Nvidia working on AMD SAM GPU performance boost facsimile

[HEXUS]

However its PCIe bus tech will accelerate its GPUs in both AMD and Intel CPU systems.

Read more.

[ET3D]

That's good to see. We're used to seeing NVIDIA show custom solutions and AMD go for open standards, but here it seems like AMD locked this solution to Ryzen 5000 for no good reason but marketing. If NVIDIA presents something that works on all CPUs, perhaps we'll see AMD expand the use of SAM to at least Ryzen 3000.

[Nifl]

So Nvidia is trying to say they have been already working on it and have not released or announced it yet even though their R&D budget is way bigger? I smell BS.

[AGTDenton]

I smell BS.

AKA fear of finally losing a sizable market share.


Even if nVidia reacts, they're still going to lose bang for buck
One thing is good, however, competition, competition, competition! In the style of Ballmer

[QuorTek]

So if AMD get the same boost on workstations and servers and super computers, where money is earned... then it can be grim for both NVIDIA and Intel both in one, however Intel is working on GFX now as well so both the big x86 companies has GFX.

[Spud1]

One thing is good, however, competition, competition, competition!

Exactly. That's the positive here - as one of the companies adds something different, the others react with an equivalent and continues onwards improving the products for all of us no matter which manufacturer you buy

SAM itself isn't massively interesting given how meager the gains seem to be in the real world, but every frame counts!

[mattburnzy]

That's good to see. We're used to seeing NVIDIA show custom solutions and AMD go for open standards, but here it seems like AMD locked this solution to Ryzen 5000 for no good reason but marketing. If NVIDIA presents something that works on all CPUs, perhaps we'll see AMD expand the use of SAM to at least Ryzen 3000.

Has this been confirmed that its just marketing?

[LSG501]

Does seem a little strange that they haven't already implemented something that is in the pcie spec if it gave such benefits to performance....5-10% performance gains isn't exactly small when it's essentially just code that's being changed, but then I suppose that's what happens when you sit on your laurels due to no real competition....

[DanceswithUnix]

So Nvidia is trying to say they have been already working on it and have not released or announced it yet even though their R&D budget is way bigger? I smell BS.

They are bound to have been working on it, this has been a part of the PCIe specification for years. So at some point Intel CPUs are bound to support it, and Nvidia cards.

AIUI you need both CPU support as that is the PCIe root bridge and BIOS support to set up the cards. That's before the drivers can start supporting it, and there is probably some OS support needed as well.

Looking on my machine here on Linux using "lspci -v" as part of the output I can see:

0c:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Vega 10 XL/XT [Radeon RX Vega 56/64] (rev c3) (prog-if 00 [VGA controller])
        Subsystem: ASUSTeK Computer Inc. Device 0555
        Flags: bus master, fast devsel, latency 0, IRQ 85
        Memory at e0000000 (64-bit, prefetchable) [size=256M]
        Memory at f0000000 (64-bit, prefetchable) [size=2M]
        I/O ports at d000 [size=256]
        Memory at fcc00000 (32-bit, non-prefetchable) [size=512K]
        Expansion ROM at 000c0000 [disabled] [size=128K]
        Capabilities: [48] Vendor Specific Information: Len=08 <?>
        Capabilities: [50] Power Management version 3
        Capabilities: [64] Express Legacy Endpoint, MSI 00
        Capabilities: [a0] MSI: Enable+ Count=1/1 Maskable- 64bit+
        Capabilities: [100] Vendor Specific Information: ID=0001 Rev=1 Len=010 <?>
        Capabilities: [150] Advanced Error Reporting
        Capabilities: [200] Resizable BAR <?>
        Capabilities: [270] Secondary PCI Express
        Capabilities: [2a0] Access Control Services
        Capabilities: [2b0] Address Translation Service (ATS)
        Capabilities: [2c0] Page Request Interface (PRI)
        Capabilities: [2d0] Process Address Space ID (PASID)
        Capabilities: [320] Latency Tolerance Reporting
        Kernel driver in use: amdgpu
        Kernel modules: amdgpu

Note in the middle of the "Capabilities" there lurks a resizable BAR" yet only has a mapped memory region of 256M so doesn't appear to be in use. That's on a Vega 56, so not exactly new, but on an X470 motherboard. Connecting to my server which has a B550 chipset which should support it, the GPU in there is an old R7 360 I had kicking around and that doesn't appear to have a resizable BAR so I can't see if support should work there.

With the Zen 3 and RDNA 2 launches so close AMD have a lot of control over the whole ecosystem, so it seems a good opportunity to get this stuff out there and properly working. Hopefully now AMD has made some noise on this, support will spread.

[Tabbykatze]

That's good to see. We're used to seeing NVIDIA show custom solutions and AMD go for open standards, but here it seems like AMD locked this solution to Ryzen 5000 for no good reason but marketing. If NVIDIA presents something that works on all CPUs, perhaps we'll see AMD expand the use of SAM to at least Ryzen 3000.

They didn't "lock" it down for no good reason, the main reason they were able to do it is that they have both a CPU and GPU division that they can work on internally for it. The actual next step for this is to be CXL which is a vendor ambiguous technology in PCIe 5.0 and above.

AMD would have to tip off both Nvidia and Intel what they are doing and with their financial horsepower would overcome what AMD is trying to be a forward player in the market.

"no good reason" is disingenuous to why they did it at best, it doesn't seem like it's just a microcode update and boom the CPU can now address all of GPU memory instead of the 256MB BAR of PCIe, it requires a little bit more. As AMD has both CPU and GPU, worked a ton on CCIX (competitor to CXL which lost), they were in a prime opportunity to perform acceleration in this area.

[Iota]

Does seem a little strange that they haven't already implemented something that is in the pcie spec if it gave such benefits to performance....5-10% performance gains isn't exactly small when it's essentially just code that's being changed, but then I suppose that's what happens when you sit on your laurels due to no real competition....

I'd probably think there is something stopping them from doing so, otherwise they would have done this already. Perhaps Intel doesn't play nicely, or it's unstable etc etc.

[Sumanji]

Hoping this forces AMD's hand into enabling SAM on older chipsets and CPU's... actually mildly annoyed at AMD if SAM was an attempt to artificially wall off a needle-moving tech to their newer kit...

[LSG501]

I'd probably think there is something stopping them from doing so, otherwise they would have done this already. Perhaps Intel doesn't play nicely, or it's unstable etc etc.

Could be compatibility issues but, and you can call me cynical here, they've just had no real competition and sat on it for just this occasion now they've got competition. They'll get it out quick smart but notice how the news has been timed to make people question if AMD is the best choice again... admittedly neither's a good choice at the moment because there's basically no stock lol

[lumireleon]

QUESTION: can the AMD cpu use the GDDR6 as RAM for other compute tasks instead of gaming??

[Tabbykatze]

QUESTION: can the AMD cpu use the GDDR6 as RAM for other compute tasks instead of gaming??

Could they not before?

What do you mean by your question?

[Corky34]

Note in the middle of the "Capabilities" there lurks a resizable BAR" yet only has a mapped memory region of 256M so doesn't appear to be in use. That's on a Vega 56, so not exactly new, but on an X470 motherboard. Connecting to my server which has a B550 chipset which should support it, the GPU in there is an old R7 360 I had kicking around and that doesn't appear to have a resizable BAR so I can't see if support should work there.

It also says it couldn't completely decode that register, i guess because AMD hasn't actually told anyone.