AMD Piledriver rumours ... and expert conjecture

Page 204 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
Reynod

Reynod

Administrator
Mar 26, 2007
14,921
16
42,215
We have had several requests for a sticky on AMD's yet to be released Piledriver architecture ... so here it is.

I want to make a few things clear though.

Post a question relevant to the topic, or information about the topic, or it will be deleted.

Post any negative personal comments about another user ... and they will be deleted.

Post flame baiting comments about the blue, red and green team and they will be deleted.

Enjoy ...
 
Sort by date Sort by votes
When an APU is CF'd with a dedicated gpu, does it matter if the dgpu has gddr3 or gddr5 memory? I need a DEFINITIVE answer from somewhere credible, maybe even from AMD themselves(I've been trying to find an answer on da googles forever).
 
Upvote 0 Downvote
Nope..............either way, both types will have to downclock significantly to be in tune with the APU memory, so in such a situation just stick to using the discrete only.

No point buying a powerful low end, just to anchor it down with your IGP
Click to expand...

Another question: What are the differences between asymmetric crossfire and 'regular' crossfire?
 
Upvote 0 Downvote
Nope..............either way, both types will have to downclock significantly to be in tune with the APU memory, so in such a situation just stick to using the discrete only.

No point buying a powerful low end, just to anchor it down with your IGP
Click to expand...


What the ... f......

Ok going to try to be polite and just say this is a crock of sh!t.

The dGPU does NOT "slow itself down" to match the APU. I know because unlike the above posters I actually have an A8 based laptop that contains an A8-3550MX and a 7690M. With ACF activated the 7690M DOES NOT clock down, its working 100%. What the ATI driver does is a form of non-uniform load balancing, it treats the two GPU's the same as your OS treats two CPU "cores". It will place load on the dGPU primary and when the dGPU is at 100% it will try place additional load on the APU.

At worst you get no additional performance benefit, at best you get a slight increase. The only time I've ever even read about "worse" performance is when there was a driver problem and the issue was solved with either updated drivers or downloading newer capability profiles.

The above posters are speaking either out of ignorance of malice, don't know which.
 
Upvote 0 Downvote
Now for the technical difference between CFX and ACF. Both are "crossfire" but operate in radically different modes.

CFX is a straight 50/50 split of the graphics workload between both GPUs. It does this in a similar way as SLI, by emulating a single larger GPU with double the graphics resources.

ACF use's the name crossfire but is completely different. Instead of emulating a single large GPU it instead tries to dynamically split the workload between the two target GPUs similar to how OS's split threads between multiple CPUs. The GPU's are only required to use the same architecture (VLIW5 / VLIW4 / GCN) and have the same amount of memory storage available. The graphics driver will attempt to dynamically load balance each GPU with the stronger one receiving the larger load. This load balancing can cause issues with DX9 but works extremely well in DX10/DX11.

-=Edit=-

Forgot to mention it's now called "Dual Graphics" when doing ACF between an APU and a dGPU. It's been optimized at the driver level to work best on DX 10/11 products, it's so-so on DX9 products.
 
Upvote 0 Downvote
I thought the onboard graphics processor did the 2D and the discrete GPU did the 3d?
Click to expand...


No, and more no.

APU's are just a low powered Radeon GPU glued to a low power Phenom II (Trinity is Bulldozer-E / Piledriver) core. It is a fully capable graphics processor and can be treated as such.

How are people just inventing stuff then taking it as fact?

Not a dig at you smp but I see so much false info passed off as fact from the blue face painted crowd that it's appalling. Making something up, then passing it around as fact to everyone you (pejorative) does not make something true.
 
Upvote 0 Downvote
I got confused
with Hybrid SLI I believe that the mGPU handles 2d and the discrete handles 3D

"HybridPower
HybridPower unleashes graphics performance when needed and switches to quiet, low-power quiet operation for everyday computing. Combine an NVIDIA Hybrid SLI-enabled GPU into a NVIDIA Hybrid SLI-enabled motherboard (mGPU) for the ultimate control. Dial up performance for demanding 3D games and applications; downshift to the mGPU to reduce noise and extended battery life for everyday computing tasks like browsing the Web, word processing, or watching High Definition videos."
source- http://www.nvidia.com/object/hybridsli_notebook.html

looks like AMD does it differently
that is why I added a question mark and didnt state it as fact
 
Upvote 0 Downvote
Again wasn't a dig at you, just too much false info being spread so wanted to nip it in the bud.

AMD calls it "Switchable Graphics", either fixed mode (older style) or dynamic (newer style).

When you start an application the control panel asks which power profile to assign that application to (Low / High) then remembers it for future reference. Low power profile is using the onboard APU as the graphics processor, High Power profile use's either the dGPU (ACF off) or both together (ACF on). In fixed mode you can assign which is the primary GPU from the BIOS, in dynamic mode it will assume the APU is primary and keep the dGPU powered down unless it's needed.

The 6620g is basically a Turks GPU configured as 400:20:8, runs at 444Mhz and supports up to DDR3-1600.

The biggest problem with the Llano APU's, and I doubt Trinity will be any different, is the OEM cooling solution provided. The chips maybe branded @45W but they can go a helluva lot higher. Using dynamic software overclocking you can scale it from 800Mhz to 3.0Mhz (any higher and it gets too hot for the notebook cooler), though the onboard iGPU is stuck at whatever speed the manufacturer sets it at. If AMD ever allows dynamic clock manipulation of the bonded iGPU they would have one amazing hobbyist chip.
 
Upvote 0 Downvote
Now for the technical difference between CFX and ACF. Both are "crossfire" but operate in radically different modes.

CFX is a straight 50/50 split of the graphics workload between both GPUs. It does this in a similar way as SLI, by emulating a single larger GPU with double the graphics resources.

ACF use's the name crossfire but is completely different. Instead of emulating a single large GPU it instead tries to dynamically split the workload between the two target GPUs similar to how OS's split threads between multiple CPUs. The GPU's are only required to use the same architecture (VLIW5 / VLIW4 / GCN) and have the same amount of memory storage available. The graphics driver will attempt to dynamically load balance each GPU with the stronger one receiving the larger load. This load balancing can cause issues with DX9 but works extremely well in DX10/DX11.

-=Edit=-

Forgot to mention it's now called "Dual Graphics" when doing ACF between an APU and a dGPU. It's been optimized at the driver level to work best on DX 10/11 products, it's so-so on DX9 products.
Click to expand...

So this means that gddr3 and gddr5 affects the performance when activating dual graphics?

Also, could you link me the 'A8-3550MX and a 7690M' laptop?
 
Upvote 0 Downvote
So this means that gddr3 and gddr5 affects the performance when activating dual graphics?

Also, could you link me the 'A8-3550MX and a 7690M' laptop?
Click to expand...

HP-DV6z-6c00 is my model number.

A8-3550MX
8GB DDR3-1333 memory (new HP bios is encrypted and has 1600 locked out)
ATI 7690M (rebadged 6770M).

Read again over what I wrote. Each GPU is treated differently and loaded until it's at 100%. Memory only effects that GPU's performance. The APU having DDR3 will have ZERO effect on the dGPU having GDDR5.
 
Upvote 0 Downvote
HP-DV6z-6c00 is my model number.

A8-3550MX
8GB DDR3-1333 memory (new HP bios is encrypted and has 1600 locked out)
ATI 7690M (rebadged 6770M).

Read again over what I wrote. Each GPU is treated differently and loaded until it's at 100%. Memory only effects that GPU's performance. The APU having DDR3 will have ZERO effect on the dGPU having GDDR5.
Click to expand...

Alright, thanks.
 
Upvote 0 Downvote
Just checked HP's website and I died a little inside.

They now have the DV6z-7000 but no DV7z variant.

CPU's include
A6-4400
A8-4500M
A10-4600M

dGPU's include (there is where I died inside)
7670M 1GB
7730M 2GB

And now DDR3-1600 support.

Can get an A10 + 7670M / 7730M with the 1920x1080 screen for under $900. Go with the 1366x768 screen and it's about $750.

I really hate how they didn't offer the 7750M which is the exact same chip as a 7730M but with GDDR5 memory. The 7730M is severely hampered by memory bandwidth as it's limited to DDR3 only. HP's higher models have better GPU offerings so their artificially keeping this as a low end model. From what I can find they don't offer a DV7z-7000 line, only the DV7t-7000's but there is a single DV7-7010 that has an A10 in it but there is no customize option. In the near future there should another model added.
 
Upvote 0 Downvote
No, and more no.

APU's are just a low powered Radeon GPU glued to a low power Phenom II (Trinity is Bulldozer-E / Piledriver) core. It is a fully capable graphics processor and can be treated as such.

How are people just inventing stuff then taking it as fact?

Not a dig at you smp but I see so much false info passed off as fact from the blue face painted crowd that it's appalling. Making something up, then passing it around as fact to everyone you (pejorative) does not make something true.
Click to expand...

Don't mind king smp, he doesn't even have a 6 core cpu.
 
Upvote 0 Downvote
I made it that way on purpose....
malmental style.. 😛
Click to expand...


You must leave your old Malmental ways behind you and embrace being Legendary

and anyway I did end my statement about Hybrid CFx with a question mark
I was thinking of notebooks and where they switch between onboard and discrete for 2d and 3d
doing that make sense but pairing an oboard with a discrete for 3d work doesnt seem like it would work well
Is there any benchmarks on using an ACF setup?
 
Upvote 0 Downvote
You must leave your old Malmental ways behind you and embrace being Legendary

and anyway I did end my statement about Hybrid CFx with a question mark
I was thinking of notebooks and where they switch between onboard and discrete for 2d and 3d
doing that make sense but pairing an oboard with a discrete for 3d work doesnt seem like it would work well
Is there any benchmarks on using an ACF setup?
Click to expand...


Tons. The effect varies depending on your program, as I've already explained. It works the same as having multiple CPUs.

In DX9 benchmarks my dGPU will be fully loaded with the 6620g spiking up on occasion, minimal gains but no lowering of performance. In DX10/11 benchmarks both GPUs are loaded and gains are noticeable, still not very significant but that's to be expected. The 6620g isn't exactly a powerful GPU and then 7690M is so much stronger that the iGPU only adds a few percentage points worth of power. Also the iGPU shares the same TDP as the CPU which limits performance unless you play with core speeds and know how to use K10stat.
 
Upvote 0 Downvote
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom