / under-volting all unused cores to save TDP. This gives you a higher headway then what you'd reach trying to simultaneously overclock everything at once. It also means you can get ridiculously better single ~ dual core performance increase's.

Things to note,
#1 Windows is schizophrenic about task scheduling. It'll keep moving a thread around the CPU to whatever core it sees as "unused", in practice this is what prevents turboboost and other dynamic overclocking from engaging. Windows is preventing the other cores from going idle and thus enabling the overhead for the overclock. You gotta be smarter then windows is, use the processor affinity flag to force your program to work on the cores your going to overclock, this will let the other cores go idle.

#2 Other power management programs will often try to "outsmart" you and mess up your dynamic overclocking efforts. Disable all additional power management programs, your taking direct control over your CPU's speed.

Results:
I've got my 3550MX (2.0 Ghz stock, 2.7 boost) to run at 3.0 Ghz on up to two cores while running 800Mhz on the other two. Also got it to run 2.2~2.5Ghz on all four simultaneously though it gets very hot doing so. This resulted in a very large increase in benchmark numbers beating out many desktop CPU's at single threaded tasks. Amazing for a 45w CPU without any L3.

 

[palladin9479]

Also guys there are two sides to a GPU. One is the hardware, GPU's are just giant SIMD arrays that crunch lots of simple math. Intel is rapidly gaining parity with ATI / NVidia on this front, their still a year+ away but they will get there. The second is the software drivers, this is an area that Intel doesn't have much experience in (graphics drivers for 3D acceleration). History is littered with companies who made good hardware but messed up the software drivers, by the time they got the drivers to work their product was obsolete. Drivers are not something you can suddenly create through brute manpower, it takes time to release, have problems reported and subsequently fixed then released again. You have to work with other software companies to ensure your drivers include accelerators and profiles for those products. ATI / Nvidia have massive profile database's they've built over a decade+ of work. Intel is still years away from being close to those guys in the driver department.

So 1+ year until hardware parity, 3+ years until software parity. And while it's the "thing" to hate on AMD, their graphics units are made by ATI. AMD didn't dissolve ATI, they kept their divisions and teams intact, only the logo / name was replaced. This also explains why each generation of APU seems to use the previous generation of GPU's. It takes time from when ATI finish's the GPU for release and AMD is able to modify it to be coupled with their CPUs.

*Note*
I use ATI to refer to the graphics division of AMD which is run separately from their CPU division.

 

[palladin9479]

Here is a link to PSCheck and a quick guide to using it. Mind you these are all LN2 overclockers so their instructions are tailored for insane overclocks.

http://www.xtremesystems.org/forums/showthread.php?276190-AMD-PSCHECK-for-Bulldozer-UPDATE-NEWER-VERSION-amp-INSTRUCTION-HOW-TO-USE-IT

A better guide on how to use PSCheck, though these guys are a bit extreme.

http://www.overclockers.com/forums/showthread.php?t=691683

Now remember these are with LN2 so unrealistic, but 5Ghz+ on air is possible. You won't get all 8 "cores" running at 5Ghz, disable every other core and you could easily get two "cores" boosted to 5ghz while the other two are set to idle speeds. As most "games" we use rarely use more then two to four cores, this shouldn't be an issue.

7.7Ghz on a FX-6100
http://hwbot.org/submission/2221454_

8.0Ghz on a GX-8120
http://hwbot.org/submission/2222100_i.m.o.g._cpu_frequency_fx_8120_8060.51_mhz

 

[blazorthon]

There's another performance enhancement that you can do. There is a little known piece of software called PSCheck, it's AMD's tool to control the p-stats and voltages of each core / module on a CPU. Basically K10stat for Bulldozer uArch CPUs.

Now a little bit of info first, everything starting with the K10 arch and forward has multiple multipliers / voltages for each CPU core. That number in the BIOS only represents the initial speed for everything upon system boot, its the worst place to do overclocking (best place for voltages / enable cores). Using special software you can manipulate the exact multipliers / voltages of each of the p-stats on a CPU and then force different cores into different p-states.

The above means you can overclock just two cores while under-clocking / under-volting all unused cores to save TDP. This gives you a higher headway then what you'd reach trying to simultaneously overclock everything at once. It also means you can get ridiculously better single ~ dual core performance increase's.

Things to note,
#1 Windows is schizophrenic about task scheduling. It'll keep moving a thread around the CPU to whatever core it sees as "unused", in practice this is what prevents turboboost and other dynamic overclocking from engaging. Windows is preventing the other cores from going idle and thus enabling the overhead for the overclock. You gotta be smarter then windows is, use the processor affinity flag to force your program to work on the cores your going to overclock, this will let the other cores go idle.

#2 Other power management programs will often try to "outsmart" you and mess up your dynamic overclocking efforts. Disable all additional power management programs, your taking direct control over your CPU's speed.

Results:
I've got my 3550MX (2.0 Ghz stock, 2.7 boost) to run at 3.0 Ghz on up to two cores while running 800Mhz on the other two. Also got it to run 2.2~2.5Ghz on all four simultaneously though it gets very hot doing so. This resulted in a very large increase in benchmark numbers beating out many desktop CPU's at single threaded tasks. Amazing for a 45w CPU without any L3.

This is exactly the sort of proactive engagement into actually learning about how to fully utilize what we have that I enjoy. People can go about mocking either company, but there are all sorts of performance tricks that can have tremendous results. With AMD, they seemed to have screwed up pretty badly with Bulldozer, but it simply needs a little more work than other CPUs did in the past to let it shine. Since people are lazy, this get's overshadowed by it's stock performance and other such metrics even among the enthusiasts. I am convinced that although this is a fair way of looking at things for the people whom don't know any better about this topic, it's inexcusable for any true enthusiast to just pretend that this stuff doesn't matter. The FX CPUs can be made to be much faster for most situations than they are now while being more power efficient too. People can whine and/or complain all they want to, but this is an unavoidable fact. With Vishera, this will likely make them more than competitive with Intel's K edition CPUs and probably let them keep up with Haswell's higher end multiplier-locked CPUs.

 

[blazorthon]

I rather have a better CPU then to go through all the 'trickeration' you speak of.
it's a FAIL chip no matter what spin you attempt to put on it.

It only takes a few minutes to do this. It's no different from needing to overclock to get great performance. Most of use here at Tom's have had at least some experience with overclocking and I'd have to assume that you have, but you object to something that helps even more than overclocking and is easier to do?

 

[Bricktop]

respect to intel who are competing with AMD on integrated GPU & laying on the massive pressure, If AMD does
not achieve at-least 50/60% gpu performance, its good bye AMD.. Intel isn't playing around like AMD is, by not scavenging its desktop gpu sales, So minor updates from AMD wont work anymore with haswell.

AMD is at an disadvantage from 28nm vs 22nm but AMD still shouldn't be losing on the GPU front as intel wont on the CPU front... as soon as one takes GPU/CPU lead in both, its game over for the other

AMD has some time left. They are getting decent attention in the server market, and their desktop/mobile consumer market share hasn't got much smaller over the last few years.

Will they be able to compete for the CPU crown again? Doubtful. They may even lose the IGP crown in 3 years (Intel's SkyLake) or less. But, because they are more affordable than Intel and because their IGP can boost the performance of AMD discrete graphics, both Microsoft and Sony are looking at AMD for APU's and graphics cards in the next-gen gaming consoles. Dominating the gaming console market would be a jab at both Intel and Nvidia. It would reestablish AMD's legitimacy in the CPU/APU market, which would be followed by better press coverage. The future of AMD really hinges on getting contracts for the next-gen gaming consoles.

 

[army_ant7]

@palladin9479 and blaz:
You two, and some others, are doing a great favor for AMD as it seems. I applaud you two. And in addition to, and to complement, what I've learned from blaz, the things you've shared are very interesting and helpful, so you have my thanks palladin9479. 🙂 You two, just keep on rocking on!