AMD promises 100 gigabyte/second memory Lower power, higher capacity, less PCB real estate: that's the list of wins being touted by AMD for its coming High Bandwidth Memory (HBM) architecture. The previously-sketchy details of HBM have been filled out in this AMD white paper. The architecture is slated to roll in some of the Radeon 300 series cards when they land, …
Um mantle being open, no DX12 in the works when anoucing mantle, no new hardware needed for freesync when new scaler in monitor was needed, Bulldozer making them back competitive with intel, power of their APU's (they are even getting sued by investors over this one). I could go on but you get the point.
You have most of Mantle in Vulkan, isn't that open enough? "In the works" is a silly ambiguous term that might as well mean "it's still being only designed", don't know about "Freesync" or what they said about it (wasn't that about additional hardware?), Bulldozer was a bet on software going the effective parallel route instead of not-so-efficient (but easy for the developers) mostly single thread design, which clearly didn't work out, what's with the APUs (no apostrophe there)? The A10-7850 holds up pretty well despite one if its main features not being used yet again (HSA).
Yes, I do get the point. You confuse vague marketing-made promises with expectations.
AMD was a mess, but much less than people make it out to be. That led to CEO change and now that the initial confusion is almost over, they might come back somehow. Or not, we'll see. Regardless, as far as GPU hardware goes, I don't recall any underdelivering.
cyke1-
You obviously have not done your homework and are spewing headline type foolishness.
As an example Mantle is open and improves performance significantly because the poorly written code does not take full advantage of AMD high core count CPUs. DX12 actually incorporates the operating concepts of DX12 to improve AMD CPU performance and illustrate that the code was the primary restriction on high core count AMD CPU performance. Bulldozer was a poorly designed and executed product that showed up years late and it still delivered a good performance bump for the price. AMD's APUs are extremely power efficient and have superior graphics to any X86 APU in existence. AMD is NOT getting sued for the power of their APUs. In fact Intel has been exposed for lying about the products TDP where as AMD openly and accurately states their product TDPs.
AMD investors are suing for some products being late to market based on road map projections - which clearly state that they are subject to change. AMD's 10Q filings also state that future performance is based on existing knowledge. AMD investors are unlikely to prevail in court as a result, but any damn fool can sue in hopes of Jackpot Justice. Investing is not a guarantee of success and that is why all entities warn investors before they invest. Trying to use the judicial system to compensate you for your bad investing decisions if foolhardy.
Now if you actually want to learn something about HBM and why this works well and has been working well you might want to read the article at SemiAccurate.com on AMD's HBM. Then you won't need to post ignorant, technically clueless hate comments.
Bulldozer gave approximately zero average improvement over Phenom 2 in typical workloads. The SSE improvements are still going unused because SSE is so fragmented devs don't bother. The fx8370 I'm running is faster than the Phenom 2 it replaced... The clock hike and 2 more 'cores' explain almost all that increase, IPC is barely changed.
It's initially going into devices that stream to/from memory so latency isn't much of an issue. However 'latency = clocks*clock period' and lower clock rates don't guarantee higher latency, though rounding usually will, latency is a constant the clocks*period simply has to match or exceed.
In normal CPU's it's going to hit streaming modes less often and that might be a problem, increasing the average latency. We'll just have to wait and see.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
