Should AMD release an updated 8 core FX (improved cores, smaller process)?

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Should AMD offer a minor FX 8 core update?

  • definitely yes

  • definitely not

  • not sure

  • if it's not an Intel I don't care


Results are only viewable after voting.

dark zero

Platinum Member
Jun 2, 2015
2,655
140
106
Ok, let's start to become realist and not biasing Intel or AMD.

The situation of AMD is bad, However they have enough cash to land a last try.

Seeing the current situation... I see that:
- Zen alone will come at 2017, servers might get lucky and enters on Q4 2016 or Q1 2017
- Zen APU might come at H2 2017 let alone 2018. HBM2 will get.mainstream recently at 2H 2017.
- Intel Cannonlake will come at H2 2017 but considering how the desktop and laptop market are not in good shape and 10nm are still inmature, they might launch on servers first and few months later on desktop or even Q1 2018.
- Intel Kabylake is supposed to enter just on Q3 2016 to counter Pascal and Artic Islands (if there is not a shortage, Intel iGPU will get sodomized here)

Artic Islands and Pascal will come at any means by Q3 2016, but seeing that nVIDIA is using nVILiNK on HPC, the situation might say something from the greens on consumers. Artic might come with shortage BTW.

But anyways, the important thing is not that. Is that there might be a years empty on AMD CPU. They have 2 options:
- Release nothing, except the cards, and cut dramatically the prices of their current products
- Release the last generation of Bulldozer

We saw in leaks that AMD is doing the 2nd by releasing Bristol Ridge (Quads) and Stoney Ridge (low power duals)

Now we recieved lots of info that the socket used will be AM4. And the chips were supposed to be compatible with FM2.

However we saw also that Zen is using AM4 too... And seeing that is a HPC... And also, seeing that PGA is not good performer after all, AMD might move to LGA.

Also, AMD must test some of their iGPU advantage ... Because Tonga is already released on Kaveri. And Intel outmatched it.


Finally, 28nm already reached at their top with HDL. And HDL only works up to 35 watts... Bristol will go up to 95 w... And 14nm seems only made for Zen.

And there are rumors that AMD send an order to TSMC to do chips different than Artic.


Now my predictions about next year on AMD CPU comes:

- Bristol Ridge and Stoney Ridge will get released at 16nm made by TSMC

- It will come at Q2 or H2 2016 to compete Kabylake iGPU and to come with competitive CPU.

- Both chips will have a cut version of Artic Islands GPU, making the trope Roy debuted in Super Smash Bros and to explain the 40% GPU improvement.

- That means that AMD will go to have 10% advantage against Kabylake IGPU.

- Also that means that there won't be a low tier on dGPU from them since Artic Island minimun memory is 4Gb of HBM2 (nVIDIA will have a low tier card with 2GB on HBM1) and those chips are the new minimun.

- Both will use only DRR4 and LPDDR3 with high clocks (2133 on LPDDR3 and 3000 Mhz on DDR4)

- Since Zen is towards HPC and AMD will unify everything, AMD will move everything to LGA

- Bristol and Stoney are targeted to have Single Core Performance similar to Ivy Bridge and theorical Broadwell Multicore performance.

Now being specific on the Ridge family, except Summit.

- Bristol will be targeted to desktops and medium Steam machines.
It is suposed to get up to 95 Watts.
The max. GHZ avaliable might be up to 3.5 Ghz.
The brands will be A8, A10 and Fx (being the last one OCable)
Expecting from 10% up to 20% cost increase compared to the previous generation.
Goes to compete against Kabylake.


- Stoney will get to low cost machines, small factors machines and laptops replacing Carrizo and Carrizo-L.
It is supossed to have the range from 10 to 35 watts.
The max. GHZ avaliable might be up to 3.2 Ghz on turbo and 2.8 Ghz on normal work. Not hard to reach since It's a Dual Core.
The brands used could be E1 (or even Duron), E2 (Or Sempron), A4 (or even Athlon and A6 (don't think that Phenom will be used here) and all.of them won't be OCable
Expecting only up to 10 % cost increase on their chips compared to the previous gen.
Goes to compete Atom and Kabylake-U

And Well... That's my predictions by following AMD news. They don't have money, but they can't being stuck on the same tech.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
I meant from an overall performance, satisfaction of use stand point. Obviously it's a bad part from a business perspective, and doesn't exist for a reason.

Well, Bulldozer/Piledriver FX weren't really a star in terms of sales in the channel and OEMs weren't fan of the thing either, so I think from an overall satisfaction of use stand point FXs were bad products. That's the only conclusion we can get from what people voted with their wallet at least. This new Steamroller/Excavator FX wouldn't really address the shortcomings that made the previous FXs a bad product.

Sure, such a product might cater to AMD fans like yourself, but there aren't many people wanting to disregard the ancient platform, the irregular gaming performance and the extra power consumption.
 

R0H1T

Platinum Member
Jan 12, 2013
2,582
163
106
Well, Bulldozer/Piledriver FX weren't really a star in terms of sales in the channel and OEMs weren't fan of the thing either, so I think from an overall satisfaction of use stand point FXs were bad products. That's the only conclusion we can get from what people voted with their wallet at least. This new Steamroller/Excavator FX wouldn't really address the shortcomings that made the previous FXs a bad product.
Arguably the single biggest problem with FX was that it ran too hot & was also a power hog, 14nm FF should solve one if not both & I bet it'd clock way higher than 28nm Carrizo. If priced right the Excavator core does have plenty of room in the current (desktop) mainstream segment especially where IGP isn't what buyers are looking for.

Sure, such a product might cater to AMD fans like yourself, but there aren't many people wanting to disregard the ancient platform, the irregular gaming performance and the extra power consumption.
Wait didn't a bunch of non AMD fans throw around this argument that IGP in desktops is just a waste of die space, not good enough to do serious gaming, & people would still buy discrete anyway? Sure a straight Piledriver port would be a disaster but not an Octa core Excavator chip which is clocked fairly high, whether that actually makes business sense from AMD's PoV is a separate debate.
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
22,688
12,632
136
Maybe Via will make a CPU based on the AM3+ Socket.

Wait a minute.., I have an AM3+ system that's not very funny :) Maybe the new Zen could still drop into the AM3+ socket like an AM2 CPU will and still use DDR3. Oh come on, why not?

AM3+ doesn't have the pin configuration to handle all the power planes required of AM4 CPUs. AM4 is meant to have the best of AM3+ and FM2+ all wrapped up in one tidy socket: enough capacity to handle 125W+ CPUs and support for all the separate power planes of AMD's modern APUs. There will be other platform improvements, such as dedicated PCI-e lanes for storage.

Arguably the single biggest problem with FX was that it ran too hot & was also a power hog, 14nm FF should solve one if not both & I bet it'd clock way higher than 28nm Carrizo. If priced right the Excavator core does have plenty of room in the current (desktop) mainstream segment especially where IGP isn't what buyers are looking for.

As I have stated before, even XV on 28nm + HDL has some clockspeed ranges where it would make a thrifty, desirable 4M chip had it been launched alongside Carrizo. There's a target market for something like that, too. AMD didn't specifically need 14nm just to pull off a chip like that. What they needed was the will, the foresight, and the money to make something like that happen. They came up short in at least one of those departments.

Wait didn't a bunch of non AMD fans throw around this argument that IGP in desktops is just a waste of die space, not good enough to do serious gaming, & people would still buy discrete anyway?

Plenty of "I don't like APUs" AMD fans said the same thing.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
Arguably the single biggest problem with FX was that it ran too hot & was also a power hog, 14nm FF should solve one if not both & I bet it'd clock way higher than 28nm Carrizo. If priced right the Excavator core does have plenty of room in the current (desktop) mainstream segment especially where IGP isn't what buyers are looking for.

I think you are expecting too much from Samsung 14nm node, let alone from Globalfoundries cheapskate copy of Samsung 14nm node.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
Err... Apple wants a word with you

No, it doesn't. The performance target AMD is pursuing with Zen, or that it would need to pursue for a revamped Steamroller/Excavator FX doesn't have much to do with the performance targets Apple was pursuing with A9. That said, I think people are expecting too much for AMD on 14nm.
 

superstition

Platinum Member
Feb 2, 2008
2,219
221
101
Arguably the single biggest problem with FX was that it ran too hot & was also a power hog...
Slow RAM performance as well, eh?

What about cache latency?

I wonder if a bit of minor tinkering could speed up its RAM performance or if the problem is the ancient nature of the 970 and 990 chipsets.


When I've read about FX the weaknesses I've heard of are these:

1) RAM performance

2) floating point performance

3) power consumption at the high clocks needed to try to stay with Intel

4) too deep pipeline (to get those high clocks) — like Netburst

5) too much die space devoted to having many weak cores rather than fewer large/powerful cores — not efficient except in specific workloads that can leverage them (like 7-zip).

6) less than optimal turbo (see #5 also)

7) too much reliance on automated tools which increases die area and power consumption. This seems specious because Intel uses them heavily as well, right? Also, CMT is supposed to reduce the die space needed, right?

8) caches are high in latency, not so efficient

9) AM3+ socket uses 65nm chips, lacks native USB-3 and PCI-E 3.0 (not really the processor's fault, but still...)


strengths seem to be:

1) ability to have very high clock speeds, which makes overclocking fun for enthusiasts, especially those who want to invest in water

2) multi-threaded performance is good considering the cost of the chips for consumers (particularly things like a $100 8320E from Microcenter with a $40 off motherboard bundle — which can overclock to 4.5 GHz on air). Of course, this low pricing can be seen as a drawback for AMD's profitability.

3) no die wasted on iGPU

4) large die area provides more space to dissipate heat

5) AMD didn't cheap out and uses actual solder

6) all are unlocked

7) CMT allegedly provides performance with fewer transistors/die space, in comparison with SMT

8) Modules provide better integer performance, but at the expense of floating-point

9) longevity of AM3+ socket means people can feel more comfortable buying a high-end motherboard instead of a bunch of lower midrange motherboards, making overclocking a better experience — plus the drawbacks are fairly minor unless you're going to run a triple or quadruple GPU setup
 
Last edited:

superstition

Platinum Member
Feb 2, 2008
2,219
221
101
Ok, let's start to become realist and not biasing Intel or AMD.

The situation of AMD is bad, However they have enough cash to land a last try.

Seeing the current situation... I see that:
- Zen alone will come at 2017, servers might get lucky and enters on Q4 2016 or Q1 2017
- Zen APU might come at H2 2017 let alone 2018. HBM2 will get.mainstream recently at 2H 2017.
- Intel Cannonlake will come at H2 2017 but considering how the desktop and laptop market are not in good shape and 10nm are still inmature, they might launch on servers first and few months later on desktop or even Q1 2018.
- Intel Kabylake is supposed to enter just on Q3 2016 to counter Pascal and Artic Islands (if there is not a shortage, Intel iGPU will get sodomized here)

Artic Islands and Pascal will come at any means by Q3 2016, but seeing that nVIDIA is using nVILiNK on HPC, the situation might say something from the greens on consumers. Artic might come with shortage BTW.

But anyways, the important thing is not that. Is that there might be a years empty on AMD CPU. They have 2 options:
- Release nothing, except the cards, and cut dramatically the prices of their current products
- Release the last generation of Bulldozer

We saw in leaks that AMD is doing the 2nd by releasing Bristol Ridge (Quads) and Stoney Ridge (low power duals)

Now we recieved lots of info that the socket used will be AM4. And the chips were supposed to be compatible with FM2.

However we saw also that Zen is using AM4 too... And seeing that is a HPC... And also, seeing that PGA is not good performer after all, AMD might move to LGA.

Also, AMD must test some of their iGPU advantage ... Because Tonga is already released on Kaveri. And Intel outmatched it.


Finally, 28nm already reached at their top with HDL. And HDL only works up to 35 watts... Bristol will go up to 95 w... And 14nm seems only made for Zen.

And there are rumors that AMD send an order to TSMC to do chips different than Artic.


Now my predictions about next year on AMD CPU comes:

- Bristol Ridge and Stoney Ridge will get released at 16nm made by TSMC

- It will come at Q2 or H2 2016 to compete Kabylake iGPU and to come with competitive CPU.

- Both chips will have a cut version of Artic Islands GPU, making the trope Roy debuted in Super Smash Bros and to explain the 40% GPU improvement.

- That means that AMD will go to have 10% advantage against Kabylake IGPU.

- Also that means that there won't be a low tier on dGPU from them since Artic Island minimun memory is 4Gb of HBM2 (nVIDIA will have a low tier card with 2GB on HBM1) and those chips are the new minimun.

- Both will use only DRR4 and LPDDR3 with high clocks (2133 on LPDDR3 and 3000 Mhz on DDR4)

- Since Zen is towards HPC and AMD will unify everything, AMD will move everything to LGA

- Bristol and Stoney are targeted to have Single Core Performance similar to Ivy Bridge and theorical Broadwell Multicore performance.

Now being specific on the Ridge family, except Summit.

- Bristol will be targeted to desktops and medium Steam machines.
It is suposed to get up to 95 Watts.
The max. GHZ avaliable might be up to 3.5 Ghz.
The brands will be A8, A10 and Fx (being the last one OCable)
Expecting from 10% up to 20% cost increase compared to the previous generation.
Goes to compete against Kabylake.


- Stoney will get to low cost machines, small factors machines and laptops replacing Carrizo and Carrizo-L.
It is supossed to have the range from 10 to 35 watts.
The max. GHZ avaliable might be up to 3.2 Ghz on turbo and 2.8 Ghz on normal work. Not hard to reach since It's a Dual Core.
The brands used could be E1 (or even Duron), E2 (Or Sempron), A4 (or even Athlon and A6 (don't think that Phenom will be used here) and all.of them won't be OCable
Expecting only up to 10 % cost increase on their chips compared to the previous gen.
Goes to compete Atom and Kabylake-U

Any idea if Kaby Lake is supposed to finally get H.265 encoding and decoding?

Is Zen supposed to have it?
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,002
126
Well, Bulldozer/Piledriver FX weren't really a star in terms of sales in the channel and OEMs weren't fan of the thing either, so I think from an overall satisfaction of use stand point FXs were bad products. That's the only conclusion we can get from what people voted with their wallet at least. This new Steamroller/Excavator FX wouldn't really address the shortcomings that made the previous FXs a bad product.

Sure, such a product might cater to AMD fans like yourself, but there aren't many people wanting to disregard the ancient platform, the irregular gaming performance and the extra power consumption.


I know there isn't much that will ever make you say AMD's product is decent. But again, I said with improved cache (there is certainly room for improvement), a better manufacturing process than the current 32/28nm, and of course the core improvements. Right now AMD struggles to compete with i3's in many workloads, this would probably give them a product they could sell at higher than i3-competitive prices. The problem is it'd still be second best to Intel's CPU's in most cases, and how many AM3+ users are there left? I understand why it's not a good product from a business standpoint. But from a user's perspective I think it'd be a fine performer and power use would likely be much improved.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
I know there isn't much that will ever make you say AMD's product is decent. But again, I said with improved cache (there is certainly room for improvement), a better manufacturing process than the current 32/28nm, and of course the core improvements. Right now AMD struggles to compete with i3's in many workloads, this would probably give them a product they could sell at higher than i3-competitive prices. The problem is it'd still be second best to Intel's CPU's in most cases, and how many AM3+ users are there left? I understand why it's not a good product from a business standpoint. But from a user's perspective I think it'd be a fine performer and power use would likely be much improved.

Maybe you didn't see my comments about Brazos on the last thread about it.... But all in all, we differ regarding Bulldozer chips because you see them as decent chips with a few flaws here and there, and I see it as a failed concept that is unfixable no matter how much engineering efforts AMD puts into it, meaning that they would still compete against i3 and have i7 costs even with all improvements you listed.
 
Mar 10, 2006
11,715
2,012
126
Any idea if Kaby Lake is supposed to finally get H.265 encoding and decoding?

Is Zen supposed to have it?

Skylake already has H.265 decode/encode.

Zen is a CPU core, and the initial parts will not have dedicated media blocks, so no. You'll have to wait for the Zen-based APUs that launch in 2017.
 

tential

Diamond Member
May 13, 2008
7,348
642
121
Skylake already has H.265 decode/encode.

Zen is a CPU core, and the initial parts will not have dedicated media blocks, so no. You'll have to wait for the Zen-based APUs that launch in 2017.

Hopefully Arctic Islands has h.265 decode/encode as well as HDMI 2.0 and is out at the same time Zen is....
Since we know Fiji doesn't..
 
Mar 10, 2006
11,715
2,012
126
Hopefully Arctic Islands has h.265 decode/encode as well as HDMI 2.0 and is out at the same time Zen is....
Since we know Fiji doesn't..

AMD already has H.265 decode in Carrizo, and I suspect its Zen-based APUs will have H.265 encode. I would be surprised if AMD didn't bake this tech into its future dGPUs.