Wednesday, May 3rd 2017

AMD "Vega 10" Bears Core-Config Similarities to "Fiji"
A Linux patch for AMD's GPU drivers reveals that its upcoming "Vega 10" graphics processor bears numeric core-configuration similarities to the "Fiji" silicon which drives the enthusiast-segment Radeon R9 Fury series graphics cards. The patch bears configuration values which tell the software how to utilize the resources on the GPU, by spelling them out. The entry "gfx.config.max_shader_engines = 4," for example, indicates that "Vega 10" features four shader engines, like "Fiji." Another entry "Adev-> gfx.config.max_cu_per_sh = 16" signifies the number of GCN compute units (CUs) per shader engine. Assuming the number of stream processors per CU hasn't changed from 64 in the "Vega" architecture, we're looking at a total stream processor count of 4,096. This could also put the TMU count at 256.
At earlier reveals of the "Vega 10" package, you notice a large, somewhat square GPU die neighboring two smaller rectangular memory stack dies, which together sit on a shiny structure, which is the silicon interposer. The presence of just two memory stack dies sparked speculation that "Vega 10" features a narrower 2048-bit memory interface compared to the 4096-bit of "Fiji," but since the memory itself is newer-generation HBM2, which ticks at higher clocks, AMD could run them at double the memory clock as "Fiji" to arrive at the same 512 GB/s bandwidth. The 4,096 stream processors of "Vega 10" are two generations ahead of the ones on "Fiji," which together with 14 nm process-level improvements, could run at much higher GPU clocks, making AMD get back into the high-end graphics segment.
Sources:
aceCrasher (Reddit), ComputerBase.de
At earlier reveals of the "Vega 10" package, you notice a large, somewhat square GPU die neighboring two smaller rectangular memory stack dies, which together sit on a shiny structure, which is the silicon interposer. The presence of just two memory stack dies sparked speculation that "Vega 10" features a narrower 2048-bit memory interface compared to the 4096-bit of "Fiji," but since the memory itself is newer-generation HBM2, which ticks at higher clocks, AMD could run them at double the memory clock as "Fiji" to arrive at the same 512 GB/s bandwidth. The 4,096 stream processors of "Vega 10" are two generations ahead of the ones on "Fiji," which together with 14 nm process-level improvements, could run at much higher GPU clocks, making AMD get back into the high-end graphics segment.
63 Comments on AMD "Vega 10" Bears Core-Config Similarities to "Fiji"
I have a lot of experience overclocking all types of GPU's: 560 Ti, 6950, 7970, 7950, 6850, 6870, 650 Ti, 390X, Fury, and on and on!!! Every single card gained just as much, and usually more performance if you overclocked its RAM. How much? My 7970 gained more from overclocking ram than from overclocking its core. That's how much.
Again, show me benchmarks that compares it and shows the benefit. I have yet to see one.
Found one from Anandtech, have not found any other yet. But this speaks volumes, a card with quite a small bus width gets minimal gains by overclocking the VRAM. (From 7Ghz to 7.8Ghz)
Memory bandwidth is the least of it's problems.
The funny thing about all this, is that *across the board* on a large benchmark suite it is STILL beneficial to have a 'too strong' core than it is to have too much bandwidth. Can you have too much bandwidth? Yes, because high bandwidth translates into an efficiency drop, and efficiency is TDP, and the more of that budget you can reserve for the GPU core, the more performance you can extract from it.
So you can harp on all you want about Nvidia's tight VRAM, as long as the GPU still is king of the road, which it is by a margin of about 50-60% at this point versus AMD's offerings, that is what counts. Not to mention the fact that only a portion of the game engines lean heavily on VRAM, while all game engines want a fast core. Nvidia's simply got a better balance going on than AMD, and this has been the case since Kepler.
Again, the resulting performance is what matters, all the rest is irrelevant, unless you are looking at very specific engines and situations, such as 4K gaming where the Fury X excelled, but still couldn't really beat a 980ti because the latter could get much better OC - a direct result of using a tight VRAM bus.
The VW Up! 1.0L was an I3... 3 cylinders.
Your logic is flawed. The point is definitely there....But, flawed nonetheless. :P
On Topic: This article is trash. Please take it down...
On topic: Although its somewhat interesting I couldn't care less about it. I just want to know the performance. If its got the performance then sign me up for Vega.
My R9 290 is super power hungry if I even try to overclock it a little.
They need to focus more on the core architecture. Memory bandwidth and speed is not currently an issue even at 4k resolutions for most games.
And at 4k with two cards id like more speed and memory bandwidth please.
Shareing similar numbers of inner hardware is both obvious since its their Ip they are building upon and have researched the optimisation of to be able to advance it.
You wouldn't scrap all that went into making any other device and start totally from scratch.
And if a cinfig works well for you why change.
So...guess what? Faster memory is needed or it won't scale.
Same for the communism remark.
If 2 products offer the same, you got to do something to make 1 more attractive so people will go for that one instead.
A lower price point would help a lot, now im not saying half would be the smart move as that means they are heavily competing with themselves, but if they are 200 dollars under Nvidia's equivalent, who would ever buy the Nvidia card?
You all seem to think its Ferrari that makes the most money selling 5 cars for 1 million and not Ford selling 5 million cars for 10.000.
Selling tons more at a lower price is much better then selling a few at a higher price.
Frankly, I don't believe you did any experimental testing, and I doubt you even understand what you are talking about.
Your chart suggests memory made a major difference! If you can't see your own chart supports my argument, talking to you is a waste of time.
PS: Out of curiosity, would you consider me an AMD 'fanboy'?
Either this is bs or AMD actually comes out with a new pipeline micro architecture (which is very unlikely).
Rumors point to Vega being AMD's first arch to use tiled rasterization. All evidence points to it too with a bigger die size per SP, more integrated ROP's, and some pretty crazy geometry and efficiency claims. Fingers crossed that they pull it off!
‘the bare bones of the plot’
‘one feels vulnerable in baring one's soul to another’