NVIDIA GTX 1650 GDDR6 vs. GDDR5 Benchmark: Big Uplift in Performance
video description
Date: 2020-05-06
Related videos
Comments and reviews: 10
DrSuperGood
The lowered clock of the GDDR6 card is likely indirectly due to the increased performance of the higher bandwidth GDDR6 memory. When using GDDR5 memory it is possible that much of the GPU core was sitting idle waiting for data. Idle silicon uses less power than active silicon which is the basis of how power virus applications like Furmark work. The GDDR6 memory allows less of the GPU core to be idle waiting for data at any given time. This indirectly increases power usage making it more likely that the GPU will run into a clock limiting factor. This would also explain why the GPU clock frequency with GDDR6 is a lot less stable as it is bouncing off a limit. The limit may not necessarily be power or current but could also be thermal in the sense that GPU boost frequency depends on GPU temperature.
reply
The lowered clock of the GDDR6 card is likely indirectly due to the increased performance of the higher bandwidth GDDR6 memory. When using GDDR5 memory it is possible that much of the GPU core was sitting idle waiting for data. Idle silicon uses less power than active silicon which is the basis of how power virus applications like Furmark work. The GDDR6 memory allows less of the GPU core to be idle waiting for data at any given time. This indirectly increases power usage making it more likely that the GPU will run into a clock limiting factor. This would also explain why the GPU clock frequency with GDDR6 is a lot less stable as it is bouncing off a limit. The limit may not necessarily be power or current but could also be thermal in the sense that GPU boost frequency depends on GPU temperature.
reply
Patrick
It's so fun watching this. I'm still using a i7-6700K machine, but I've bough an msi Meg X570 Unify motherboard, and when I can afford it (maybe with my stimulus check, I'm getting a Ryzen 7 3700x to put in it. My 64 GB of G. Skill TridentZ DDR4 3200 will work fine with it, and I'll keep using my ZOTAC GeForce GTX 980 Ti 6GB AMP! Extreme. My power supply will be happy to provide less power I suppose, it's a SeaSonic Snow Silent-1050 80 PLUS PLATINUM Modular Power Supply. My 3 m. 2s will plug right in and my two Western Digital RE WD6001FXYZ HDD 6TB 7200 RPM 128MB Cache SATA 6. 0Gb/s are still great to go. Do you see any problems just trading everything from my old board to the new board?
reply
It's so fun watching this. I'm still using a i7-6700K machine, but I've bough an msi Meg X570 Unify motherboard, and when I can afford it (maybe with my stimulus check, I'm getting a Ryzen 7 3700x to put in it. My 64 GB of G. Skill TridentZ DDR4 3200 will work fine with it, and I'll keep using my ZOTAC GeForce GTX 980 Ti 6GB AMP! Extreme. My power supply will be happy to provide less power I suppose, it's a SeaSonic Snow Silent-1050 80 PLUS PLATINUM Modular Power Supply. My 3 m. 2s will plug right in and my two Western Digital RE WD6001FXYZ HDD 6TB 7200 RPM 128MB Cache SATA 6. 0Gb/s are still great to go. Do you see any problems just trading everything from my old board to the new board?
reply
SWIRF
the lowered clocks is another example of the 5600 xt bios situation but not on the same scale. if the gpu's can handle better clocks running 24/7 for years. having them lower is just greedy, or atleast shady. If encroachment on other products you make is such an issue stop making 30 GPU's that are 5% - 10% different in performance and cramming them into a price point. What like 90%+ of consumers leave everything stock? Give people the maximum performance of the product out of the gate. Well, then you couldn't dupe people who don't do the research into spending more money on your X K oc and gaming models.
reply
the lowered clocks is another example of the 5600 xt bios situation but not on the same scale. if the gpu's can handle better clocks running 24/7 for years. having them lower is just greedy, or atleast shady. If encroachment on other products you make is such an issue stop making 30 GPU's that are 5% - 10% different in performance and cramming them into a price point. What like 90%+ of consumers leave everything stock? Give people the maximum performance of the product out of the gate. Well, then you couldn't dupe people who don't do the research into spending more money on your X K oc and gaming models.
reply
ChintzyPC
Note on the thermal solution for the memory: With GDDR6 on my 1660Ti SC Ultra I'm able to overclock the memory to +1290 fully stable (+1450 is when it gets unstable. However that is after placing thermal pads with 17W/mK on the opposite side of the PCB between it and the backplate. Also note this is without any added heatsinks (as it's the same not direct-die cooled heatsink layout as the card in this vid. Seems like there needs to be some more testing in regards to GDDR6 flip-chip thermals and whether direct-die is better than cooling from the back of the PCB.
reply
Note on the thermal solution for the memory: With GDDR6 on my 1660Ti SC Ultra I'm able to overclock the memory to +1290 fully stable (+1450 is when it gets unstable. However that is after placing thermal pads with 17W/mK on the opposite side of the PCB between it and the backplate. Also note this is without any added heatsinks (as it's the same not direct-die cooled heatsink layout as the card in this vid. Seems like there needs to be some more testing in regards to GDDR6 flip-chip thermals and whether direct-die is better than cooling from the back of the PCB.
reply
Doug
I bought a 1650 for one of my computers which is in a 3U rack case. It's essentially the fastest card I can fit in the case, as any card which requires separate power won't fit. The top of all of these cards is physically right up against the top of the case, so there's absolutely zero room for anything to stick out above the cards. If someone made a card where the power comes out the back instead of the top, I'd happily upgrade the machine to a much faster card. But I haven't found anything that will work without doing any soldering to add my own power connector.
reply
I bought a 1650 for one of my computers which is in a 3U rack case. It's essentially the fastest card I can fit in the case, as any card which requires separate power won't fit. The top of all of these cards is physically right up against the top of the case, so there's absolutely zero room for anything to stick out above the cards. If someone made a card where the power comes out the back instead of the top, I'd happily upgrade the machine to a much faster card. But I haven't found anything that will work without doing any soldering to add my own power connector.
reply
Igor
Im not surprised at all. when you compare gtx 1070 to rtx 2060, both have 1920 cuda cores but the latter is 20% faster. The rtx 2060 managed to match or even surpass the gtx 1080 in some titles now, even though the architecture is slightly better on low level apis than pascal thats not enough to justify such crazy gains in performance, lets not forget rtx 2060 has also less memory bus width (192 vs 256 bits) which in theory should help balancing things between them but in reality rtx 2060 is still considerable faster.
reply
Im not surprised at all. when you compare gtx 1070 to rtx 2060, both have 1920 cuda cores but the latter is 20% faster. The rtx 2060 managed to match or even surpass the gtx 1080 in some titles now, even though the architecture is slightly better on low level apis than pascal thats not enough to justify such crazy gains in performance, lets not forget rtx 2060 has also less memory bus width (192 vs 256 bits) which in theory should help balancing things between them but in reality rtx 2060 is still considerable faster.
reply
Agost
The most interesting data here is about GDDR6 power consumption, about 2-2. 5W more than GDDR5 per chip and it's only 12Gbps. Obviously it gets worse since on this model they're not cooled properly, increasing power consumption by a bit, but I'm kinda confident saying GDDR6 12Gbps pulls 2W more than GDDR5 8Gbps per chip. +8W on a 4GB card and +16W (or more) on a 8GB card is actually a lot; I wonder how much power the 11GB of 14Gbps GDDR6 are pulling on the 2080Ti, like 50-60W?
reply
The most interesting data here is about GDDR6 power consumption, about 2-2. 5W more than GDDR5 per chip and it's only 12Gbps. Obviously it gets worse since on this model they're not cooled properly, increasing power consumption by a bit, but I'm kinda confident saying GDDR6 12Gbps pulls 2W more than GDDR5 8Gbps per chip. +8W on a 4GB card and +16W (or more) on a 8GB card is actually a lot; I wonder how much power the 11GB of 14Gbps GDDR6 are pulling on the 2080Ti, like 50-60W?
reply
SlavjanA
Considering the 30% extra GPU power draw headroom provided by the lower power draw of G6, you'd think it'd clock higher. But based on these results, and its stock specs, it seems as though either Nivida's using lower binned GPUs for this variant, or perhaps more likely on account of the small die size and maturity of the node, they're potentially pumping more voltage through it to essentially choke it.
reply
Considering the 30% extra GPU power draw headroom provided by the lower power draw of G6, you'd think it'd clock higher. But based on these results, and its stock specs, it seems as though either Nivida's using lower binned GPUs for this variant, or perhaps more likely on account of the small die size and maturity of the node, they're potentially pumping more voltage through it to essentially choke it.
reply
Daniel
even knowing OC targets are highly variable between cards, and how that might skew numbers in a not scientific way, I would have like to have seen these numbers also shown compared to a MAX OC of the new card, to show how much more performance could actually be available. edit: P. S. my point here is to determine if the core clock downgrade is potentially not affecting OC potential for the card.
reply
even knowing OC targets are highly variable between cards, and how that might skew numbers in a not scientific way, I would have like to have seen these numbers also shown compared to a MAX OC of the new card, to show how much more performance could actually be available. edit: P. S. my point here is to determine if the core clock downgrade is potentially not affecting OC potential for the card.
reply
Setzer
Hey guys, it'd be really interesting to compare the laptop gtx 1650ti (just a gddr6 version of the laptop gtx 1650 with 1024 cores) to the desktop gtx 1650 gddr5 and gddr6 variants. Perhaps clock them all similarly for better testing. It'd be really interesting what the full gtx 1650 die with gddr6 can do. Maybe it'll be close to a rx470 in performance. Or even close to a gtx 1060 3gb.
reply
Hey guys, it'd be really interesting to compare the laptop gtx 1650ti (just a gddr6 version of the laptop gtx 1650 with 1024 cores) to the desktop gtx 1650 gddr5 and gddr6 variants. Perhaps clock them all similarly for better testing. It'd be really interesting what the full gtx 1650 die with gddr6 can do. Maybe it'll be close to a rx470 in performance. Or even close to a gtx 1060 3gb.
reply
Add a review, comment
Other channel videos
