8 ms·

AMD's Hidden Strategy

30 points by x0054 9y ago

anonymousab 9y ago
Pressed the button to show the full article and was greeted with a full screen demand to download their stock news app. > Welcome to Seeking Alpha, please complete your setup with our FREE app Do the metrics really say that these approaches gain more users than they lose?
- dvfjsdhgfv 9y ago
  Maybe they had this switched off but just switched it on when they submitted their website to HN?
- brador 9y ago
  You can't measure lost users, only installs. 1 > 0 so it's inevitable to show an increased number of installs with popup vs without.
  - sgarman 9y ago
    You can with A/B testing though?
- PascLeRasc 9y ago
  That and breaking the article into 6 pages. If NYT, WaPo, and Medium can fit longform text into one page this site can too.
- DrScump 9y ago
  I get a popup for "Register for FREE to finish reading this article" when trying to go beyond page 2 (I didn't).
hnov 9y ago
AMP version without login required https://www.google.com/amp/s/seekingalpha.com/amp/article/4142634-amds-hidden-strategy https://www.google.com/amp/s/seekingalpha.com/amp/article/41...
- netsharc 9y ago
  It un-amps itself when I load this. Even the print trick doesn't work, it just prints the current page, so for a 6-page article you'd have to print 6 times (and each virtual page probably takes up 1 full paper page + 1 line of a 2nd paper page...). Although, who still prints articles anyway? Guess whose article won't be read past page 1...
  - LamaOfRuin 9y ago
    Firefox reader view.
jitl 9y ago
Here’s the tl;dr: On the server, AMD is making a security play w/ EPYC arch. - AMD not vulnerable to Meltdown - SEV allows encryption of a VM’s memory, further mitigating read attacks On the consumer side, AMD is selling on-package Vega to Intel. This should increase market penetration, and thus get more games optimized for the AMD arch.
- dogma1138 9y ago
  It claims that Meltdown can be used to read from other memory address spaces and break cross VM boundaries which is categorically false. Spectre v2 does allow you to execute intra VM and VM to host attacks Meltdown does not. SeV will not help in this regard. Games are already optimized for AMD GPUs and pretty darn well due to consoles I’m not entirely sure how much difference would the KBL-G provide considering that at least in the mobile space it’s going after the MX150 it needs 100W to compete with the 1050 so that would be on desktop only.
  - unitboolean 9y ago
    May be he meant docker instances? I heard that Meltdown allows one docker instance to access another one even without permission.
    - dogma1138 9y ago
      Don’t know what they meant but it surely doesn’t affect the serverless model or PASS/IASS nearly as bad as Spectre. Spectre is also much much harder to mitigate against Meltdown sucked but it was easily fixed and it will also be easily fixed in silicon Spectre especially variant 1 will be with us for a long time and no one knows what still is lying hidden. AMDs main advantage is that none of the researchers really bothered reversing their branch predictor so none of the BP training reliant PoCs work on their hardware but there is nothing in their design that mitigates against Spectre specifically. And judging by the guide they’ve released for how to manage branch prediction post Spectre at least the technical team is worried about what would happen when someone starts paying attention to them.
generic_T 9y ago
I found it very surprising that the article didn't mention machine learning applications. Machine learning and scientific computing is where a company like NVIDIA really shines. The reason for this is CUDA. Nearly all deep learning frameworks, such as tensorflow, use CUDA to run on GPUs. Until AMD comes up with their version of CUDA, and convinces the authors of deep learning frameworks to support it, they will be behind. NVIDIA.
- fulafel 9y ago
  AMD has been working on CUDA for a while now, it's called HIP. Here's Tensorflow for it: https://github.com/ROCmSoftwarePlatform/hiptensorflow https://github.com/ROCmSoftwarePlatform/hiptensorflow I have no experience so can't comment on how production-ready it is.
  - paulmd 9y ago
    HIP is fine for very basic programs but you have always been able to convert those pretty trivially with find-and-replace. The bigger problem has always been middleware. Effectively, HIP needs to duplicate everything that NVIDIA provides, 1:1, and it doesn't. More complex libraries (particularly ones that touch hardware features like texture units/etc) cannot be insta-converted with HIP either. It's impractical, for the most part. I've always been of the opinion that what AMD needs is something like GPU Ocelot that can transpile PTX rather than working at a source level. Unfortunately I'm sure that would be legally frought for AMD.
    - generic_T 9y ago
      I'm just getting started with machine learning and deep learning, so I'm certainly no expert. I wonder if building frameworks on top of OpenCL would be a viable alternative?
      - paulmd 9y ago
        Sure. And like I said, for relatively trivial stuff you can almost do a find-and-replace of CUDA keywords with OpenCL keywords and get something that will run with relatively minimal tweaking. The problem is the middleware - i.e. when you go to write a program, you want to write your financial simulation or whatever, not a RNG library. Once that stuff gets written, OpenCL will be perfectly viable. It's just sort of a slog now, vs a huge amount of stuff available for CUDA, both from NVIDIA and from others. Another, more general problem is that OpenCL is not really a GPGPU language, it's a general framework for dispatching tasks to heterogeneous processors. You can hypothetically use OpenCL to dispatch tasks to a FPGA or a DSP or anything else. This means there's a relatively silly amount of boilerplate compared to CUDA, which is stylistically annoying but not really complex. But in general it's also up to the vendor to implement their own SDK, compiler, etc, and there's no guarantee that what runs on one OpenCL platform will run on another. What is commonly referred to as "OpenCL" is actually the AMD APP SDK and the NVIDIA OpenCL SDK, which both implement their own version of parallel C++ dispatched according to the OpenCL function interface/semantics. So in other words, OpenCL suffers from a bit of Architecture Astronaut syndrome. AMD solved the problem of making a framework that can dispatch tasks to any coprocessor that implements their interface, NVIDIA solved the problem of making a GPGPU language. As such, AMD sometimes struggles to fulfill the cross-platform promise of their language and suffers in conciseness. For learning CUDA, I really like the Thrust language. You can write GPGPU programs that are literally 10 lines or less. It looks like regular C++ STL but you can execute it on the GPU. You can easily call __device__ functions using functors iterating across your dataset, or get raw_pointers into a device_vector and then pass them into a normal kernel launch, so there's a lot of flexibility there to tune stuff as you want to make it more complex. http://docs.nvidia.com/cuda/thrust/index.html http://docs.nvidia.com/cuda/thrust/index.html For a non-trivial program, here's my master's thesis, which I did in Thrust. Not really machine learning but at least you can see what you're getting into. https://github.com/pholvs/PandemicThrust https://github.com/pholvs/PandemicThrust For AMD, there is supposedly an equivalent called "Bolt Framework" but I have not used it personally. As is customary for AMD's libraries, after release it was abandoned and it has not been updated in several years, best of luck. https://hsa-libraries.github.io/Bolt/html/ https://hsa-libraries.github.io/Bolt/html/ https://github.com/HSA-Libraries/Bolt https://github.com/HSA-Libraries/Bolt
loeg 9y ago
Note that if you don't make it to page 6, there is a disclaimer that the author is long AMD.
rogerbinns 9y ago
Can anyone comment on mainline kernel drivers for an entire AMD based laptop? For example if you get a Thinkpad you get Intel CPU and GPU, but other parts like wifi and bluetooth are Intel too. All of these are covered by mainstream Linux kernel drivers. (I'm thinking of years past when Broadcom wifi etc were a problem under Linux.)
- mixedCase 9y ago
  I have a Kaveri era HP laptop (read: old). No issues with wifi. Fairly sure it had an Atheros chip which has amazing Linux support.

30 points by x0054 9y ago

Pressed the button to show the full article and was greeted with a full screen demand to download their stock news app. > Welcome to Seeking Alpha, please complete your setup with our FREE app Do the metrics really say that these approaches gain more users than they lose?

dvfjsdhgfv 9y ago
Maybe they had this switched off but just switched it on when they submitted their website to HN?
brador 9y ago
You can't measure lost users, only installs. 1 > 0 so it's inevitable to show an increased number of installs with popup vs without.
- sgarman 9y ago
  You can with A/B testing though?
PascLeRasc 9y ago
That and breaking the article into 6 pages. If NYT, WaPo, and Medium can fit longform text into one page this site can too.
DrScump 9y ago
I get a popup for "Register for FREE to finish reading this article" when trying to go beyond page 2 (I didn't).

hnov 9y ago

AMP version without login required https://www.google.com/amp/s/seekingalpha.com/amp/article/4142634-amds-hidden-strategy https://www.google.com/amp/s/seekingalpha.com/amp/article/41...

netsharc 9y ago
It un-amps itself when I load this. Even the print trick doesn't work, it just prints the current page, so for a 6-page article you'd have to print 6 times (and each virtual page probably takes up 1 full paper page + 1 line of a 2nd paper page...). Although, who still prints articles anyway? Guess whose article won't be read past page 1...
- LamaOfRuin 9y ago
  Firefox reader view.

jitl 9y ago

Here’s the tl;dr: On the server, AMD is making a security play w/ EPYC arch. - AMD not vulnerable to Meltdown - SEV allows encryption of a VM’s memory, further mitigating read attacks On the consumer side, AMD is selling on-package Vega to Intel. This should increase market penetration, and thus get more games optimized for the AMD arch.

dogma1138 9y ago
It claims that Meltdown can be used to read from other memory address spaces and break cross VM boundaries which is categorically false. Spectre v2 does allow you to execute intra VM and VM to host attacks Meltdown does not. SeV will not help in this regard. Games are already optimized for AMD GPUs and pretty darn well due to consoles I’m not entirely sure how much difference would the KBL-G provide considering that at least in the mobile space it’s going after the MX150 it needs 100W to compete with the 1050 so that would be on desktop only.
- unitboolean 9y ago
  May be he meant docker instances? I heard that Meltdown allows one docker instance to access another one even without permission.
  - dogma1138 9y ago
    Don’t know what they meant but it surely doesn’t affect the serverless model or PASS/IASS nearly as bad as Spectre. Spectre is also much much harder to mitigate against Meltdown sucked but it was easily fixed and it will also be easily fixed in silicon Spectre especially variant 1 will be with us for a long time and no one knows what still is lying hidden. AMDs main advantage is that none of the researchers really bothered reversing their branch predictor so none of the BP training reliant PoCs work on their hardware but there is nothing in their design that mitigates against Spectre specifically. And judging by the guide they’ve released for how to manage branch prediction post Spectre at least the technical team is worried about what would happen when someone starts paying attention to them.

generic_T 9y ago

I found it very surprising that the article didn't mention machine learning applications. Machine learning and scientific computing is where a company like NVIDIA really shines. The reason for this is CUDA. Nearly all deep learning frameworks, such as tensorflow, use CUDA to run on GPUs. Until AMD comes up with their version of CUDA, and convinces the authors of deep learning frameworks to support it, they will be behind. NVIDIA.

fulafel 9y ago
AMD has been working on CUDA for a while now, it's called HIP. Here's Tensorflow for it: https://github.com/ROCmSoftwarePlatform/hiptensorflow https://github.com/ROCmSoftwarePlatform/hiptensorflow I have no experience so can't comment on how production-ready it is.
- paulmd 9y ago
  HIP is fine for very basic programs but you have always been able to convert those pretty trivially with find-and-replace. The bigger problem has always been middleware. Effectively, HIP needs to duplicate everything that NVIDIA provides, 1:1, and it doesn't. More complex libraries (particularly ones that touch hardware features like texture units/etc) cannot be insta-converted with HIP either. It's impractical, for the most part. I've always been of the opinion that what AMD needs is something like GPU Ocelot that can transpile PTX rather than working at a source level. Unfortunately I'm sure that would be legally frought for AMD.
  - generic_T 9y ago
    I'm just getting started with machine learning and deep learning, so I'm certainly no expert. I wonder if building frameworks on top of OpenCL would be a viable alternative?
    - paulmd 9y ago
      Sure. And like I said, for relatively trivial stuff you can almost do a find-and-replace of CUDA keywords with OpenCL keywords and get something that will run with relatively minimal tweaking. The problem is the middleware - i.e. when you go to write a program, you want to write your financial simulation or whatever, not a RNG library. Once that stuff gets written, OpenCL will be perfectly viable. It's just sort of a slog now, vs a huge amount of stuff available for CUDA, both from NVIDIA and from others. Another, more general problem is that OpenCL is not really a GPGPU language, it's a general framework for dispatching tasks to heterogeneous processors. You can hypothetically use OpenCL to dispatch tasks to a FPGA or a DSP or anything else. This means there's a relatively silly amount of boilerplate compared to CUDA, which is stylistically annoying but not really complex. But in general it's also up to the vendor to implement their own SDK, compiler, etc, and there's no guarantee that what runs on one OpenCL platform will run on another. What is commonly referred to as "OpenCL" is actually the AMD APP SDK and the NVIDIA OpenCL SDK, which both implement their own version of parallel C++ dispatched according to the OpenCL function interface/semantics. So in other words, OpenCL suffers from a bit of Architecture Astronaut syndrome. AMD solved the problem of making a framework that can dispatch tasks to any coprocessor that implements their interface, NVIDIA solved the problem of making a GPGPU language. As such, AMD sometimes struggles to fulfill the cross-platform promise of their language and suffers in conciseness. For learning CUDA, I really like the Thrust language. You can write GPGPU programs that are literally 10 lines or less. It looks like regular C++ STL but you can execute it on the GPU. You can easily call __device__ functions using functors iterating across your dataset, or get raw_pointers into a device_vector and then pass them into a normal kernel launch, so there's a lot of flexibility there to tune stuff as you want to make it more complex. http://docs.nvidia.com/cuda/thrust/index.html http://docs.nvidia.com/cuda/thrust/index.html For a non-trivial program, here's my master's thesis, which I did in Thrust. Not really machine learning but at least you can see what you're getting into. https://github.com/pholvs/PandemicThrust https://github.com/pholvs/PandemicThrust For AMD, there is supposedly an equivalent called "Bolt Framework" but I have not used it personally. As is customary for AMD's libraries, after release it was abandoned and it has not been updated in several years, best of luck. https://hsa-libraries.github.io/Bolt/html/ https://hsa-libraries.github.io/Bolt/html/ https://github.com/HSA-Libraries/Bolt https://github.com/HSA-Libraries/Bolt

loeg 9y ago

Note that if you don't make it to page 6, there is a disclaimer that the author is long AMD.

rogerbinns 9y ago

Can anyone comment on mainline kernel drivers for an entire AMD based laptop? For example if you get a Thinkpad you get Intel CPU and GPU, but other parts like wifi and bluetooth are Intel too. All of these are covered by mainstream Linux kernel drivers. (I'm thinking of years past when Broadcom wifi etc were a problem under Linux.)

mixedCase 9y ago
I have a Kaveri era HP laptop (read: old). No issues with wifi. Fairly sure it had an Atheros chip which has amazing Linux support.