Author

Topic: Nvidia Titan V - does 70mh/s on ETH STOCK (Read 567 times)

sr. member
Activity: 2674
Merit: 328
December 12, 2017, 07:12:14 AM
#5
Volta's memory bus is 3072 vs 2048 bit width of Vega. It is 1.5x more bandwidth if using same freq. moreover I think NVIDIA is doing better with timings tuning

Im sure the timings differ and that could boost performance, but considering the Vega 64 can do 43mh/s maxed out, just tighter timings cant be pushing a 90% increase in peak performance, can they? (bear in mind these are stock timings, so they wouldn't be as aggressive as the STILT's straps).
Memory bus can be an additive factor, but then the GTX 1060 run a 192bit memory bus and outperform the 256bit RX 470/480/570/580 cards at same clocks (without bios mods ofcource).

Something tells me the Vega 56/64's are crapping out due to some other reason ...


I don't think 1060 wins vs AMD cards even at stock...some are even, but new ones with Hynix memory definitely
don't come close
legendary
Activity: 1510
Merit: 1003
December 12, 2017, 04:07:51 AM
#4
Volta's memory bus is 3072 vs 2048 bit width of Vega. It is 1.5x more bandwidth if using same freq. moreover I think NVIDIA is doing better with timings tuning

Im sure the timings differ and that could boost performance, but considering the Vega 64 can do 43mh/s maxed out, just tighter timings cant be pushing a 90% increase in peak performance, can they? (bear in mind these are stock timings, so they wouldn't be as aggressive as the STILT's straps).
Memory bus can be an additive factor, but then the GTX 1060 run a 192bit memory bus and outperform the 256bit RX 470/480/570/580 cards at same clocks (without bios mods ofcource).

Something tells me the Vega 56/64's are crapping out due to some other reason ...
Memory bus width is the key for eth if it is not compute bounded. Also timings are very important. Gtx1060 6g with shitty high latency Hynix gddr5 memory is doing only 17mhs stock for example.
hero member
Activity: 751
Merit: 517
Fail to plan, and you plan to fail.
December 12, 2017, 04:01:41 AM
#3
Volta's memory bus is 3072 vs 2048 bit width of Vega. It is 1.5x more bandwidth if using same freq. moreover I think NVIDIA is doing better with timings tuning

Im sure the timings differ and that could boost performance, but considering the Vega 64 can do 43mh/s maxed out, just tighter timings cant be pushing a 90% increase in peak performance, can they? (bear in mind these are stock timings, so they wouldn't be as aggressive as the STILT's straps).
Memory bus can be an additive factor, but then the GTX 1060 run a 192bit memory bus and outperform the 256bit RX 470/480/570/580 cards at same clocks (without bios mods ofcource).

Something tells me the Vega 56/64's are crapping out due to some other reason ...
legendary
Activity: 1510
Merit: 1003
December 12, 2017, 03:36:42 AM
#2
Volta's memory bus is 3072 vs 2048 bit width of Vega. It is 1.5x more bandwidth if using same freq. moreover I think NVIDIA is doing better with timings tuning
hero member
Activity: 751
Merit: 517
Fail to plan, and you plan to fail.
December 12, 2017, 02:52:52 AM
#1
Here is BBT's video doing a full roundup :

https://www.youtube.com/watch?v=CW8LOQ2yvO4

This card ofcource does not make economic sense, but it does show that HBM2 even at 850mhz stock clocks can push 70+ mh/s on Ethash, and upto 83mh/s when overclocked even in the preliminary tests done by BBT Carter in the above video.
What I'm curious about now is, what the hell is keeping the HBM2 on the Vega's back? Isn't all HBM2 made by Samsung? That would rule out different manufacturers and different timings as the cause.
Anyone have any insights into this?
Jump to: