AMD CPU speculation... and expert conjecture

[blackkstar]

No one is mentioning the political side of all these APIs either.

DirectX: Microsoft doesn't care about desktop users anymore judging by what they've done with Windows 8. It took massive screaming and kicking from desktop users to get a button that opens up a start screen. After endless whining about wanting the start menu back. People want old features back so badly that third parties have gone far out of their way to create their own replacements. But my point is that Microsoft no longer cares about us. They see heaps of money in phones and tablets and they see nothing in the HEDT crowd.

OpenGL: Governed by the lethargic Khronos Group. Trying to get anything done with OpenGL is a disaster. Do you want a new extension? Get ready to whine and wait a year before it becomes a part of OpenGL. Khronos group also does a very poor job of keeping OpenGL spec clean. It is a cluttered messed with 5 different ways to do the same thing.

Third party APIs: They are definitely going to cause fragmentation and create problems, but as far as OpenGL and DirectX are concerned, it's almost necessary at this point. DirectX is preferred for its ease of use and short development times over OpenGL, but it's tied to Windows. Some game developers are rather scared after seeing what Microsoft did with Windows 8 and how they treated HEDT customers. Staying purely DirectX for the future means you're going to be at Microsoft's whim, and if they decide that Windows 9 should be even more focused on mobile with HEDT as an afterthought, you're going to want an alternative to DirectX to escape Windows.

Valve is very forward thinking about this. Everyone who is supporting Mantle is also forward thinking about this as well.

But relying on AMD was sort of a last resort. Game developers wanted something that didn't suck to write like OpenGL yet wasn't tied down to Microsoft's whims of "Windows 7 is very popular but lets scrap the whole Aero desktop environment and make something for touch screens!" The people who wanted out of OpenGL disaster and Microsoft lock-in approached Nvidia and Intel. Some of them just wanted DirectX to not be forgotten and ignored, so they went to Microsoft.

The screams for help fell on deaf ears. AMD was the only one who really listened. I'm presuming Microsoft doesn't want to throw resources at a new version of DirectX. PC gaming doesn't help them with their goal of taking over the living room. And I would have assumed that if DX12 was a priority for Microsoft, that they would have made efforts to make sure Xbone supported DirectX. DirectX 11 is nearly 5 years old by now, they should have had time to work on it beforehand. Not to mention that DirectX had a year or two update cycle before it fell off the deep end and got ignored with DirectX 11.1

Things will get fragmented though. Nvidia has absolutely zero interest in creating things for gamers that can't be locked down on Nvidia hardware exclusively.

It sucks but our options at this point are:

1. Be stuck at Microsoft's whims when they've demonstrated that HEDT PC gaming is not a priority for them anymore.
2. Use the train wreck that is OpenGL
3. Fracture hardware vendors with software layers.

3 seems the best to me. At least it opens up some options (like OSX and Linux gaming) while not being a disaster. I also am very fond of the folks who design the GPUs being in charge of the software API that runs on that hardware as opposed to it being some obscure group that knows nothing of GPU design.

I'd envision at some point game developers could say "wow, I wish the GPU did this!", AMD/Nvidia will go "there's not a good way to do this in software" and then the next GPU family will have support for that.

EDIT: Case in point, look at some of the options Mantle is opening up. It is making AFR obsolete and opening up doors like allowing several different GCN clusters to do different things. That's something I don't think Khronos or Microsoft are capable of designing, simply because they lack the proximity to the hardware and they function as large, lethargic groups that only do things when AMD or Nvidia whine enough to them.

 

[blackkstar]

And the ARM 1 up-man-ship continues.

http://semiaccurate.com/2014/06/03/cavium-thunder-x-ups-arm-core-count-48-single-chip/

That has been my fear for AMD all along as they dabble with standard cores while these other companies have been at it for years and have custom cores. And they have a reputable motherboard vendor Gigabyte making the boards.

If a company like Facebook or Google really want to go all in on ARM will they choose a 48 core or an 8 core? Go big or go home right?

I'm really loving what ARM is doing to Intel in these markets. Intel marketing spent years laughing at the whole "MOAR COARS" thing and "MOAR COARS" is cleaning Intel's clock, and it looks like it's going to get even worse in servers.

 

[The Q6660 Inside]

MSi is teasing a new AMX board at computex, you can clearly see the lack of a square cut in the middle, and spread out pin covers as an indication of the AM3+ socket layout. From what I see, it screams "MICRO ATX AMD FX BOARD".

Edit: Here we go: http://www.thinkcomputers.org/sapphire-unveils-atomic-990fx-motherboard/

 

[gamerk316]

It sucks but our options at this point are:

1. Be stuck at Microsoft's whims when they've demonstrated that HEDT PC gaming is not a priority for them anymore.
2. Use the train wreck that is OpenGL
3. Fracture hardware vendors with software layers.

3 seems the best to me. At least it opens up some options (like OSX and Linux gaming) while not being a disaster. I also am very fond of the folks who design the GPUs being in charge of the software API that runs on that hardware as opposed to it being some obscure group that knows nothing of GPU design.

But here's the problem: Each GPU vendor would introduce their own technologies and API's. So one of the following will happen:

1: Devs will code to the lowest common denominator
2: Devs will code to whoever gives them the most financial incentive
3: Devs will code for the majority GPU maker
4: Devs will code for whoever has the easiest to develop development studio

See the problem? Three of the outcomes favor NVIDIA over everyone else. If it comes down to a war of cash, API's, Development Suite, or Market Share, NVIDIA wins hands down. That's why AMD had best pray they haven't started a major API war, because they don't have the resources to win such a war.

 

[colinp]

Interesting that there are a few new AM3+ designs coming out all of a sudden. They wouldn't be doing it if there wasn't a market for them or something coming. If an mATX 9-series board comes along, then I'll pick the cobwebs off of my wallet for one.

 

[Cazalan]

And the ARM 1 up-man-ship continues.

http://semiaccurate.com/2014/06/03/cavium-thunder-x-ups-arm-core-count-48-single-chip/

That has been my fear for AMD all along as they dabble with standard cores while these other companies have been at it for years and have custom cores. And they have a reputable motherboard vendor Gigabyte making the boards.

If a company like Facebook or Google really want to go all in on ARM will they choose a 48 core or an 8 core? Go big or go home right?

I'm really loving what ARM is doing to Intel in these markets. Intel marketing spent years laughing at the whole "MOAR COARS" thing and "MOAR COARS" is cleaning Intel's clock, and it looks like it's going to get even worse in servers.

They do have moar coars coming. Haswell 18 core (32 thread), Phi 62/72 core. They won't be cheap of course. $2500-$5000 a pop.

 

[blackkstar]

But here's the problem: Each GPU vendor would introduce their own technologies and API's. So one of the following will happen:

1: Devs will code to the lowest common denominator
2: Devs will code to whoever gives them the most financial incentive
3: Devs will code for the majority GPU maker
4: Devs will code for whoever has the easiest to develop development studio

See the problem? Three of the outcomes favor NVIDIA over everyone else. If it comes down to a war of cash, API's, Development Suite, or Market Share, NVIDIA wins hands down. That's why AMD had best pray they haven't started a major API war, because they don't have the resources to win such a war.

I do see the problem. But I don't see a good solution. Microsoft doesn't care about us anymore. They've basically stood and watched Apple go from a joke to one of the most valuable tech companies by selling fancy phones and tablets.

There is no good future. Our future is either fragment and figure something out or let MS and Khronos screw us over.

The problem is that Nvidia has already expressed that they don't care about an alternative API. They have been beating the drum for DirectX improvements (with dubious marketing slides) and talking about how OpenGL can be as good as Mantle when it comes to performance.

But none of those situations address the primary problems with DirectX and OpenGL. It seems like a lot of people are thinking that Mantle is just free FPS. It is free FPS, but the original intent of Mantle is to escape Microsoft's vendor lock in and their attitude toward PC gaming as well as OpenGL's vast problems.

AMD tried to address those problems by making going Mantle as easy as possible to port to from DirectX. They are hoping that DirectX will remain somewhat dominant for a while and Mantle will exist just as a cheap supplementary thing.

I find Star Swarm to be an example of where Mantle will take us. It will eventually start offering features that you can't find in DirectX. Star Swarm allows for far more units to be on screen with Mantle because it leverages Mantle's strong points.

Right now games like BF4 are just using it as extra performance for free.

I'd imagine that Mantle, if it sticks around, will eventually warp rather far from DirectX.

Which is leaving me thinking that the future we're going to see will be game developers pushing Nvidia to support Mantle or something compatible.

But I do think all of that is a long ways away, at least a few years.

We needed something to take us away from relying on Microsoft and the direction they are taking Windows in. It has definitely freaked out a lot of developers who don't want to be stuck making Android and iOS games for the rest of their careers.

There is no solution. I don't think that 4 years ago anyone would have expected Microsoft to abandon PC gamers and the traditional desktop market like they have and they wouldn't have expected OpenGL to be such a nasty mess right now.

But this whole SteamOS/Linux/Mantle/etc push is a lot of people getting really freaked out about Microsoft releasing Windows 9 with a unified tablet/desktop/phone user environment and being complete crap for keyboard and mouse.

To be honest with you, I would have been happy to see Nvidia embrace it first and then AMD to tag along. Even if it meant the new API being a huge performance blow to AMD, it would have meant a better API and no more reliance on Windows for gaming.

 

[WarCrysis]

I hate hearing peoples opinions on stuff like this API business. Sure openGL could be better then all of them.....

but DX is the standard. AMD is the ones changing the game and making msft pay attention again..

like openGL even stands a chance.. Tell me.. Who's going to invest the money in to open GL to make it a better option?? Its like ea making games for the Wii u.. They won't make there money back..

but then there is retards like some people that are like oh if they did this or this or this they would be so much better.......... Yea and if Obama got rid of this health care mess it would be great. But really. Your smarter then that. Dx has been unchallenged till mantle.. Look how fast they respond... They do care, but tell me. If its not worth the investment would you do with your own money?? The answer is no.

thank you "if" people. Come back to reality

 

[WarCrysis]

Btw this a AMD thread so stfu with this crap

 

[juanrga]

And the ARM 1 up-man-ship continues.

http://semiaccurate.com/2014/06/03/cavium-thunder-x-ups-arm-core-count-48-single-chip/

That has been my fear for AMD all along as they dabble with standard cores while these other companies have been at it for years and have custom cores. And they have a reputable motherboard vendor Gigabyte making the boards.

If a company like Facebook or Google really want to go all in on ARM will they choose a 48 core or an 8 core? Go big or go home right?

ARM license two 64bit Cortex cores: A57 (latency optimized) and A53 (throughput optimized). The A57 scales up to a 16-cores on die. The A53 cores scales up to what ARM calls the "sea of cores": 64-cores on die.

Why is not AMD releasing a chip with lots of A53 cores?

AMD is using small number of A57 cores because will use large number of GCN cores for throughput. In fact, my information is that the K12 core (also a latency optimized core) will scale up to 16 cores.

It is important to remark that Cavium is openly claiming that their new SoC is not aimed at micro-servers, but will compete against Intel Xeons on top-servers (Opteron share is almost inexistent). In fact the fastest Xeon CPU peaks about 0.6 TFLOP/s and this Cavium SoC must produce about 1 TFLOP/s. Apparently the top SoC will be rated at 95W TDP. Top Xeon CPUs alone are rated at ~130W. Thus Cavium Thunder-X would be much faster than the fastest x86 CPU today but consuming less power.

Other details of the Thunder-X are the 78K of L1-I and 32K of L1-D along with 16MB of L2 cache, support for quad-channel DDR4-2400MHz and up to a 1TB of DRAM on board.

 

[juanrga]

And no surprises... AMD releases mobile Kaveri, which includes the mentioned FX-branded APUs.

http://www.theinquirer.net/inquirer/news/2348053/computex-amd-launches-kaveri-apu-for-mobile-pcs-and-notebooks

 

[palladin9479]

I firmly believe that anyone in the IT industry should learn some ASM, just so they can fully understand exactly how CPU's (or any processor really) "see's" the world. It really gives a different point of view.

Anyhow, the whole OGL legacy issue isn't so much Kronos's fault but the result of who's involved. OGL is used in the CAD/CAM industry along with many non-PC / non-gaming related fields like science and simulations. Those industry's have spent an inordinate amount of money developing libraries and systems specific to that legacy OGL and they have zero interest in any new gaming technology. Kronos's can't remove that legacy code as it would break all those industrial applications and force an extremely large financial burden on it's members. That's why they developed OpenGL-ES instead for portable gaming and such. It's had a majority of the legacy code stripped out and from what I've been told is easier to program for. I expect to see it become more popular for general purpose gaming in the next five years or so.

Gamer about the standards war, those frequently happen and will always happen. They are necessary to ensure constant progress. You will get several vying standards and eventually the market will select one out, it will be a better over-all standard then it's competitors.

 

[de5_Roy]

mobile kaveri previews/first looks

AMD FX-7600P Kaveri Review: FX Rides Again...In A Mobile APU?
http://www.tomshardware.com/reviews/fx-7600p-kaveri-apu,3842.html

moar

anandtech
http://www.anandtech.com/show/8119/amd-launches-mobile-kaveri-apus
techreport
http://techreport.com/review/26528/a-first-look-at-amd-kaveri-apu-for-notebooks

after finishing the toms article, seems like reviewers weren't given much time to use the kaveri f.f.r.d.

 

[Fearnot]

And no surprises... AMD releases mobile Kaveri, which includes the mentioned FX-branded APUs.

http://www.theinquirer.net/inquirer/news/2348053/computex-amd-launches-kaveri-apu-for-mobile-pcs-and-notebooks

Yeah what you just posted was this countdown https://ad.moontoast.com/552fa6aa-e1bd-11e3-bc12-123139027524 . Everyone could tell it was mobile Kaveri with the clues.

http://ifitcanreachspace.com/ So is this countdown the new gpu's? Heard there is another countdown somewhere else also.

 

[colinp]

The reviews are starting to come in for mobile Kaveri.

http://www.tomshardware.co.uk/fx-7600p-kaveri-apu,review-32965.html

http://anandtech.com/show/8119/amd-launches-mobile-kaveri-apus

http://hexus.net/tech/reviews/cpu/70653-amd-fx-7600p-28nm-kaveri/?page=3

Usual comparisons against i7 laptops (News at 10! Kaveri gets smoked by an i7 in CPU benchmarks!), but Anand has some comparisons against Trinity and Richland chips, and it looks like a decent enough step up. Hexus even compares against desktop Kaveri, and it holds up pretty well, despite the much lower TDP, but you have to ignore that 15w i5...

Here's what I'd like to see when proper laptop reviews come out:

Battery life comparisons
Crossfire usage
Temp charts

And I'd also really love to see a quality design rather than a chunky, heavy, creaky design (cough! HP!).

 

[de5_Roy]

Usual comparisons against i7 laptops (News at 10! Kaveri gets smoked by an i7 in CPU benchmarks!), but Anand has some comparisons against Trinity and Richland chips, and it looks like a decent enough step up.

amd wants reviewers to compare mobile kaveri with mobile i7. the point is to show how much of mobile i7's performance kaveri delivers and how favorably mobile kaveri compares to mobile i5. another one is to mock the hd4400-4600 igpu. notice that iris and iris pro comparisons are conspicuously absent, so were vce vs quick sync. this is a preview anyway, amd seems to have rushed the paper launch to counter haswell refresh and devil's canyon instead of polishing the software side (mantle, enduro, igpu etc).

 

[colinp]

amd wants reviewers to compare mobile kaveri with mobile i7. the point is to show how much of mobile i7's performance kaveri delivers and how favorably mobile kaveri compares to mobile i5. another one is to mock the hd4400-4600 igpu. notice that iris and iris pro comparisons are conspicuously absent, so were vce vs quick sync. this is a preview anyway, amd seems to have rushed the paper launch to counter haswell refresh and devil's canyon instead of polishing the software side (mantle, enduro, igpu etc).

Maybe the situation is different elsewhere, but aside from the Macbook Pro, laptops with Iris Pro are rarer than hen's teeth in the UK. Are they not also all 47W chips and stonkingly expensive?

 

[de5_Roy]

Maybe the situation is different elsewhere, but aside from the Macbook Pro, laptops with Iris Pro are rarer than hen's teeth in the UK.

possible. apple was the one who influenced intel into building hd3k-iris pro. some of the skus are shockingly under $400 - 2c/4t with iris/pro igpus.

my criticisms of amd's tactics aside, mobile kaveri has the highest potential since llano came out in 2011. i'd like to see fx7600p in gaming laptops like the one msi built. heck, fx7600p would be great fit for small steam boxes based on gigabyte's brix gaming barebones platform. another potential candidate would be zotac's small boxes, or sphere-thingies.

 

[colinp]

possible. apple was the one who influenced intel into building hd3k-iris pro. some of the skus are shockingly under $400 - 2c/4t with iris/pro igpus.

my criticisms of amd's tactics aside, mobile kaveri has the highest potential since llano came out in 2011. i'd like to see fx7600p in gaming laptops like the one msi built. heck, fx7600p would be great fit for small steam boxes based on gigabyte's brix gaming barebones platform. another potential candidate would be zotac's small boxes, or sphere-thingies.

Agree; fingers crossed the equipment makers also agree.

BTW, just noticed that you'd already posted some links to reviews just before me; apologies, I wasn't ignoring you, just sleepy!

 

[juanrga]

And no surprises... AMD releases mobile Kaveri, which includes the mentioned FX-branded APUs.

http://www.theinquirer.net/inquirer/news/2348053/computex-amd-launches-kaveri-apu-for-mobile-pcs-and-notebooks

Yeah what you just posted was this countdown https://ad.moontoast.com/552fa6aa-e1bd-11e3-bc12-123139027524 . Everyone could tell it was mobile Kaveri with the clues.

http://ifitcanreachspace.com/ So is this countdown the new gpu's? Heard there is another countdown somewhere else also.

As I mentioned before

Some more data about http://ifitcanreachspace.com/

When countdown hits zero the page will change to "AMD's most advanced APU ever. Welcome to Earth".

I also commented in another post about the new Tonga GPU (R9 M295X) for mobile and gave some link to specifications.

I believe that AMD could release a nice Steambox with Mobile Kaveri + R9 M295x in a very small form factor, similar to the hand sized Brix Gaming Box (mobile Richland A85557M + R9 M275x).

Seeing the potential of Steamroller in mobile, this makes one to ask why AMD dropped desktop Kaveri TDP from 100W to 95W and clocks from 4GHz to 3.7GHz. Of course, one can upgrade clocks again at 4GHz on K chips, but makes one think about the decision.

 

[de5_Roy]

when i started reading the tr article, i noticed this:
http://core0.staticworld.net/images/article/2014/06/kaveri-lineup2-100310098-orig.png
all the 19w kaveris have 8x gen 2 pcie lanes while 35w kaveris get full 16x gen 3.0 lanes. lower kaveris aren't getting eyefinity, dual gfx and trueaudio enabled. i sorta understand why dual gfx and eyefinity will be disabled, but why true audio. and what about the psp? from at article:

Besides offering different levels of performance, AMD is also differentiating their Kaveri APUs based on other features. Eyefinity and TrueAudio support will be limited to the A10 and FX APUs; the A6/A8 APUs lose this functionality. Similarly, the A6 does not have Dual Graphics functionality

ASRock Launches First Socket AM3+ Motherboard with M.2 Slot
http://www.techpowerup.com/201653/asrock-launches-first-socket-am3-motherboard-with-m-2-slot.html

Apacer Mashes a DDR3 DIMM with an M.2 SSD
http://www.techpowerup.com/201652/apacer-mashes-a-ddr3-dimm-with-an-m-2-ssd.html
yo dawg...

this is not amd-related but something this revolutionary needs posting. behold, DDR5 ram. on a gfx card, no less:
http://www.techpowerup.com/img/14-06-04/103b.jpg
did i say gfx card? make that two
http://www.techpowerup.com/img/14-06-04/104a.jpg
what's next, gskill claiming transfer rate and clockrate are the same thing...!?

Agree; fingers crossed the equipment makers also agree.

BTW, just noticed that you'd already posted some links to reviews just before me; apologies, I wasn't ignoring you, just sleepy!

nah, no problem about the links.

 

[gamerk316]

I firmly believe that anyone in the IT industry should learn some ASM, just so they can fully understand exactly how CPU's (or any processor really) "see's" the world. It really gives a different point of view.

Anyhow, the whole OGL legacy issue isn't so much Kronos's fault but the result of who's involved. OGL is used in the CAD/CAM industry along with many non-PC / non-gaming related fields like science and simulations. Those industry's have spent an inordinate amount of money developing libraries and systems specific to that legacy OGL and they have zero interest in any new gaming technology. Kronos's can't remove that legacy code as it would break all those industrial applications and force an extremely large financial burden on it's members. That's why they developed OpenGL-ES instead for portable gaming and such. It's had a majority of the legacy code stripped out and from what I've been told is easier to program for. I expect to see it become more popular for general purpose gaming in the next five years or so.

Gamer about the standards war, those frequently happen and will always happen. They are necessary to ensure constant progress. You will get several vying standards and eventually the market will select one out, it will be a better over-all standard then it's competitors.

http://www.xkcd.com/927/

 

[Ags1]

Anyhow, the whole OGL legacy issue isn't so much Kronos's fault but the result of who's involved. OGL is used in the CAD/CAM industry along with many non-PC / non-gaming related fields like science and simulations. Those industry's have spent an inordinate amount of money developing libraries and systems specific to that legacy OGL and they have zero interest in any new gaming technology. Kronos's can't remove that legacy code as it would break all those industrial applications and force an extremely large financial burden on it's members. That's why they developed OpenGL-ES instead for portable gaming and such. It's had a majority of the legacy code stripped out and from what I've been told is easier to program for. I expect to see it become more popular for general purpose gaming in the next five years or so.

OpenGL has had two alternate pipelines for some time - the legacy pipeline dreamed up in the dark ages of the '90s, and the programmable pipeline. OpenGL ES strips away the legacy stuff and is massively successful. Of all the posters bashing OpenGL here, who has actually programmed with it?

Anyway, it's great to see mobile Kaveri launch, but I was still hoping for GDDR5 support, so I'm a little disappointed.

 

[cdrkf]

Anyhow, the whole OGL legacy issue isn't so much Kronos's fault but the result of who's involved. OGL is used in the CAD/CAM industry along with many non-PC / non-gaming related fields like science and simulations. Those industry's have spent an inordinate amount of money developing libraries and systems specific to that legacy OGL and they have zero interest in any new gaming technology. Kronos's can't remove that legacy code as it would break all those industrial applications and force an extremely large financial burden on it's members. That's why they developed OpenGL-ES instead for portable gaming and such. It's had a majority of the legacy code stripped out and from what I've been told is easier to program for. I expect to see it become more popular for general purpose gaming in the next five years or so.

OpenGL has had two alternate pipelines for some time - the legacy pipeline dreamed up in the dark ages of the '90s, and the programmable pipeline. OpenGL ES strips away the legacy stuff and is massively successful. Of all the posters bashing OpenGL here, who has actually programmed with it?

Anyway, it's great to see mobile Kaveri launch, but I was still hoping for GDDR5 support, so I'm a little disappointed.

I'm pretty sure I read the other day some suggestions that there is a mobile variant of Kaveri planned with quad channel DDR3 memory (it would certainly make sense as the APU is horribly memory constrained). It was probably just rumours though unfortunately.

 

[de5_Roy]

I'm pretty sure I read the other day some suggestions that there is a mobile variant of Kaveri planned with quad channel DDR3 memory (it would certainly make sense as the APU is horribly memory constrained). It was probably just rumours though unfortunately.

for quad channel at least 4 so-dimm slots woulda been needed. coulda diminished mobile kaveri's value proposition. amd didn't reveal mobile kaveris' pricing details, though.
gddr5, otoh, woulda been awesome. treat the igpu as discreet gfx with it's own high bandwidth vram - coulda erased memory bottleneck.