AMD will launch AM4 platform in March 2016 says industry source

[ShintaiDK]

Have you seen any Compute performance Benchmarks of the new Iris pro 6200 in any of the reviews ???

I havent, and it seams OpenCL drivers are excellent as always on the Intel iGPUs.

Current official Windows 10 64bit driver (non Beta)
https://downloadcenter.intel.com/download/25484/Intel-Graphics-Driver-for-Windows-7-8-1-10-15-40-
Driver Version: 15.40.10.64.4300 & 15.40.10.32.4300
DATE: October 16, 2015

And then you believe Apple will not jump in to the opportunity to use ZEN with Polaris.

The 4300 isn't the latest non beta or official for that matter. Both 4331 and 4352 are officially released.

And the official one when searched for with Skylake is 4352.
https://downloadcenter.intel.com/pr...-530-for-6th-Generation-Intel-Core-Processors

But its easy to miss in the eager.

If you believe so much in Zen, put your money behind it. Because you will make a fortune then. And it would be AMDs first real success since K8.

 

[AtenRa]

The 4300 isn't the latest non beta or official for that matter. Both 4331 and 4352 are officially released.

And the official one when searched for with Skylake is 4352.
https://downloadcenter.intel.com/pr...-530-for-6th-Generation-Intel-Core-Processors

But its easy to miss in the eager.

If you believe so much in Zen, put your money behind it. Because you will make a fortune then. And it would be AMDs first real success since K8.

Go to the end of that page and see what products are supported by this driver.

Intel Iris Pro 6200 is not one of them sadly.

 

[ShintaiDK]

Go to the end of that page and see what products are supported by this driver.

Intel Iris Pro 6200 is not one of them sadly.

No its 4332 for Iris Pro 6200. Not 4300.
https://downloadcenter.intel.com/pr...6200-for-5th-Generation-Intel-Core-Processors

 

[AtenRa]

yea nice one unfortunately that is not the iGPU alone.

nice one, post the link pls

No its 4332 for Iris Pro 6200. Not 4300.
https://downloadcenter.intel.com/pr...6200-for-5th-Generation-Intel-Core-Processors

If you choose Windows 10 64bit the current official driver for the Iris pro 6200 is the 4300

 

[ShintaiDK]

Yet AMD is still behind. No wonder Apple uses Intel.

 

[Shivansps]

This. Basically this means making the APU more complex, because the HBM will work as a frame buffer and you would still use DDR memory as a main memory for the system otherwise costs will skyrocket.

Given that AMD is still reeling from putting HBM on very high end SKUs, I will be quite surprised if they are able to put HBM on cheap mainstream SKUs. Zen might improve their cost structure a bit but I doubt squander this cost improvement with HBM won't help them.

Im also thinking about the GPU core used, with HBM providing not less than 100GB/s, it has to be at least a 12CU R7 360 core, that is able to CF with another R7 360, that whould be interesting. That for the top bin "A10", if that happen i whould guess "A8" whould be a 10CU "R7 250X" maybe whiout HBM, "A6" 8CU "R7 250", and "A4" a R7 240.

 

[Fjodor2001]

AMD track record is that the next APU flops more than what it replaces. Carrizo APU for laptops have been a complete flop.

Yet somehow all this changes every single time with the next APU.

Do you really think it's reasonable to compare Carrizo to Zen in terms of what performance improvements can be expected over its predecessor?

Carrizo was an incremental uArch update that still stayed on 28 nm. Zen is a completely new uArch generation and goes from 28 to 14 nm.

 

[mrmt]

Im also thinking about the GPU core used, with HBM providing not less than 100GB/s, it has to be at least a 12CU R7 360 core, that is able to CF with another R7 360, that whould be interesting. That for the top bin "A10", if that happen i whould guess "A8" whould be a 10CU "R7 250X" maybe whiout HBM, "A6" 8CU "R7 250", and "A4" a R7 240.

It is important to separate what is an interest technical solution and an interesting product to market. How would in terms of production costs this solution compare against future GDDR5X solutions for the same performance level?

Trinity, Kaveri and Carrizo backfired on AMD because this comparison didn't make sense when compared to multiple combinations of Intel processors or Intel processors + dGPU, Zen APUs will backfire for the same reason if they don't solve the cost issues, and so far there is nothing pointing out for this direction.

 

[ShintaiDK]

Do you really think it's reasonable to compare Carrizo to Zen in terms of what performance improvements can be expected over its predecessor?

Carrizo was an incremental uArch update that still stayed on 28 nm. Zen is a completely new uArch generation and goes from 28 to 14 nm.

Why would anything change that didn't change the last 10 years happen? Its a company run into the ground and is expected to perform miracles?

 

[Shivansps]

It is important to separate what is an interest technical solution and an interesting product to market. How would in terms of production costs this solution compare against future GDDR5X solutions for the same performance level?

Trinity, Kaveri and Carrizo backfired on AMD because this comparison didn't make sense when compared to multiple combinations of Intel processors or Intel processors + dGPU, Zen APUs will backfire for the same reason if they don't solve the cost issues, and so far there is nothing pointing out for this direction.

I dont really think Kaveri APU or any APU failed for that, they failed for the same reason that every AMD product has failed so far, bad CPU perf compared to competition. On top of that, the problem was that until A8-7600 and A10-7850K showed up, the IGP perf was just not good enoght, and they are now just barely enoght.

The price diff on a R7 250 DDR3 (7850K IGP) -> R7 360 (the minimum ideal for AM4 APU) is just $20, and R7 360 level IGP is well inside the "good enoght" for 1080P that may even be a side-grade or upgrade to anyone with a 750TI and below, so they need to step up the game in both CPU and IGP, main concern is CPU, a Kaveri/FX level CPU perf with a R7 360 gona fail BAD, they need to get close to an I5, at least a SB i5.

The main problem for AMD is the transition to 4K.

 

[frozentundra123456]

The problem with APUs is that they fit perfectly the old saying "jack of all trades and master of none." Worse cpu performance than intel and worse graphics performance than even a low/mid range dgpu. And I would argue that the igp performance is still not really "good enough" unless you are willing to limit yourself in the games you play or to sub 1080p. APUs are not necessarily a *bad* product. I would agree with the APU brigade that yes, you *can* game on an APU. But why would you when such better alternatives are available at equal or very minimal additional cost?

And you are right, more demanding games, 4K and the next round of die shrunk dgpus are just going to raise the goalposts drastically higher for APUs. But that hasnt stopped the hype train. We have gone right past the Zen station and now are picking up speed for Zen APUs.

 

[Shivansps]

Still if the IGP is at least R7 360 level i think they will be fine, but they need to get a good cpu perf as well, 12CU and maybe even 14 seems possible.

 

[cbn]

Even with A10, Core i3 + GT 730 GDDR5 would be faster.

Haven't tried this setup, got any links ??? because the GT730 with GDDR5 only has a 64bit memory controller.

The only review I ever found was one at tipidpc, but they took the info down.

I do own the card though. So if you want me to do some gaming benchmarks with my Athlon x 4 860K I would be happy to do so. Then we can compare to your A10-7850K.

P.S. The memory bus is only 64 bit, but the bandwidth @ 40 GB/s is still higher than an APU with dual channel DDR3/4 2400 (which has to share bandwidth with the CPU).

Well i dont currently have the A10 but only the A8-7600, but i would really like to know how this GPU will do. we should talk in pm in a few days and try to make a new topic

Here are my Athlon x4 860K/GT730 GDDR5 Firestrike Results:

http://www.3dmark.com/3dm/10105262

Score: 1493
GPU score: 1614
Physics score: 4439

And here is the Firestrike score for a stock A10-7870K with DDR3 2800 (edit: It was run at DDR3 2133 speeds for the test)

http://www.tweaktown.com/reviews/7270/amd-series-a10-7870k-kaveri-refresh-apu-review/index6.html

Score: 1351
GPU score: 1498
Physics score: 4902

So going by GPU score GT 730 GDDR5 is 7.7% faster than a 512sp iGPU @ 866 Mhz with DDR3 2800 2133. Now granted the Athlon x 4 860K CPU doesn't throttle, but I don't know how much that factors into this benchmark. (EDIT: tested Athlon x 4 860K @ 2.8 Ghz with GT 730 GDDR5 and the GPU score came out the same--> http://forums.anandtech.com/showpost.php?p=37957735&postcount=485 )

 

[maddie]

The problem with APUs is that they fit perfectly the old saying "jack of all trades and master of none." Worse cpu performance than intel and worse graphics performance than even a low/mid range dgpu. And I would argue that the igp performance is still not really "good enough" unless you are willing to limit yourself in the games you play or to sub 1080p. APUs are not necessarily a *bad* product. I would agree with the APU brigade that yes, you *can* game on an APU. But why would you when such better alternatives are available at equal or very minimal additional cost?

And you are right, more demanding games, 4K and the next round of die shrunk dgpus are just going to raise the goalposts drastically higher for APUs. But that hasnt stopped the hype train. We have gone right past the Zen station and now are picking up speed for Zen APUs.

That sounds a lot like a general purpose computing device AKA a PC. What a hilarious statement.

 

[deasd]

Here are my Athlon x4 860K/GT730 GDDR5 Firestrike Results:

http://www.3dmark.com/3dm/10105262

Score: 1493
GPU score: 1614
Physics score: 4439

And here is the Firestrike score for a stock A10-7870K with DDR3 2800

http://www.tweaktown.com/reviews/7270/amd-series-a10-7870k-kaveri-refresh-apu-review/index6.html

Score: 1351
GPU score: 1498
Physics score: 4902

So going by GPU score GT 730 GDDR5 is 7.7% faster than a 512sp iGPU @ 866 Mhz with DDR3 2800. Now granted the Athlon x 4 860K CPU doesn't throttle, but I don't know how much that factors into this benchmark.

I've heard from somewhere that APU(Kaveri and older gen) limit the IMC to 2133Mhz internally, so even if you overclock DDR3 far ahead 2133Mhz there's no improvement in graphic performance.

 

[frozentundra123456]

Well, if you arent familiar with the saying it is not a compliment. Maybe you would find a more blunt term like mediocre at everything humorous as well.

 

[cbn]

Based on the info I have posted in this thread here, here and here I feel pretty confident Bristol Ridge in A8 APU and A10 APU format would be DOA on desktop (in the same way socket AM1 was pretty much DOA).

Most (and preferably all) of Bristol Ridge 4C 512sp and 4C 384sp dies should go to 35W mobile if at all possible.

This would leave harvested Bristol Ridge in the form of 4C without iGPU and 2C with 384sp for AM4. (With some leftover 4C 512sp and 4C 384sp dies possibly making it on AM4 around the time Zen APU launches)

Does anyone have objections? If so please present your case.

 

[DrMrLordX]

I've heard from somewhere that APU(Kaveri and older gen) limit the IMC to 2133Mhz internally, so even if you overclock DDR3 far ahead 2133Mhz there's no improvement in graphic performance.

Eh. What you're hearing is people parroting one reviewer's estimation that DDR3-2133 CL9 was the "sweet spot" for Kaveri memory performance, which is sort of half-assed. Kaveri still has more memory bandwidth available from DDR3-2400 CL10 when using the iGPU, so you do actually pick up a few frames by increasing memory speed. You can see continued gains from even higher memory clockspeeds, assuming you can achieve them with any degree of stability.

Does anyone have objections? If so please present your case.

Why bother? None of us run AMD anyway. They're gonna do what they're gonna do, and we all get to sit back and witness the results.

Bristol Ridge is coming to the desktop, and there's nothing you can do about it.

 

[AtenRa]

Yet AMD is still behind. No wonder Apple uses Intel.

I have asked you to post the link of that slide but you havent.

So, its a review of the Core i7-5775C from PCworld posted in early August 2015.

from the review,

http://www.pcworld.com/article/2950...ktop-broadwell-has-one-neat-trick.html?page=3

Note: I tested the A10-7870K with its RAM set to DDR3/1600 to match the Intel systems and also grabbed the latest Catalyst driver available.

Also, no mention of the Windows used for the review, could be win 7 or win 8 but certainly not Win 10.

 

[AtenRa]

Here are my Athlon x4 860K/GT730 GDDR5 Firestrike Results:

http://www.3dmark.com/3dm/10105262

Score: 1493
GPU score: 1614
Physics score: 4439

And here is the Firestrike score for a stock A10-7870K with DDR3 2800

http://www.tweaktown.com/reviews/7270/amd-series-a10-7870k-kaveri-refresh-apu-review/index6.html

Score: 1351
GPU score: 1498
Physics score: 4902

So going by GPU score GT 730 GDDR5 is 7.7% faster than a 512sp iGPU @ 866 Mhz with DDR3 2800. Now granted the Athlon x 4 860K CPU doesn't throttle, but I don't know how much that factors into this benchmark.

Just because they use a 2800MHz memory doesnt mean they run the memory at 2800MHz,

From the review you posted.

For the benchmarks, I decided that out of the box performance was just as important as clock for clock performance. I decided to test all the CPUs at stock, but I also gave the A10-7870K a 4.7GHz CPU overclock and a 2400MHz memory overclock, as these two were very easy to attain on the APU.

Only the overclocked results on that review use 2400MHz memory, the out of the box performance" it is run at 2133MHz.

edit : from the same teaktown review,

Windows 10 iGPU performance

With proper 2133MHz ram on Win 10 and A10-7870K iGPU performance is faster than Core i7 5775C in FireStrike.

Should be fun to see DX-12 Gaming Benchmarks between the two.
But the real battle from H2 2016 will be between Kabylake and Bristol Ridge and entry-level dGPU Polaris.

 

[The Stilt]

Obviously the iGPU in 7870K is superior to anything Intel currently has. Without the bandwidth restrictions it would be > 25% faster than it currently is and obviously even more at higher speeds.

Bristol Ridge will slightly increase the lead due the higher GPU frequencies, DDR-2400 support and DCC introduced in Tonga.

Despite that, unless AMD has been able to solve the bandwidth issue in products to be released in 2016 / 2017 the game of APUs is pretty much over. If Intel keeps up the same rate they have been improving their iGPUs during the past two years, by then end of 2016 they´ll have outperformed in AMD in every iGPU segment.

Even if Raven Ridge support DDR4-3200 it won´t be even remotely enough to allow significant improvements in the iGPU performance. Even the current generation desktop APUs can easily utilize > 96GB/s worth of bandwidth. AMD needs either a very fast on-die frame buffer or HBM, nothing else will basically do.

 

[AtenRa]

Obviously the iGPU in 7870K is superior to anything Intel currently has. Without the bandwidth restrictions it would be > 25% faster than it currently is and obviously even more at higher speeds.

Bristol Ridge will slightly increase the lead due the higher GPU frequencies, DDR-2400 support and DCC introduced in Tonga.

Despite that, unless AMD has been able to solve the bandwidth issue in products to be released in 2016 / 2017 the game of APUs is pretty much over. If Intel keeps up the same rate they have been improving their iGPUs during the past two years, by then end of 2016 they´ll have outperformed in AMD in every iGPU segment.

Intel has the same memory bandwidth problem as AMD, only their $300-400+ iGPUs Iris Pro have eDRAM, the rest are only using dual channel DDR-4. At the same memory bandwidth AMD is still 30-50% faster than current Intel Skylake. As you said next year Bristol Ridge will increase performance by lets say 10-20% ???
But in 2017 with ZEN APUs at 14nm, high end APU SKUs will definitely need HBM or any other high bandwidth memory otherwise they will be under performing like crazy.

 

[mrmt]

I dont really think Kaveri APU or any APU failed for that, they failed for the same reason that every AMD product has failed so far, bad CPU perf compared to competition. On top of that, the problem was that until A8-7600 and A10-7850K showed up, the IGP perf was just not good enoght, and they are now just barely enoght.

Even if AMD had more CPU performance it would be still selling dies that are the double of the competition in terms of size but at a higher price, meaning that it would still be in a bad competitive terms, just better than the currently disastrous terms of today, and that would still mean bad margins or lost market share if Intel decided to turn the margins screws.

Just look at happened to the cat family, Bay Trail wiped out the cat family despite the cat family not having a performance handicap, that happened because of the better cost structure and Intel willingness to get the market for itself (and pay the price for it), the same will happen if AMD go for better CPU performance but without a better cost structure to back it up.

 

[cbn]

Bristol Ridge is coming to the desktop, and there's nothing you can do about it.

Yes, I realize Bristol Ridge is coming to desktop.

But specifically the 4C 512sp and 4C 384sp dies don't make much sense on AM4. 512sp desktop iGPU in mid to late 2016 is not relevant for gaming when today's low profile cards are already faster.

For 35W mobile though, I can see Bristol Ridge quad core with iGPU with a potential niche against either 28W Intel GT3e or 15W Intel + GM108 DDR3 (ie, GT940M). In this scenario, the investment AMD made in integration makes a lot more sense.

 

[AtenRa]

Just look at happened to the cat family, Bay Trail wiped out the cat family despite the cat family not having a performance handicap, that happened because of the better cost structure and Intel willingness to get the market for itself (and pay the price for it), the same will happen if AMD go for better CPU performance but without a better cost structure to back it up.

Need to remind you that 28nm Jaguar SOC was almost the same die size as 22nm Intels BayTrail ???? And we both know very well 22nm was far more expensive at the time of the BayTrail release than 28nm planar.

Intel payed more than 10B to come to the results they have today in the mobile/Tablet market. That is a shrinking market share that will come back to almost zero in the end of this year.

They cannot do the same for their big cores, that is were they get there profits from.