What are AMD Teasing?
- Thread starter WYP
- Start date
SPS
Moderator
No the bottleneck is still very much there with one card, however, the reason you don't experience it so much as a consumer is because devs avoid it like the plague. Any graphics/GPGPU programmer knows that data should rarely be transferred between the CPU and GPU. Eliminating this bottleneck would be awesome for games programmers because it allows us to use the GPU for more highly parallel tasks where computed data is required by the CPU every frame.
shambles1980
New member
the last few new things have been refreshes..
but i would imagine its an apu.
but here is hoping that its not a cpu or apu. and its a good performance gpu
BigBlue1
Banned
This is true, as we have seen with Mantle there are benefits to be had by removing the bottleneck even with a single card. With Mantle, DX11.3, DX12, NVLink and whatever the AMD equivalent will be there are bound to be some good advancements
There is no AMD equivalent...at least not until they buy an nVidia one and reverse engineer it like they do with all the other advancements in the market. lol
shambles1980
New member
those are mostly api limits that are being removed though.
you cant really call that a bottleneck "well i couldnt"
to me a bottleneck on the pcie would be the card wants to use 32Gb/s but the pcie lanes provided can only supply 16Gb/s which would effectivly bottleneck the card.
with an api im not sure what id call it. "bloated? slow down?" dunno something like that. and its usually mostly because the api is making checks and dumping things out for you to try and make sure you dont have issues.
dx 12 seems to have said Direct x will no longer do these checks for you. so when you program for dx 12 make sure that you load and unload properly..
this gives a more direct access to the programer and speeds up the whole api. but may be a bit more involved for the programers.
still that may not even be what your talking about. but from scimming over bigblue's posts and reading the replies i am assuming im atleast in the same general area of conversation.
you cant really call that a bottleneck "well i couldnt"
to me a bottleneck on the pcie would be the card wants to use 32Gb/s but the pcie lanes provided can only supply 16Gb/s which would effectivly bottleneck the card.
with an api im not sure what id call it. "bloated? slow down?" dunno something like that. and its usually mostly because the api is making checks and dumping things out for you to try and make sure you dont have issues.
dx 12 seems to have said Direct x will no longer do these checks for you. so when you program for dx 12 make sure that you load and unload properly..
this gives a more direct access to the programer and speeds up the whole api. but may be a bit more involved for the programers.
still that may not even be what your talking about. but from scimming over bigblue's posts and reading the replies i am assuming im atleast in the same general area of conversation.
SieB
New member
Yeah, sure....
SPS
Moderator
CPU/GPU data transfer is a hardware bottleneck. It's the slowest part.
This is true, as we have seen with Mantle there are benefits to be had by removing the bottleneck even with a single card. With Mantle, DX11.3, DX12, NVLink and whatever the AMD equivalent will be there are bound to be some good advancements
But the bottleneck in batches/draw calls is to do with CPU load, not bus bottlenecks.
So mantle etc are about reducing cpu load proper, or doing more effective multi threading.
Bus bandwidth on pci bottlenecks, I'm less sure about.
Does anyone actually know the frame by frame pcie data traffic loads and what it consists of and why more bandwidth alleviates a current measurable bottleneck?
I'm not sure I understand it right, but is local vram used for storing textures and geometry as well as frame buffers etc, or is texture/geometry pushed to gpu from system ram per frame?
From tests I gathered it was the former, in which case in not sure why it'd be a huge issue with current bus bandwidth.
If its the latter then I'm not sure but we'd probably be bus bandwidth bound constantly which isn't the case!
Hence my confusion on that point hehe
SieB
New member
But the bottleneck in batches/draw calls is to do with CPU load, not bus bottlenecks.
So mantle etc are about reducing cpu load proper, or doing more effective multi threading.
Bus bandwidth on pci bottlenecks, I'm less sure about.
Does anyone actually know the frame by frame pcie data traffic loads and what it consists of and why more bandwidth alleviates a current measurable bottleneck?
I'm not sure I understand it right, but is local vram used for storing textures and geometry as well as frame buffers etc, or is texture/geometry pushed to gpu from system ram per frame?
From tests I gathered it was the former, in which case in not sure why it'd be a huge issue with current bus bandwidth.
If its the latter then I'm not sure but we'd probably be bus bandwidth bound constantly which isn't the case!
Hence my confusion on that point hehe
To be honest I don't fully understand it myself :lol:
But basically to my understanding, and I (probably am) could be wrong.
The the bottleneck is between the GPU and CPU, the CPU has to work harder limiting how well the GPU can perform. With the bottleneck removed things that usually get used by the CPU can be offloaded onto the GPU increasing performance.
Ah I think that description sounds like batch prep overhead. So its not the bus bandwidth at fault but CPU physical load at fault.
Old API are not so efficient so CPU load is high, nor are those API good at load spreading on Multi core processors.
So basically cpu is doing loads of jobs, including frame prep for gpu. But its likely running on one core with other tasks too so suddenly when the cpu core is flat out the gpu is waiting on the cpu doing x frame preps per second.
This is why OC 2500k at high mhz are still so good for games performance.
Mantle is more efficient. Not sure how, but prob a combo of good multi threading AND off loading even more batch tasks from cpu AND just more efficient processing code.
All of a sudden we could find an i7 better for gaming.
But also gpu speed will be more the core factor for fps, with the need to match cpu to gpu less important bs currently.
No idea on pci bus bandwidth but I'm guessing its no more an issue than anything else... it just tends to creep up in line with vram size and so on.
Hmmm, will try find out for sure on the latter and let you know my findings hehe
Old API are not so efficient so CPU load is high, nor are those API good at load spreading on Multi core processors.
So basically cpu is doing loads of jobs, including frame prep for gpu. But its likely running on one core with other tasks too so suddenly when the cpu core is flat out the gpu is waiting on the cpu doing x frame preps per second.
This is why OC 2500k at high mhz are still so good for games performance.
Mantle is more efficient. Not sure how, but prob a combo of good multi threading AND off loading even more batch tasks from cpu AND just more efficient processing code.
All of a sudden we could find an i7 better for gaming.
But also gpu speed will be more the core factor for fps, with the need to match cpu to gpu less important bs currently.
No idea on pci bus bandwidth but I'm guessing its no more an issue than anything else... it just tends to creep up in line with vram size and so on.
Hmmm, will try find out for sure on the latter and let you know my findings hehe
Wraith
Bettyswollocks
Yup spot on, but you also missed speed bottlenecking
caused by having a crap motherboard (slow bus) or slow RAM. It's all about the bandwidths. CPU frequency, RAM frequency, GPU/DRAM frequencies, motherboard bus frequencies. There will always be a bottleneck somewhere.
Yup spot on, but you also missed speed bottlenecking
It's all about the bandwidths. CPU frequency, RAM frequency, GPU/DRAM frequencies, motherboard bus frequencies. There will always be a bottleneck somewhere.
Drive speeds too? I always wondered how an SSD made such a difference when a normal mech shouldn't really be the bottleneck anyway.
Wraith
Bettyswollocks
Very good point, thanks for that addition I completely forgot about SATA II/III/Express speeds, what about the new M.2 PCIe.... :huh:
Well its not spot on in the case of batch prep.
The bottleneck IS the CPU, not the bus.
That is why clocked low core count gpus are vogue for uber fast gaming machines.
But as said that may all suddenly change with Mantle etc... an i7 3.5ghz might blow a 5.0ghz 2500k out of the water on uber high power GPU systems.
I get your general point, but its an important differentiation in this case. These new API will change a lot about spec/balance of cpu vs gpu moving forward!
PCIe bus, as far as I can see isn't slowing down much at all on any half reasonably balanced system... And generally systems are balanced. Ie no C2D on z99 boards, or voodoo 2 cards on pcie etc hehe
Bottlenecks vary in size and user significance though.
No system will be perfectly balanced because of that.
What matters are ones that impact you.
Generally the nasty one that does is the one being addressed now with Mantle etc... CPU physical overhead.
Imo most other stuff is way beyond being a bottleneck than it ever used to be.
Ie the same mainstream mobo I run can also a bunch of titan gpu with oodles more VRAM and fps etc. No mobo bottleneck there for anyone spending either £50 to £2500 on Gpu hehe
A good gaming PC is cheaper and faster than ever, most bottlenecks are gone and for most gamers the limit is their wallet not hardware... Well it will be soon. Till we go to 4k mainstream 3D 144hz vr that is haha
If AMD can get Mantle games out there on a new faster GPU and require their gamers to only need a modest cpu to boot, saving them £££ on total system costs for same perf as nvidia, they'll be onto a winner.
Mantle etc are the keys to the biggest and Imo only significant bottleneck we've seen for a while.
Dave
The bottleneck IS the CPU, not the bus.
That is why clocked low core count gpus are vogue for uber fast gaming machines.
But as said that may all suddenly change with Mantle etc... an i7 3.5ghz might blow a 5.0ghz 2500k out of the water on uber high power GPU systems.
I get your general point, but its an important differentiation in this case. These new API will change a lot about spec/balance of cpu vs gpu moving forward!
PCIe bus, as far as I can see isn't slowing down much at all on any half reasonably balanced system... And generally systems are balanced. Ie no C2D on z99 boards, or voodoo 2 cards on pcie etc hehe
Bottlenecks vary in size and user significance though.
No system will be perfectly balanced because of that.
What matters are ones that impact you.
Generally the nasty one that does is the one being addressed now with Mantle etc... CPU physical overhead.
Imo most other stuff is way beyond being a bottleneck than it ever used to be.
Ie the same mainstream mobo I run can also a bunch of titan gpu with oodles more VRAM and fps etc. No mobo bottleneck there for anyone spending either £50 to £2500 on Gpu hehe
A good gaming PC is cheaper and faster than ever, most bottlenecks are gone and for most gamers the limit is their wallet not hardware... Well it will be soon. Till we go to 4k mainstream 3D 144hz vr that is haha
If AMD can get Mantle games out there on a new faster GPU and require their gamers to only need a modest cpu to boot, saving them £££ on total system costs for same perf as nvidia, they'll be onto a winner.
Mantle etc are the keys to the biggest and Imo only significant bottleneck we've seen for a while.
Dave
Last edited by a moderator:
nepas
New member
Stop trolling,there is no need to try and start any flame war(and you just look stupid as feck):huh:
Hmm an intel 4790k and a gpu capable of performing similar to the 290x/780ti SOC WOOT! Can only wish.
Anyways on a more serious note, I still dont understand why people are causing such a big ruckus over AMD being late the the game, its as simple as dont want to wait, go Nvidia. its really up to you... AMD is generally only ever going to be a product that has more value as opposed to performance from the Nvidia equivalent, this is something that we have seen.
Anyways on a more serious note, I still dont understand why people are causing such a big ruckus over AMD being late the the game, its as simple as dont want to wait, go Nvidia. its really up to you... AMD is generally only ever going to be a product that has more value as opposed to performance from the Nvidia equivalent, this is something that we have seen.
Last edited:
RizeAllard
New member
I hope we will see a brand new gpu which can compete with the new flagship of Nvidia in the terms of energy efficency mainliy. But i would really like to hear about some new cpu (apu) architecture as well
Waiting for the wednesday with high hopes!
http://www.youtube.com/watch?v=C6GSQTobbQk
But i would really like to hear about some new cpu (apu) architecture as well Waiting for the wednesday with high hopes!
http://www.youtube.com/watch?v=C6GSQTobbQk
