Skip to main content

Tom's Hardware's AMA With AMD, In Its Entirety

The Reference Heat Sink, AGP, Frame Pacing, And APU13

Q. Why not use a more robust "reference design" for stock  coolers, incorporating some of the brilliant solutions proven out by your card partners and other 3rd parties in regards to noise and greater heat dissipation...especially at the high end?

A. Our reference solutions are designed to accommodate every system that fully complies with the ATX and PCIe mechanical and electrical specifications. Further, we have OEM and system integrator customers that prefer a blower-style assembly. In order to accommodate all parties interested in purchasing a reference GPU, our hand is guided in the direction of the designs you see.

Q. Why don’t graphics card makers like yourselves release a legacy AGP card on the newest technology? You could saturate the bandwidth available in AGP with a GCN card that's probably so efficient it'd be single slot!

A. I think it's mostly because AGP is a fully dead standard across the industry. The ROI would likely be negative.

Q. I like what AMD is doing with regard to mantle and the R9 290x and am entertaining the prospect of buying a couple, but I am reluctant to do so until the runt/dropped frame issue has been completely addressed for single GPU, crossfire setups and Eyefinity systems. Can you tell us what progress AMD has made with regard to this and what AMD is still working on to address any existing issues?

A. 1) Frame pacing was 100% resolved on single-GPU systems in early 2013. 2) Frame pacing for multi-GPU systems, at any resolution (e.g. Eyefinity/4K), is fully resolved in hardware on the 290 and 290X.  3) Frame pacing for the R9 270, R9 280 and HD 7000 Series systems will require a software solution. Our engineers are working on that right now, and we intend to release a driver this quarter to resolve the issue.

Q. When designing the 8000 series, what is the default resolution the cards are made for? (1080p, 1440p, 4k resolution)

A. The R9 290/290X: 4K.  The R9 280: 1440/1600p.  The R9 270X:/R7 260X: Max settings 1080p and high settings 1080p. From a hardware perspective, these are our design goals for the products.

Q. Can we have real numbers about the Mantle gain? Games are still running perfect at maximum with a 7970ghz, if the gain is not really high to hold the next graphic step it will be a bit useless for high end cards. Is it designed to improve latency? Add more graphic effects?  Does the next cpu will have better single core performance?

A. Real numbers regarding Mantle will be published at the AMD Developer Summit running November 11-13 in San Jose. I will say that we are not undertaking such a mammoth effort to yield 3-4% performance - that would be a waste of time.

Q. Regarding the AMD developer summit, will there be more developer / game support announced or will it be the same ones from the GPU 14 event?  Do you think there will be enough support for Mantle on the 290 / 290X or will it take longer than usual to implement and see the advantages on this current generation of cards?

A. We are planning to announce more titles than what you saw at GPU14. Depending on the schedule, those announcements may come slightly after APU13.

Q. With the introduction of Mantle and your Gaming Evolved Program, I would like to know which game companies (aside the ones you already told us) are going to collaborate with your teams to provide better optimized games to the Radeon line in the future?

A. We're partnered with: Eidos Montreal, DICE, Square Enix, Rebellion, Codemasters, EA, Crytek, Irrational Games, Nixxes, and more. We have an excellent relationship with almost every major studio out there, and we're in frequent contact to assist with performance optimizations for Radeon.

  • jumpmanxt
    Does AMD have an official download link for their cleanup utility? I can't get the link in the article to work (
  • tomfreak
    I is sad.... I ask if Radeon 7790 support TrueAudio or not, my question got passed -.-

    Radeon 7790 is the same chip with 260x.
  • ronch79
    WHy wasn't AMD asked about the future of their CPU division? Do they plan to offer FX SKUs using the Steamroller core? What about Excavator? Any plans to continue making x86 CPUs after Excavator?
  • InvalidError
    11927410 said:
    WHy wasn't AMD asked about the future of their CPU division?
    Because the AMA panel was from the GPU division and deferred CPU-related questions to AMD's upcoming forum and interviews elsewhere. In other words, they probably were not authorized to discuss CPU-related questions.
  • sarinaide
    Nice interview, its eye opening to see how little people in position to know actually knew about AMD and its projects. The automatic knee-jerk reaction is to assume a problem when there is no problem other than the assumption of problems.

  • goodguy713
    well this was pretty informative actually. the temp issue with the 290 and 290x seems like it can be solved just by gaming out side in 30 degree weather.. lol
  • goodguy713
    all joking aside i honestly think I will buy a 290 here soon been waiting for amazon to get more in stock
  • MajinCry
    Gah. Forgot about the AMA. I'd have asked if there would be any performance benefit from wrapping, say, DirectX 9 to Mantle, due to the low-level aspect of the latter.
  • InvalidError
    11928558 said:
    Gah. Forgot about the AMA. I'd have asked if there would be any performance benefit from wrapping, say, DirectX 9 to Mantle, due to the low-level aspect of the latter.
    Sounds like a silly idea to me since the point of Mantle is to bypass DXn's complexity and overhead in the first place by giving programmers lower-level access to the hardware.

    If you make DXn libraries that translate the DXn API to Mantle, you lose pretty much all advantages Mantle is intended to provide since you have to jump through all DXn hoops on top of dealing with Mantle stuff.

    With DXn, you get: Game -> DXn APIs -> driver hardware abstraction layer -> low-level drivers -> hardware
    With Mantle, you get: Game -> Mantle -> low-level drivers -> hardware
    With "Mantlified DXn" you get: Game -> DXn API -> DXn translation layer -> Mantle -> low-level drivers -> hardware

    Converting DXn to Mantle is likely less efficient since it introduces an extra middleware layer that may not be able to execute calls as efficiently as a middleware layer that speaks native hardware language - the hardware abstraction layer has liberties in how to translate DXn calls to hardware that Mantle would not have.
  • joditas
    This is the original version

    I appreciate THG for hosting this but many questions are simply left unanswered. If AMD wants to capture more market share, not just maintain, they need to do better in areas where they are lacking.