Vishera on AM3+ for 2012 and 2013!

I'm interested to see how far can they scale up their APU designs actually. If Intel can scale up Sandy Bridge all the way up to the performance enthusiast sector with the 2700K, why can't AMD do the same thing either?
 
R3MF said:
AMD analyst day:

http://www.anandtech.com/show/5491/amds-2012-2013-client-cpugpuapu-roadmap-revealed

i am disappointed in their desktop roadmap.

ditching sepang/terramar means no on-die PCIe 3.0 controller (with socket FM2), and relying on an aging chipset stuck with off-die PCIe 2.0 (socket AM3+). also means no triple-channel memory in the enthusiast space either.

and not only for 2012, but for much of 2013 as well!
Click to expand...

I guess the change in the server lineup could be related to either design or foundry issues.

Also from a feature/performance standpoint, does AMD really need on-die PCI Express in the server sector? While it would be great to have, it is not a must have feature.

yimingwuzere said:
I'm interested to see how far can they scale up their APU designs actually. If Intel can scale up Sandy Bridge all the way up to the performance enthusiast sector with the 2700K, why can't AMD do the same thing either?
Click to expand...

First: Because AMD is not Intel.
Second: Technically, AMD has APUs that run the gamut from tablets all the way to 'performance' laptops.

I'm still waiting on a Opteron APU, that would be great in HPC applications:
Think of this but with Opteron APUs (4/8 core, 400+ GCN shaders, 125W) instead of Interlagos Opterons, running with a bevy of Tahiti-based FireSTREAM cards...
 
JMccovery said:
I guess the change in the server lineup could be related to either design or foundry issues.

Also from a feature/performance standpoint, does AMD really need on-die PCI Express in the server sector? While it would be great to have, it is not a must have feature.
Click to expand...

it is not the server sector i care about, rather it is the desktop, and if they have the PCIe controller on-die in one it will be identical in the other.

more to the point, the change leaves AMD reliant on a chipset that they never intended to upgrade to PCIe 3.0, thus the hi end desktop will be lumbered with (off-die) PCIe 2.0.
 
Yeah, it sucks, but I can see where Read doesn't want to keep throwing away money to try to conquer an insurmountable force (Intel). AMD won't get another chance like the Pentium 4, EVER.

Warning, Personal Rant:
If there is anyone reading this, that can affect design choices at AMD, PLEASE, shitcan the 10-series chipsets until they have PCI-E 3.0, and release the upcoming SB1050 as a SB955 to go with the current 9-series northbridges.
 
agreed.

i can easily be persuaded to buy a 95W eight-core piledriver CPU if the price is reasonable, but i want a northbridge with PCIe 3.0 and a southbridge with native USB 3.0.

it is not a lot to ask.
 
I think is best to wait and see what will happen.
I think they have learned their lesson with bulldozer.
Creating a big hype and then fail to deliver.
Maybe things will be better with the second gen fx cpus.
Who knows only time will tell.
I think they are trying to bring back some money first in the company and then to focus on the enthusiast level.
As for PCIe 3.0 there is little to gain from it now.
First they need better cpus for desktops.
 
Last edited:
R3MF said:
it is not the server sector i care about, rather it is the desktop, and if they have the PCIe controller on-die in one it will be identical in the other.

more to the point, the change leaves AMD reliant on a chipset that they never intended to upgrade to PCIe 3.0, thus the hi end desktop will be lumbered with (off-die) PCIe 2.0.
Click to expand...

and i have absolutely no problem with that. in the end it keeps prices down and there quite frankly isn't anything PCI-e 3.0 on die does better than off die PCI-e 2.0. so i'll enjoy not having to fork over the extra cash for a feature i'll never use nor take advantage of anytime in the near future.
 
R3MF said:
agreed.

i can easily be persuaded to buy a 95W eight-core piledriver CPU if the price is reasonable, but i want a northbridge with PCIe 3.0 and a southbridge with native USB 3.0.

it is not a lot to ask.
Click to expand...

What is so great about PCI-E 3.0? AMD is not going to implement something unless it makes a tangible difference. For example, Intel always made the jump to faster memory, but amd didnt to keep their prices lower. In testing the performance difference did not justify the cost. I believe this is the same scenario.
 
buttons said:
What is so great about PCI-E 3.0? AMD is not going to implement something unless it makes a tangible difference. For example, Intel always made the jump to faster memory, but amd didnt to keep their prices lower. In testing the performance difference did not justify the cost. I believe this is the same scenario.
Click to expand...

Actually, with their GPUs doing so well it makes quite a bit of sense. Using a 7970 there can be some performance degradation on particular games at x8 PCIE 2 where in PCIE 3 x8 that goes away (PCIE 2 x8 is PCIE 3 x4. x16 PCIE 2 = x8 PCIE 3.0). It only applies to particular titles so it's not a huge deal atm, but it's certainly something that should be on the agenda. That's not the biggie, though...

Remember AMD introduced GCN and it deserves merits for its computational ability as well. If they're agenda is full GPU/CPU integration and HSA by 2014 then PCIE 3.0 is absolutely required or at least something else with massive bandwidth. In reality, it should be AMD leading the charge with regards to GPU connectivity considering they're leading the charge with GPU performance. What good is having an incredible video card if you're putting it in a PCIE slot that just hampers its performance?

http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/25
 
Last edited:
pelo said:
Actually, with their GPUs doing so well it makes quite a bit of sense. Using a 7970 there can be some performance degradation on particular games at x8 PCIE 2 where in PCIE 3 x8 that goes away (PCIE 2 x8 is PCIE 3 x4. x16 PCIE 2 = x8 PCIE 3.0). It only applies to particular titles so it's not a huge deal atm, but it's certainly something that should be on the agenda. That's not the biggie, though...

Remember AMD introduced GCN and it deserves merits for its computational ability as well. If they're agenda is full GPU/CPU integration and HSA by 2014 then PCIE 3.0 is absolutely required or at least something else with massive bandwidth. In reality, it should be AMD leading the charge with regards to GPU connectivity considering they're leading the charge with GPU performance. What good is having an incredible video card if you're putting it in a PCIE slot that just hampers its performance?

http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/25
Click to expand...

Currently, utilizing PCI Express 2.0 vs 3.0 incurs a performance penalty of 10% at the most for computational programs and about 5% for games (mainly the x8 vs x16 results).

It is possible that by the time AMD launches something to replace the AMx line, we may need the extra bandwidth that PCI-E 3.0 provides.
 
The difference currently isn't significant. But if people are willing spend an extra $20-40 on a 6870 than a 6850 or +$20 on 560Ti > 6950, or even $550 for a 7970(s) then spending the extra bit of money on a PCIE 3.0 board isn't exactly unreasonable. In fact it just makes sense. AMD boards generally offer the x16+x16 or x16+x8 slots at cheaper price now, but Intel is embracing PCIE 3.0. That difference between platforms also has to do with the lanes as well and not just whether it's 2.0 vs 3.0, so the difference in terms of GPU bandwidth and performance is marginal when factoring in price. Still, though, that's not the point.

With AMD's current agenda I'd think the impact of the extra bandwidth is something they should definitely be embracing and leading in. They're obviously very dedicated to GPU performance and their GCN architecture is also suited for the HPC world, so that 10% in computation is actually quite significant when you consider just how much $$ the HPC world splurges on GPUs. Not to mention it would also allow for breathing room for current platforms with future GPUs. Intel platform + AMD GPU? not exactly something that should sit well with them.

My guess is we'll see PCIE 3.0 in SoC form before or around the time we see it on a chipset and that'll probably be because of their transition toward APUs and specialized modular designs. It's not a matter of 'oh that's nice to have for the future' anymore, but rather something that AMD should be leading in.
 
What's the point of PCIe on die, anyways? Does it really give you a massive FPS boost or something? All I ever see it do is lock down CPUs so you can't overclock them unless they have an unlocked multiplier.
 
sdlvx said:
What's the point of PCIe on die, anyways? Does it really give you a massive FPS boost or something? All I ever see it do is lock down CPUs so you can't overclock them unless they have an unlocked multiplier.
Click to expand...

it's more efficient in wattage and total system cost. so, not a whole lot in the realm of what enthusiasts are concerned about.
 
If power targets are met with decent CPU and GPU performance (sr > pd > bd in 1t performance), I may upgrade my SFF system next year with a Kaveri APU, especially if HSA support pans out (C++ AMP or whatever).

sdlvx: what venm11 said. It can also become more efficient since the memory controller is on the CPU. An external PCIe controller required a regular bus connection (HyperTransport on AMD CPUs), but that can be moved to a wider/higher bandwidth bus internally. It doesn't necessarily limit overclocking, just if PCIe ratios or PCIe clock speed lock are not provided. Most CPUs I've seen are not clock speed limited by an integrated PCIe controller, at least not since several years old models.
 
pxc said:
sdlvx: what venm11 said. It can also become more efficient since the memory controller is on the CPU. An external PCIe controller required a regular bus connection (HyperTransport on AMD CPUs), but that can be moved to a wider/higher bandwidth bus internally. It doesn't necessarily limit overclocking, just if PCIe ratios or PCIe clock speed lock are not provided. Most CPUs I've seen are not clock speed limited by an integrated PCIe controller, at least not since several years old models.
Click to expand...

I think he was referring to the limited base clock on Llano, Lynnfield, Clarkdale and Sandy Bridge chips.
 
JMccovery said:
I think he was referring to the limited base clock on Llano, Lynnfield, Clarkdale and Sandy Bridge chips.
Click to expand...

me? i meant that HT is an extra hurdle in getting to the PCIe bus. the bridge eats up watts and costs $.

you might be thinking about what i asked about in another thread regarding the meaning of FSB.
 
AMD is kinda shooting itself in the foot with the design choice of having the fastest GPU on the market feature PCI-E 3.0, but none of their mobo/cpu parts. It basically puts a tight seal on Intel ruling the 'money doesn't matter' enthusiast market (not that the seal wasn't tight enough already). Now there is simply no excuse to run an AMD setup. Oh, I want AMD to be AMD again..
 
Unfortunately, they probably don't have the resources right now to create a new chipset for their desktop motherboards that includes PCIE 3.0. In the grand scheme of things, we're a small market and I'm sure they're focusing on laptop and server parts right now.
 
when amd will include PCIE 3.0,it will be when we will really need it.
As for people complaining now for why amd still does not include it,i bet those 2011 intel boards would be
useless,and they will tell you to buy a new board in order to really fully utilize the PCIE 3.0 speeds. :D
So why to complain about it now?
My gues is that a new chipset might launch with vishera cpus or at the end of 2012.
 
Last edited:
sdlvx said:
What's the point of PCIe on die, anyways? Does it really give you a massive FPS boost or something? All I ever see it do is lock down CPUs so you can't overclock them unless they have an unlocked multiplier.
Click to expand...

lower latency theoretically, but in reality merely that they had no plans to further develop their northbridge chips given the shift to PCIe on-die, but now with the delay that lack of development is an acute problem.

buttons said:
What is so great about PCI-E 3.0? AMD is not going to implement something unless it makes a tangible difference. For example, Intel always made the jump to faster memory, but amd didnt to keep their prices lower. In testing the performance difference did not justify the cost. I believe this is the same scenario.
Click to expand...

higher access speed and lower latency, which given the plan to initially allow gpu access to system memory (and later to have a unified address space), and thus be a real boost to gpu computing.

real world example right now - i want to run blender on the gpu via cycles and yet some of our scenes are larger than the framebuffer which is a no go.

if this is going to be possible then it absolutely needs high-bandwidth access to main memory and minimal latency, so on-die memory and on-die PCIe makes sense.
 
R3MF said:
lower latency theoretically, but in reality merely that they had no plans to further develop their northbridge chips given the shift to PCIe on-die, but now with the delay that lack of development is an acute problem.



higher access speed and lower latency, which given the plan to initially allow gpu access to system memory (and later to have a unified address space), and thus be a real boost to gpu computing.

real world example right now - i want to run blender on the gpu via cycles and yet some of our scenes are larger than the framebuffer which is a no go.

if this is going to be possible then it absolutely needs high-bandwidth access to main memory and minimal latency, so on-die memory and on-die PCIe makes sense.
Click to expand...
^Yeah it's all about being ready for changing technology.. we can't always be shortsighted.. well, if you're gonna be a new mobo next year then that's fine lol.. but I try to buy once every 3-5 years, so.. I try to keep up with the latest when I can.
 
give amd some time to get their shit straight and their new management settled. i think they are a bit conservative regarding the recent road map. since they can state a processor will be out on a certain foundry node and the foundry screws up the node so completely that the products that were to be made on that node end up getting scrapped.
 
I read some shit that they are scrapping a bunch of stuff, I hope it turns out to be fud.
 
stealthy123 said:
I read some shit that they are scrapping a bunch of stuff, I hope it turns out to be fud.
Click to expand...
That was shown by AMD in slides and mentioned in the financial analyst day presentation. Rory Read seems to be streamlining the processor product line.
 
pxc said:
That was shown by AMD in slides and mentioned in the financial analyst day presentation. Rory Read seems to be streamlining the processor product line.
Click to expand...

Well what if Rory Read is trying to make the giant sleep?in this case its intel.
Just to buy time and then bring out some crazy shit. :p
Well definitely AMD needs changes,just give the guy sometime.
He cannot do it overnight.
 
the reason the extra module was dropped is cause ddr4 is not out and the last module would be bandwidth starved not to mention a change of sockets for servers that would cost them too much.
 
The only problem i have with BD is that damn power consumption. If they can manage to decrease power usage with the added 10% - whatever% speed increase they've speculated, I'd be a happy panda. :)
 
nOrVow said:
The only problem i have with BD is that damn power consumption. If they can manage to decrease power usage with the added 10% - whatever% speed increase they've speculated, I'd be a happy panda. :)
Click to expand...

I'm pretty sure we'll see power consumption reduced as the 32nm tech is improved. I doubt that the high power consumption is an architectural flaw, but we'll see once Piledriver is released.
 
SMa4G5al said:
we may need the extra bandwidth that PCI-E 3.0 provides.
Click to expand...
You people don't really get it do you?
AMD is fighting both sides here.nvidia and intel.
Its all about money here.AMD is really short on funds now.
They might drop the desktop market completely.
And before you come here and complain about pcie 3.0,
amd will bring pcie 3.0 when nvidia will really start using the pcie 3.0 speeds.
Is it really that hard to understand this?
Vishera cpu might be the last one.and they will see if they are making any noticeable progress in the desktop area.
All will come down at the end of 2012.then they will decide.
 
Last edited:
biggston said:
You people don't really get it do you?
AMD is fighting both sides here.nvidia and intel.
Its all about money here.AMD is really short on funds now.
They might drop the desktop market completely.
And before you come here and complain about pcie 3.0,
amd will bring pcie 3.0 when nvidia will really start using the pcie 3.0 speeds.
Is it really that hard to understand this?
Vishera cpu might be the last one.and they will see if they are making any noticeable progress in the desktop area.
All will come down at the end of 2012.then they will decide.
Click to expand...

Again, AMD is NOT dropping out of the desktop market, remember the FX series isn't the only desktop chips currently in production, just like Vishera will not be the only desktop line in 2012 (means there will be a Trinity-based desktop platform, why else would there be 100W APUs).

What is happening, is that AMD isn't pursuing the high-end desktop sector as aggressively as before. It is possible that PCI-E 3.0 could launch with Trinity/FM2.

Though, I do find it interesting that there was no talk/slides dealing with the respective platforms and features for the upcoming processors.
 
biggston said:
You people don't really get it do you?
AMD is fighting both sides here.nvidia and intel.
Its all about money here.AMD is really short on funds now.
They might drop the desktop market completely.
And before you come here and complain about pcie 3.0,
amd will bring pcie 3.0 when nvidia will really start using the pcie 3.0 speeds.
Is it really that hard to understand this?
Vishera cpu might be the last one.and they will see if they are making any noticeable progress in the desktop area.
All will come down at the end of 2012.then they will decide.
Click to expand...

i think you'll find that it is you that doesn't get it.

amd is relying on HSA to carve a place for itself, and that 100% demands high-bandwidth & low-latency links between the CPU and GPU.

sepang was supposed to bring on-die PCIe 3.0 for a reason.
 
Blacklash said:
It may be interesting. Guess we will see.

Click to expand...

Uh... discrete graphics in the FX chips? With the FX-8120 and FX-8150 due to be released in Q2 2012? I don't think that chart is accurate at all.
 
JMccovery said:
Again, AMD is NOT dropping out of the desktop market, remember the FX series isn't the only desktop chips currently in production, just like Vishera will not be the only desktop line in 2012 (means there will be a Trinity-based desktop platform, why else would there be 100W APUs).

What is happening, is that AMD isn't pursuing the high-end desktop sector as aggressively as before. It is possible that PCI-E 3.0 could launch with Trinity/FM2.

Though, I do find it interesting that there was no talk/slides dealing with the respective platforms and features for the upcoming processors.
Click to expand...

You might be correct about this.We will see what kind of plans AMD has.
I was referring at the high-end sector.
Sure i know about the apu.i am just talking about high-end stuff
 
Last edited:
Tsumi said:
Uh... discrete graphics in the FX chips? With the FX-8120 and FX-8150 due to be released in Q2 2012? I don't think that chart is accurate at all.
Click to expand...

Not IN, but alongside, you know like it is currently... Separate Radeon 7K cards.
 
Tsumi said:
Did you even look at the picture?
Click to expand...

EDIT: I think I understand what you are getting at. The processors with the dotted lines around them are the ones that are launching or about to be launched. It doesn't say that the FX x1xx chips will launch in Q2, the image just shows what will constitute those platforms.

AMD may mean this:
1C = Early 2012
2C = Mid 2012
3C = Late 2012

The segments show what is a minimal recommended configuration (as the low-end 7K cards may not be out by then).
 
Last edited:
You must log in or register to reply here.
Back
Top