[News/Rumor] AMD Vega OpenCL benchmark ?

[StyM]

https://videocardz.com/67275/amd-veg...048-bit-memory

SiSoft benchmark detected 64 Compute Units on 687F:C3 device (so not C1 like in the previous leak). This device has 8GB 2048-bit memory configuration, which means two HBM2 stacks, each 4GB and 1024-bit. When it comes to core clock (344 MHz) and L2 cache (16 kB), those are definitely wrong readings, so I wouldn?t pay attention to them.

Therefore, this leak would ?confirm? that Vega has 4096 Stream Processors (64 * 64 CUs) and two stacks of HBM with 8GB in total. What we don?t know is whether 687F:C3 is the top model in RX Vega lineup, as there are clearly few different variants. So far there is no trace of cut-down chip or Vega 11.

Either way, it seems AMD engineers got really busy recently?

https://videocardz.com/67242/amd-veg...-units-spotted

[Tim]

Preliminary performance between 1080 and 1080 Ti, which I think we've all be expecting. Which could end up being between a rock and a hard place depending on nvidia's pricing. But then again, if the pricing is right, some people who are wanting to purchase a AMD card to go with Ryzen should be fairly happy. If it outperforms an 1080 for a good price, that can only be a good thing for the entire market. We've already seen an effect of the upcoming Vega on nvidia's pricing.

They really do need to get this out to market ASAP though. It's been way too long (And I REALLY mean way too long).

[vario]

Preliminary performance between 1080 and 1080 Ti, which I think we've all be expecting. Which could end up being between a rock and a hard place depending on nvidia's pricing. But then again, if the pricing is right, some people who are wanting to purchase a AMD card to go with Ryzen should be fairly happy. If it outperforms an 1080 for a good price, that can only be a good thing for the entire market. We've already seen an effect of the upcoming Vega on nvidia's pricing.

They really do need to get this out to market ASAP though. It's been way too long (And I REALLY mean way too long).

I wouldnt exactly extrapolate opencl performance with gaming performance. AMD`s were always good crunchers, but in gaming it can be different kinda thing, would be good to see how in this benchmark does fury X and rx 480.
My money is on vega not being able to continually beat 1080, just in good scenarios like vulkan/dx12.

[AbortRetryFail?]

I wouldnt exactly extrapolate opencl performance with gaming performance. AMD`s were always good crunchers, but in gaming it can be different kinda thing, would be good to see how in this benchmark does fury X and rx 480.
My money is on vega not being able to continually beat 1080, just in good scenarios like vulkan/dx12.

I suspect their focus is less on DT/Gaming and more on Enterprise. Sure -- they want a good 'game show' but AMD looks to be aiming directly at nVidia's pocketbook with their MI25 "Falcon Radeon Instinct Cluster" (built upon a Naples Platform).

[zanzabar]

They really do need to get this out to market ASAP though. It's been way too long (And I REALLY mean way too long).

i dont think amd has anything to do with when they can get it to market. they were incredibly short sighted and went HMB only on their high end and could not quickly make something like the gp102.

[dasa]

I wouldnt exactly extrapolate opencl performance with gaming performance. AMD`s were always good crunchers, but in gaming it can be different kinda thing, would be good to see how in this benchmark does fury X and rx 480.
My money is on vega not being able to continually beat 1080, just in good scenarios like vulkan/dx12.

i think its in with a chance with 4096 Stream Processors vs 2304 on the rx 580 which is supposed to catch up to the gtx 1070 performance through ncu vs gcn among other ipc improvements
time will tell

[vario]

i think its in with a chance with 4096 Stream Processors vs 2304 on the rx 580 which is supposed to catch up to the gtx 1070 performance through ncu vs gcn among other ipc improvements
time will tell

Well, optimism in regard of AMD rarely pays off ;-) . GPU division wasnt as much behind as CPU one, they just badly bet on HBM and glofo, I wouldnt expect huge IPC gains.
Similar talks were before polaris, that its gonna be MUCH more fficient, higher clocks etc. But reality was, it was sometimes just as slow as predicted by pure shader count and memory throughput. Yes, sometimes gains due to newer architecture are pretty good, but in reality only clocks made it possible to catch up to R9 ~390 on average.
RX 580, is ALMOST CERTAINLY , will be polaris v2, so just higher clocks, MAYBE gddr5x .
Look how far from the 1080 RX 480 is.
https://tpucdn.com/reviews/NVIDIA/Ge..._2560_1440.png

Of course the performance delta somewhat varies on the games used, but they would have to perform a miracle in order to get polaris up to 1080.
Rememver, AMD stated that cards based on vega will be called vega, so RX580 is surely not anything vega related.
Looking at fury X however, its quite possible Vega which is somewhat similar but with newer arch and higher clocks will reach ~1080.
1080ti, no way