Separate names with a comma.
Discussion in 'MacBook Pro' started by pooprscooper, Jun 14, 2011.
Here's to hoping
Wow I actually just read that article before coming here. It sounds promising and an external gpu could mean good things to come but don't keep your hopes up.
Theoretically an external GPU is possible but we all know that it doesn't mean that it will actually happen. It has been tried before but it has always failed. In the end, the market for such product is most likely relatively small and making a such product (mainly the software side) can be very complicated.
Probably more possible with PCI-Express 3.0 and the strong likelihood of Thunderbolt getting the equivalent of a PCI-E 2.0 x8 slot which few GPUs out today fully saturate.
Thunderbolt 2.0 is rumored to be released in ~2015 with bandwidth of 50Gb/s. PCIe 3.0 shouldn't make Thunderbolt any faster.
TB's 4xPCIe bandwidth should be easily enough to do some epic GPGPU work, and support decent framerates at medium resolution, but at very high quality (ie, large shader computational requirements, but few pixels/textures/polygons pushed through the TB connection).
Easily enough to drive a laptop screen. Or run 2xTB in parallel on the iMac, and link it to a 30" ACD.
You do realize that the standard also depends on the speed of the PCI-Express bus, right?
Yes, but the current 10Gb/s can already be achieved using PCIe 2.0. TB controllers are connected to PCIe 2.0 x4 slots which provide up to 16Gb/s. As Thunderbolt tops out at ~900MB/s in real world, the controller is clearly the bottleneck because PCIe 2.0 could easily provide over 1GB/s (2GB/s theoretically).
That means PCIe 3.0 won't change anything. Until the controller is updated to support greater speeds, 10Gb/s is what we will get.
Thunderbolt runs off PCIe lanes though which can sometimes cause issues for other devices(GPU) using PCIe when maxing out thunderbolt.
Source? The GPU has eight PCIe 2.0 lanes providing up to 32Gb/s, regardless of how much bandwidth Thunderbolt is using.
I looked and couldn't find it yet, which I really want to find. It was some article with AMD testing thunderbolt who of course would be bashing it. If thunderbolt is using 20gb/s though that only leaves 12gb/s left which might cause issues with playing a game(3.7M pixels per 27") and using raid at the same time?
Isn't Thunderbolt basically an external PCI-E interface anyway?
Yes. It's basically DisplayPort 1.1a and external PCIe slammed into one interface.
+1 if you have no idea what the above conversation is about
Btw, Hellhammer is always right
Wirelessly posted (iPhone: Mozilla/5.0 (iPhone; U; CPU iPhone OS 4_3_3 like Mac OS X; it-it) AppleWebKit/533.17.9 (KHTML, like Gecko) Version/5.0.2 Mobile/8J2 Safari/6533.18.5)
Always. Plus, he likes black metal.
Obviously if you're trying to transfer large files through the Thunderbolt connection, and power a couple of large monitors, and pass game data to and from an external GPU, you'll have problems. That's kinda obvious. But why you would want to be saturating your TB bandwidth with lots of other random stuff while playing a game, I don't know.
I don't see why having a single external GPU on a TB interface would be any different to having an extra GPU in a PCIe 2 x4 slot.
As I would imagine you could plug an external monitor directly into said external GPU, I would imagine you could happily play at even very high resolutions on an external without any TB-bottlenecking, as the actual screen data would not have to be sent back through the TB connection.
Sorry, I didn't see your reply. You actually have a point. The CPU and PCH are connected to each other with 20Gb/s DMI. The CPU provides 16 PCIe 2.0 lanes and the chipset provides eight. In MBP, the GPU is connected using half of the CPU's lanes, which is 8 lanes and total bandwidth of 32Gb/s. The remaining eight lanes (or actually four lanes, as there are four empty lanes) are used for Thunderbolt. PCH's PCIe lanes are used for stuff like AirPort etc, though not all of them are in use.
When you connect the TB controller straight to the CPU's PCIe lanes, you eliminate the DMI bottleneck in-between because the transmission will be CPU-TB, not CPU-PCH-TB. If you connected the TB controller to the PCH's PCIe lanes, then you could drive into bandwidth issues if you had an external GPU in TB and e.g. RAID 0 setup in SATA ports (they share the 20Gb/s DMI bandwidth) and you were using both at full load simultaneously. If I recall correctly, iMacs TB implementation is connected to the PCH's PCIe lanes so that could potentially cause issues if all interfaces were heavily loaded at the same time.