AMD Technology Announcements

[GloriousPain]

2016-2017, and that's dependent on Intel not catching up with good leaps in iGPU in Skylake and Cannonlake. Carrizo isn't going to be that amazing, leaving Zen to be AMD's redeeming chip, but production on Zen will not begin until late 2015 at the earliest. Samsung's 16/14nm tech has never built anything more complicated than NAND flash and the controllers for their SSDs. Their phone chips are still built on 20nm tech.

Sure, AMD is getting FreeSync out there, but that was a given.

There's still there graphics chips like Fiji which could also give them quite the boost as @ONOTech didn't specify cpu only just year of glory

[patrickjp93]

There's still there graphics chips like Fiji which could also give them quite the boost as @ONOTech didn't specify cpu only just year of glory

Yeah I might have been enjoying my whiskey a bit too much when I wrote that gem.

 

I see AMD pulling ahead of Nvidia in graphics, but on the CPU side and for massively parallel acceleration, I don't see them coming close to Intel for 2-3 years.

[mr moose]

It's impossible to offend me. Carry on. 

Really?

 

 

Marijuana should stay illegal because it causes premature death, increases the crime rate and results in schizophrenia.     :lol: 

[-rascal-]

Too bad AMD is literally the absolute shit company of the millennium with nothing of any sort of substantial value to add to the world and shitty, ultra-freaking-horrible product after product to showcase... the absolute definition of "Shit Company". The only reason such a pathetic corporation is allowed to exist is the single fact of it saving Intel from having to be subjected to federal Anti-trust break-up.. only nubs buy AMD:

 

http://www.pugetsystems.com/labs/articles/Video-Card-Failure-Rates-by-Generation-563/

 

Some old history:

1. It was AMD who started the multi-core technology on CPUs. if it wasn't for them, we could be still stuck with single-core Processors.
2. AMD had first implemented 64-bit computing on Processors in 2003. Intel followed suite in 2004.

If AMD is shit, they wouldn't be alive and celebrating their 45th anniversary earlier this year. They would be long gone like Cyrix.

[Opcode]

Consoles already have a low level api so what is the point of bringing mantle?

Portability. Games can be ported to and from the desktop and essentially support Mantle out of the box.

 

2016-2017, and that's dependent on Intel not catching up with good leaps in iGPU in Skylake and Cannonlake. Carrizo isn't going to be that amazing, leaving Zen to be AMD's redeeming chip, but production on Zen will not begin until late 2015 at the earliest. Samsung's 16/14nm tech has never built anything more complicated than NAND flash and the controllers for their SSDs. Their phone chips are still built on 20nm tech.

Sure, AMD is getting FreeSync out there, but that was a given.

Carrizo is the fist fully compliant HSA architecture. Not to mention it's 30% boost in IPC and switch to the latest GCN 1.2 architecture. It's a huge step forward for AMD. They have been holding their crown in front of Intel and I doubt that will change any time soon. By the time Zen comes out their APU's will have fusion hubs on die, possibly HBM, full HSA support, and more. Intel's far behind on a lot of things. The race for being the IPC king has backfired on them when it comes to today's demands.

[deathjester]

Too bad AMD is literally the absolute shit company of the millennium with nothing of any sort of substantial value to add to the world and shitty, ultra-freaking-horrible product after product to showcase... the absolute definition of "Shit Company". The only reason such a pathetic corporation is allowed to exist is the single fact of it saving Intel from having to be subjected to federal Anti-trust break-up.. only nubs buy AMD:

 

http://www.pugetsystems.com/labs/articles/Video-Card-Failure-Rates-by-Generation-563/

 

Literally the silliest post I have ever seen on LTT. Grats sir.

 

Welcome to PC Gaming. Nvidia is a new comer and was trash compared to 3DFX as far as gaming GPU's back then (it was 3DFX or bust for a few years) and that company is now gone. Nvidia is what the PC Gamers with no money used. They sucked. ATI is an OLD company, before it was bought from AMD and made really great autocad cards which we used for gaming before 3DFX came, with companies fighting it out for second place which included Matrox.  Nvidia eventually came out of that field with ATI.

 

AMD was also highly influential in the CPU's we have today, and DESTROYED Intel's Pentium 4 which was a complete piece of crap. We all gamed on AMD CPU's back then (if you were smart). We may be gaming on Intel CPU's now, but again, things can change. Intel might also be still be chasing the mobile market, which means AMD may very well pull equal or ahead on CPU's for gaming.

 

GPU's? They have flip flopped back and forth for years and a good brand R9 290x and a GTX 980 are pretty much dead even at 4k with the AMD card usually winning due to bandwidth. Nvidia is ahead on reference cards and power draw due to die size and that is about to change. The 780Ti's also had very high VRM temps and were also power hogs. It was not until the 970/980 that Nvidia was really that far ahead on power usage. AMD's REFERENCE r9 290/x was bad, but their aftermarket cards are still as good as it gets for 4k.

 

BTW I see you love Star Citizen. My R9 290 will be rocking in that game with Mantle. It will be CPU bound as hell per Chris Roberts. Maybe Nvidia will help you out and take up Mantle now that they made it public. Otherwise? Enjoy those FPS lows.

 

But yeah. This "nub" is rocking GTX 980 performance in most games on an OC. My GPU was 290 bucks, and runs very cool. Couldn't be happier. 

[That Norwegian Guy]

I hope AMD does well soon, but I doubt it. It's going to take a slow climb to get back, they're competing on two fronts on which both main competitors are outclassing them by a large (or even huge) margin (as I'm sure both Nvidia and Intel are holding back a lot)

 

If AMD were actually competing , we'd probably be 20 years ahead in technology advances by now, that's the amount of unused R&D I bet Nvidia and Intel are sitting on, and not releasing.

 

 

Too bad AMD is literally the absolute shit company of the millennium with nothing of any sort of substantial value to add to the world and shitty, ultra-freaking-horrible product after product to showcase... the absolute definition of "Shit Company". The only reason such a pathetic corporation is allowed to exist is the single fact of it saving Intel from having to be subjected to federal Anti-trust break-up.. only nubs buy AMD:

 

http://www.pugetsystems.com/labs/articles/Video-Card-Failure-Rates-by-Generation-563/

 

First and last post to you: get the fuck out.

[patrickjp93]

Portability. Games can be ported to and from the desktop and essentially support Mantle out of the box.

Carrizo is the fist fully compliant HSA architecture. Not to mention it's 30% boost in IPC and switch to the latest GCN 1.2 architecture. It's a huge step forward for AMD. They have been holding their crown in front of Intel and I doubt that will change any time soon. By the time Zen comes out their APU's will have fusion hubs on die, possibly HBM, full HSA support, and more. Intel's far behind on a lot of things. The race for being the IPC king has backfired on them when it comes to today's demands.

We'll see. Broadwell is Intel's Kaveri, minus AMD's advantage in GPU architecture for games. Broadwell also has a massive compute advantage which Skylake will continue to build on. As per Carrizo jumping up 30% in IPC, I reserve judgment until we start seeing a wide array of benchmarks.

[Psykomantis00]

Really?

 

 

Marijuana should stay illegal because it causes premature death, increases the crime rate and results in schizophrenia.     :lol:

Har har. 

[Opcode]

We'll see. Broadwell is Intel's Kaveri, minus AMD's advantage in GPU architecture for games. Broadwell also has a massive compute advantage which Skylake will continue to build on. As per Carrizo jumping up 30% in IPC, I reserve judgment until we start seeing a wide array of benchmarks.

One of the last things most consumers in the market today are interested in is compute performance. Even tho Kaveri already has GT3e's cards. 

 

 

 

 

 

 

Gen8 might be able to catch up to Kaveri tho Carrizo will be released around the same time bringing even more competition to Intel. I personally have my doubts that Intel will be focusing as much on their GPU architecture after Broadwell. Intel's playing catch-up in the SoC race. Carrizo will have its fusion hub built on die in which Intel is not expecting to do until Skylake. Intel's die size is much larger than Kaveri/Carrizo's as well as being on a smaller node. I'm waiting to see Carrizo's successor with die shrinks. I would assume 640 SPU's with HBM or DDR4 would be the next step for them.

[patrickjp93]

One of the last things most consumers in the market today are interested in is compute performance. Even tho Kaveri already has GT3e's cards. 

 

 

Gen8 might be able to catch up to Kaveri tho Carrizo will be released around the same time bringing even more competition to Intel. I personally have my doubts that Intel will be focusing as much on their GPU architecture after Broadwell. Intel's playing catch-up in the SoC race. Carrizo will have its fusion hub built on die in which Intel is not expecting to do until Skylake. Intel's die size is much larger than Kaveri/Carrizo's as well as being on a smaller node. I'm waiting to see Carrizo's successor with die shrinks. I would assume 640 SPU's with HBM or DDR4 would be the next step for them.

Carrizo will also be competing against Gen 9 in Skylake. Bear that in mind. Also, no, Kaveri is in the 500 Gflops range whereas Iris Pro 5200 is 832 Gflops, and the 5770k with Iris Pro 6200 is promising a whopping teraflop. Also, consumers are stupid. It's up to developers to use it. Also, we have yet to see Broadwell's die size, and you still forget Intel has 4 ALUs per core vs. AMD's 2, with a full FPU per ALU vs. AMD's current modular design. This is why AMD gets its ass kicked.

[spartaman64]

Carrizo will also be competing against Gen 9 in Skylake. Bear that in mind. Also, no, Kaveri is in the 500 Gflops range whereas Iris Pro 5200 is 832 Gflops, and the 5770k with Iris Pro 6200 is promising a whopping teraflop. Also, consumers are stupid. It's up to developers to use it. Also, we have yet to see Broadwell's die size, and you still forget Intel has 4 ALUs per core vs. AMD's 2, with a full FPU per ALU vs. AMD's current modular design. This is why AMD gets its ass kicked.

OMG newer and more expensive processors are faster ... holy shit

[Sharkyx1]

Does This mean PC will get Deep Down?

[Hatsune Miku 「 」]

Carrizo will also be competing against Gen 9 in Skylake. Bear that in mind. Also, no, Kaveri is in the 500 Gflops range whereas Iris Pro 5200 is 832 Gflops, and the 5770k with Iris Pro 6200 is promising a whopping teraflop. Also, consumers are stupid. It's up to developers to use it. Also, we have yet to see Broadwell's die size, and you still forget Intel has 4 ALUs per core vs. AMD's 2, with a full FPU per ALU vs. AMD's current modular design. This is why AMD gets its ass kicked.

Their "architecture" is not an architecture but rather a flawed design concept similar to the PS2 and PS3's cell processor. AMD's marketing term for it is CMT (madeup) or clustered multi-threading.

 

I'm not support AMD in anyway.

[Jerakl]

Too bad AMD is literally the absolute shit company of the millennium with nothing of any sort of substantial value to add to the world and shitty, ultra-freaking-horrible product after product to showcase... the absolute definition of "Shit Company". The only reason such a pathetic corporation is allowed to exist is the single fact of it saving Intel from having to be subjected to federal Anti-trust break-up.. only nubs buy AMD:

 

http://www.pugetsystems.com/labs/articles/Video-Card-Failure-Rates-by-Generation-563/

I cannot express how much of an idiot you are.

[Opcode]

Carrizo will also be competing against Gen 9 in Skylake. Bear that in mind. Also, no, Kaveri is in the 500 Gflops range whereas Iris Pro 5200 is 832 Gflops, and the 5770k with Iris Pro 6200 is promising a whopping teraflop. Also, consumers are stupid. It's up to developers to use it. Also, we have yet to see Broadwell's die size, and you still forget Intel has 4 ALUs per core vs. AMD's 2, with a full FPU per ALU vs. AMD's current modular design. This is why AMD gets its ass kicked.

Carrizo is due this holiday season which puts it up against Broadwell offerings that are coming out around the same exact time. Skylake will essentially compete with Skybridge. Which in my personal opinion Intel still will be playing catch-up as Skybridge will be a universal SoC featuring both Puma+ and Cortex-A57. Essentially one chip to rule every corner of the mobile market (where Intel doesn't have much ground). Also keep in mind tests done like shown in my previous post show Intel's architecture falling short of anything spectacular in compute (Anandtech is a reliable source). Theoretical compute performance (calculated via algorithm) doesn't always translate to real world compute performance. Intel's been known for throwing out invalid numbers before (Pentium 4) and it shows that they are at it again. Their IP 5200 is suppose to offer over 800 GFLOPS of compute power meanwhile in actual tests it only shows a fraction of Kaveri's capability. IP 6200 will be tested in compute and we can compare it with its competition Carrizo. AMD loses in the serial processing market we all already know that. Tho it's irrelevant to Intel getting their ass handed to them in compute capability. Intel's undoubtedly the king of IPC but AMD's the king of compute.

[maazster]

Carrizo will also be competing against Gen 9 in Skylake. Bear that in mind. Also, no, Kaveri is in the 500 Gflops range whereas Iris Pro 5200 is 832 Gflops, and the 5770k with Iris Pro 6200 is promising a whopping teraflop. Also, consumers are stupid. It's up to developers to use it. Also, we have yet to see Broadwell's die size, and you still forget Intel has 4 ALUs per core vs. AMD's 2, with a full FPU per ALU vs. AMD's current modular design. This is why AMD gets its ass kicked.

Kaveri is actually in the 700 Gflop range, with the 7850k.

[Nielips]

Hopefully over the next view years we will see more competition, which drives innovation, so consumers get better products. I would love to see more consumer CPU companies, but I doubt that will happen any times soon.

[patrickjp93]

Carrizo is due this holiday season which puts it up against Broadwell offerings that are coming out around the same exact time. Skylake will essentially compete with Skybridge. Which in my personal opinion Intel still will be playing catch-up as Skybridge will be a universal SoC featuring both Puma+ and Cortex-A57. Essentially one chip to rule every corner of the mobile market (where Intel doesn't have much ground). Also keep in mind tests done like shown in my previous post show Intel's architecture falling short of anything spectacular in compute (Anandtech is a reliable source). Theoretical compute performance (calculated via algorithm) doesn't always translate to real world compute performance. Intel's been known for throwing out invalid numbers before (Pentium 4) and it shows that they are at it again. Their IP 5200 is suppose to offer over 800 GFLOPS of compute power meanwhile in actual tests it only shows a fraction of Kaveri's capability. IP 6200 will be tested in compute and we can compare it with its competition Carrizo. AMD loses in the serial processing market we all already know that. Tho it's irrelevant to Intel getting their ass handed to them in compute capability. Intel's undoubtedly the king of IPC but AMD's the king of compute.

Compute tests show Intel to be the stronger iGPU. Graphics tests show AMD to be better. Furthermore, only Carrizo mobile is due out, and that's in January at the earliest. AMD won't launch desktop parts until late Q1 or Q2 at the earliest. That puts it on the heels of Skylake's launch.

Furthermore, Skybridge is not a universal SOC. It's a universal socket for accepting x86 and ARM SOCs.

[patrickjp93]

Kaveri is actually in the 700 Gflop range, with the 7850k.

Actual compute is nowhere close to its theoretical.

[FakezZ]

Too bad AMD is literally the absolute shit company of the millennium with nothing of any sort of substantial value to add to the world and shitty, ultra-freaking-horrible product after product to showcase... the absolute definition of "Shit Company". The only reason such a pathetic corporation is allowed to exist is the single fact of it saving Intel from having to be subjected to federal Anti-trust break-up.. only nubs buy AMD:

 

http://www.pugetsystems.com/labs/articles/Video-Card-Failure-Rates-by-Generation-563/

You just won the biased comment of the year award... well done. AMD might be lacking a bit when it comes to sheer performance, but price/performance ratio-wise it spanks intel in the butt...

[Opcode]

Compute tests show Intel to be the stronger iGPU. Graphics tests show AMD to be better. Furthermore, only Carrizo mobile is due out, and that's in January at the earliest. AMD won't launch desktop parts until late Q1 or Q2 at the earliest. That puts it on the heels of Skylake's launch.

Furthermore, Skybridge is not a universal SOC. It's a universal socket for accepting x86 and ARM SOCs.

I posted compute tests a few posts ago showing current GT3e being staggeringly abused by previous GCN architecture. Carrizo mobile will be available this holiday season with the desktop variant arriving H1 next year. As said with it launching at the same exact time as Broadwell so it would be just ludacris to compare it to a previous generation microprocessor. I've stated before (that you apparently want to go along with now) theoretical numbers don't mean anything. The 5960x has a theoretical peak performance of 384 GFLOPS and doesn't even live up to no wheres near that. Also Skybridge is a universal SoC sharing not only the same package but also both x86 and ARM cores on the same die. The same exact chip can be used in desktops, laptops, notebooks, tablets, cell phones, servers, etc and drive Windows, Linux, OS X, Android, iOS, etc. So its miles ahead of what Intel will be able to accomplish for a long while.

 

Actual compute is nowhere close to its theoretical.

I've been telling you this for pages now showing GT3e being beat by a large margin in compute. Gen8 may close the gap with Kaveri but Carrizo will be out at the same time period so it's in all fairness Broadwell's competition. I personally think GT3e Gen8 will close a large margin of the compute and performance gap. Tho I have my doubts of Intel taking the lead in iGPU performance any time soon.

[patrickjp93]

I posted compute tests a few posts ago showing current GT3e being staggeringly abused by previous GCN architecture. Carrizo mobile will be available this holiday season with the desktop variant arriving H1 next year. As said with it launching at the same exact time as Broadwell so it would be just ludacris to compare it to a previous generation microprocessor. I've stated before (that you apparently want to go along with now) theoretical numbers don't mean anything. The 5960x has a theoretical peak performance of 384 GFLOPS and doesn't even live up to no wheres near that. Also Skybridge is a universal SoC sharing not only the same package but also both x86 and ARM cores on the same die. The same exact chip can be used in desktops, laptops, notebooks, tablets, cell phones, servers, etc and drive Windows, Linux, OS X, Android, iOS, etc. So its miles ahead of what Intel will be able to accomplish for a long while.

 

I've been telling you this for pages now showing GT3e being beat by a large margin in compute. Gen8 may close the gap with Kaveri but Carrizo will be out at the same time period so it's in all fairness Broadwell's competition. I personally think GT3e Gen8 will close a large margin of the compute and performance gap. Tho I have my doubts of Intel taking the lead in iGPU performance any time soon.

You posted graphics tests. I don't see a compute one anywhere. Furthermore: http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/17

 

The only Kaveri compute performance comparison vs. Intel is against HD 4600 which is about 1/3 the performance of Iris Pro 5200. 

 

On top of this, Gen 8 increased the throughput per EU by 40% by decreasing the number per sub-slice by 20%. Then they added a 3rd slice, for 20% more SIMD units. Pardon me but it looks like Kaveri gets crushed, especially since Richland had better scores anyway due to better clock rates, and Carrizo's clocks fall again due to using HDL in the design process. 

 

If Carrizo wins in compute against Iris Pro 6200 in compute, it will be only because the desktop/server SKUs get HBM onboard, because otherwise it doesn't have the juice available. Then there's the problem Skylake with Gen 9 Graphics lands 2-3 months after Carrizo's launch.

[Opcode]

You posted graphics tests. I don't see a compute one anywhere. Furthermore: http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/17

 

The only Kaveri compute performance comparison vs. Intel is against HD 4600 which is about 1/3 the performance of Iris Pro 5200. 

 

On top of this, Gen 8 increased the throughput per EU by 40% by decreasing the number per sub-slice by 20%. Then they added a 3rd slice, for 20% more SIMD units. Pardon me but it looks like Kaveri gets crushed, especially since Richland had better scores anyway due to better clock rates, and Carrizo's clocks fall again due to using HDL in the design process. 

 

If Carrizo wins in compute against Iris Pro 6200 in compute, it will be only because the desktop/server SKUs get HBM onboard, because otherwise it doesn't have the juice available. Then there's the problem Skylake with Gen 9 Graphics lands 2-3 months after Carrizo's launch.

You posted the same source I did a page or two ago showing GT3e being beat by Kaveri by a huge margin. 

 

Read the source I (then you) posted again and look at Anandtech's compute scores for Kaveri (in fact I will quote my post again to make it easier on you).

 

One of the last things most consumers in the market today are interested in is compute performance. Even tho Kaveri already has GT3e's cards. 

 

 

 

 

 

 

Gen8 might be able to catch up to Kaveri tho Carrizo will be released around the same time bringing even more competition to Intel. I personally have my doubts that Intel will be focusing as much on their GPU architecture after Broadwell. Intel's playing catch-up in the SoC race. Carrizo will have its fusion hub built on die in which Intel is not expecting to do until Skylake. Intel's die size is much larger than Kaveri/Carrizo's as well as being on a smaller node. I'm waiting to see Carrizo's successor with die shrinks. I would assume 640 SPU's with HBM or DDR4 would be the next step for them.

 

I hate to bust your bubble but Intel slides show GT3e 8 only bringing roughly 20% more compute performance over Haswell's offering (direct effect of the extra 8 EU's) and this is only because GT3e 7.5 had design flaws (too low sampler throughput to keep up with the shaders). Also why bring up the whole Intel vs AMD IPC debate again? Neither of these chips pack enough iGPU horsepower in order for the CPU to be unable to feed it. Core performance has little to no adverse effect on iGPU compute performance. Side tracking trying to find a way to get around the fact that Kaveri mops the floor with Iris Pro this time around (willing to bet next time too) in compute is just silly. Tho Gen 8 should make a marginal gain on gaming performance due to the 50% increase in sampler throughput.

 

You won't see Skylake until late 2015 or even possibly pushed back to early 2016. Intel's not about to push out Broadwell's successor a month after Broadwell's release. You, I, and everyone else here knows that.

[patrickjp93]

You posted the same source I did a page or two ago showing GT3e being beat by Kaveri by a huge margin. 

 

Read the source I (then you) posted again and look at Anandtech's compute scores for Kaveri (in fact I will quote my post again to make it easier on you).

 

 

I hate to bust your bubble but Intel slides show GT3e 8 only bringing roughly 20% more compute performance over Haswell's offering (direct effect of the extra 8 EU's) and this is only because GT3e 7.5 had design flaws (too low sampler throughput to keep up with the shaders). Also why bring up the whole Intel vs AMD IPC debate again? Neither of these chips pack enough iGPU horsepower in order for the CPU to be unable to feed it. Core performance has little to no adverse effect on iGPU compute performance. Side tracking trying to find a way to get around the fact that Kaveri mops the floor with Iris Pro this time around (willing to bet next time too) in compute is just silly. Tho Gen 8 should make a marginal gain on gaming performance due to the 50% increase in sampler throughput.

 

You won't see Skylake until late 2015 or even possibly pushed back to early 2016. Intel's not about to push out Broadwell's successor a month after Broadwell's release. You, I, and everyone else here knows that.

That bench was on Iris Pro's release drivers. You run it again now on something with Intel's newer drivers Intel wins outright.

 

Intel's slides show a 20% CORE increase and a 40% per-core throughput increase. That combined with other improvements will equate to at least a 50% increase for some workloads.

 

Also, kirzanich at IDF said that is exactly what they will do, because Broadwell S won't exist, though Broadwell K will, and Skylake S is going to launch early H2 2015.

 

I'll run the same benchmark on my MBPr with the new OpenCL drivers (4950HQ).

 

you also used different benchmark versions for the last 2.

 

Well I can only get the newest version of compubench because of how the apple store works. Anyone have a source for 1.1.3? Or, does anyone have a Kaveri chip we could test the newest version on?

 

CompuBench 1.5.6 for I7 4960HQ (CPU ONLY)

 

CompuBench 1.5.6 for IRIS Pro 5200 on MBPr OSX Yosemite I7 4960HQ