How so?
Nvidia Pascal Reviews [1080XP, 1080ti, 1080, 1070ti, 1070, 1060, 1050, and 1030]
- Thread starter Love_In_Rio
- Start date
D
Deleted member 2197
Guest
MSI GeForce GTX 1080 GAMING Z 8G review
http://www.guru3d.com/articles_pages/msi_geforce_gtx_1080_gaming_z_8g_review,1.html
http://www.guru3d.com/articles_pages/msi_geforce_gtx_1080_gaming_z_8g_review,1.html
Actually I'd noticed that as well..For Nvidia it seemed like the higher end parts got quite a bit better in performance per watt..but the lower end parts not so much. I did some research and here are TPU's numbers:-
Fermi - GTX 580 has 84% of the performance per watt of 560 Ti. - https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_560_Ti/28.html
Kepler - GTX 680 has 96% of the performance per watt of 660 - https://www.techpowerup.com/reviews/Zotac/GeForce_GTX_660/28.html
Maxwell - GTX 980 has 125% the performance per watt of 960 (1080p) - https://www.techpowerup.com/reviews/ASUS/GTX_960_STRIX_OC/30.html
Looks like Pascal will be similar to Maxwell.
It has probably to do with the relative size of the shader cores: the un-core units like PCIe, display, video encoder/decoder stay the same size irrespective of the size of the shaders. In the case of the GTX 1060, the memory/shader ratio is also different compared to the 1080.
I did think of that but the difference seemed too large to make sense. A good comparison would be GTX960 at 120W vs GTX 980 at 165W for example, since 960 is exactly half of 980. This implies the un-core consumes about 75W (Taking 165W - 120W = 45W as the shader/"core" power 960), which seems far too high.
Edit: Anyone know what time the GTX 1060 reviews go live?
Last edited:
DuckThor Evil
Legend
I think the picture changes quite a bit by choosing different products for these comparisons. 560 Ti vs GTX 550 would be closer die size comparison to the Kepler and Maxwell comparisons and sees the 560 Ti be clearly on top.
Also Techpowerup never tested the reference versions for GTX 660 and 960, instead they used reference clocked custom boards for their figures, which really isn't very accurate method of doing tests, custom boards often have lower performance per watt due to various reasons.
With Kepler a 780 Ti and 680 shows very similar performance per watt and with Maxwell 980 edges the bigger chips, bigger chips do better as resolution increases. In this chart 960 and Titan X are similar.
I think comparisons should be made with reference vs reference and due to chip variance, you'd need more samples and even then you are not really measuring just the chips, because different SKUs are run with different points in their performance curves for reasons such as market positioning. Look at Fury X vs Nano for reference, same chip, but a colossal difference in performance per watt.
Yes.
D
Deleted member 2197
Guest
Last edited by a moderator:
D
Deleted member 13524
Guest
Reviews have popped up all over the place.
To summarize:
1 - In DX11 the GTX 1060 trades blows but is generally faster than a RX480
2 - In DX12 the GTX 1060 trades blows but is generally slower than a RX480
3 - In Vulkan Doom the RX 480 is ridiculously faster, though it's only one game on Vulkan so far so..
4 - Performance/watt is substantially better than the RX480 (~35W difference)
5 - Performance/dollar for the "regular" version's MSRP is similar though a bit worse than the RX 480 8GB. (EDIT: Better for DX11, worse for DX12/Vulkan)
6 - Performance/dollar for the "Founders Edition" at $300 is crap.
To summarize:
1 - In DX11 the GTX 1060 trades blows but is generally faster than a RX480
2 - In DX12 the GTX 1060 trades blows but is generally slower than a RX480
3 - In Vulkan Doom the RX 480 is ridiculously faster, though it's only one game on Vulkan so far so..
4 - Performance/watt is substantially better than the RX480 (~35W difference)
5 - Performance/dollar for the "regular" version's MSRP is similar though a bit worse than the RX 480 8GB. (EDIT: Better for DX11, worse for DX12/Vulkan)
6 - Performance/dollar for the "Founders Edition" at $300 is crap.
Last edited by a moderator:
7 - performance/production cost is way better for the 1060.
Yeah I think neither lands a knock out blow on the other one.
Point 7. should have custom AIB vs custom AIB perf/dollar.
Point 8. Reference 1060 has better cooling design than reference 480.
Point 9. No SLI vs Crossfire option 480, but personally I would prefer comparable single GPU card even at a premium due to issues with mGPU support, although this does not bother everyone.
Cheers
Point 7. should have custom AIB vs custom AIB perf/dollar.
Point 8. Reference 1060 has better cooling design than reference 480.
Point 9. No SLI vs Crossfire option 480, but personally I would prefer comparable single GPU card even at a premium due to issues with mGPU support, although this does not bother everyone.
Cheers
Last edited:
D
Deleted member 2197
Guest
Performance/dollar actually is bit better than RX 480 if you can get card at $249, which seems to be the case at least for now at Evga & BestBuy.
https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1060/28.html
http://www.evga.com/Products/Produc...ily=GeForce+10+Series+Family&chipset=GTX+1060
Last edited by a moderator:
An interesting positive for Nvidia (from a perception perspective anyway) is how close the 1060 is to the 480 and AoTS with crazy settings at 1080p
DX12 RoTR with its great performance for Nvidia (ok by this I mean overall fps not 11 vs 12) balances out the worst performance scenes-episode for Nvidia in DX12 Hitman, which surprised me as the latest DX12/async compute patch for RoTR seemed to work really well for the Fury X.
But then both of these are cut down cards, which may be affecting the 480 more than the 1060.
I am just focusing at 1080p because that is what the majority will be buying either of these cards for IMO.
Cheers
DX12 RoTR with its great performance for Nvidia (ok by this I mean overall fps not 11 vs 12) balances out the worst performance scenes-episode for Nvidia in DX12 Hitman, which surprised me as the latest DX12/async compute patch for RoTR seemed to work really well for the Fury X.
But then both of these are cut down cards, which may be affecting the 480 more than the 1060.
I am just focusing at 1080p because that is what the majority will be buying either of these cards for IMO.
Cheers
DuckThor Evil
Legend
Seems like a nice card and there are custom cards in the stores today.
DavidGraham
Veteran
Actually no, R 480 maintains strong DX12 performance in AMD supported titles, not in general. For example, Tomb Raider has a large lead with the 1060, even with the new patch. In fact even Maxwell cards are faster in DX12 compared to their AMD counterparts.
D
Deleted member 13524
Guest
Pure baseless and useless speculation until you are able to show any kind of solid proof of that.
- How much is the price-per-mm^2 on GF's 14FF and TSMC's 16FF?
- What exactly are the yields for P10 and GP104?
- How much is the cooling solution for either card? How much is the PCB for either card?
- How much is each 8Gbps DRAM module being sold for each OEM?
- Where's the official BoM + assembly cost for each card? Reference or OEM?
This is that "HBM costs more than an entire building" argument again. Please, spare us of that speculative crap that tries to pass as fact.
You're right. Hardwarecanucks has a proper DX11 vs DX12 differentiation so I changed the post in accordance:
If true, this right here is a huge advantage over the RX480. Custom RX480 cards are taking forever to appear. I think AMD mispredicted the demand for custom versions quite a lot.
The GTX 1060 does not support SLI. The cards don't even have the SLI connectors.
It will support DX12 explicit multi-adapter (in unlinked mode only?), though.
That is what I am saying in general, but some tests I have seen with the latest DX12/async patch of RoTR actually has the Fury X now about 3-5% faster than the 980 ti where before it trailed ever so slightly.
However this is not reflected with the 1060 vs 480 where it seems the 480 is more constrained.
But also AoTS is in many ways 'AMD supported' (debatable at least) and here the 1060 is very close to the 480 with difference around 3% at 1080p with crazy setting.
Cheers
Similar threads
