Details on GeForce GTX 1660 Revealed Courtesy of MSI - 1408 CUDA Cores, GDDR 5 Memory

[ArbitraryAffection]

balance the load distribution ? who needs that if you have enhanced async in RVII already

Lol, i have heard that Fiji and Vega are having some major issues getting work sent to the compute units effectively enough. Whether this is due to stalling in other parts of the pipeline (Geometry) or scheduling/driver issues IDK. But Nvidia also has a similar issue with their larger dies, but they are much, much better at it. GV100 has a similar issue in 3D workloads to Vega, look at 5120 CC TV vs 4352 CC 2080 Ti, they did some massive improvements in that regard to Turing. GV100 can also concurrently excecute integer and floating point ops. (Also check: Something about GV100 dropping Instruction Level Parallelism, and the dual-issue dispatch engines, instead focusing on Thread-level parallelism. (better for Compute I think).

Given a simple compute task; GCN is pretty damn good. It scales well to the entire Compute engine, but 3D graphics are a bit different I think. "Workload distribution" was a major problem area that Vega was supposed to address, also primitive rate / geometry (Remember Primitive Shaders?) That never came to fruition due to issues with the design, the silicon and/or the driver. That's what I heard. Vega is unfinished and missed the performance target AMD was aiming for (1080 Ti). But i digress. Ironic it's called "Graphics" core next, it should be called CCN: "Compute Core Next" XD

 

[cucker tarlson]

Lol, i have heard that Fiji and Vega are having some major issues getting work sent to the compute units effectively enough. Whether this is due to stalling in other parts of the pipeline (Geometry) or scheduling/driver issues IDK. But Nvidia also has a similar issue with their larger dies, but they are much, much better at it. GV100 has a similar issue in 3D workloads to Vega, look at 5120 CC TV vs 4352 CC 2080 Ti, they did some massive improvements in that regard to Turing. GV100 can also concurrently excecute integer and floating point ops. (Also check: Something about GV100 dropping Instruction Level Parallelism, and the dual-issue dispatch engines, instead focusing on Thread-level parallelism. (better for Compute I think).

Given a simple compute task; GCN is pretty damn good. It scales well to the entire Compute engine, but 3D graphics are a bit different I think. "Workload distribution" was a major problem area that Vega was supposed to address, also primitive rate / geometry (Remember Primitive Shaders?) That never came to fruition due to issues with the design, the silicon and/or the driver. That's what I heard. Vega is unfinished and missed the performance target AMD was aiming for (1080 Ti). But i digress. Ironic it's called "Graphics" core next, it should be called CCN: "Compute Core Next" XD

you can only go with no major arch overhaul for so long.the demand for faster gpus is there every year and more cores at lower die size approach can't last you a decade if no major changes in uarch follow.

 

[ArbitraryAffection]

you can only go with no major arch overhaul for so long.the demand for faster gpus is there every year and more cores at lower die size approach can't last you a decade if no major changes in uarch follow.

Yup, but unfortunately AMD kinda mis-judged the direction of game engine development I think, with GCN they focused heavily on Compute and number crunching for shaders but didn't beef up the geometry and with the newer larger chips, their pixel backend. Also AMD can't rreally afford to spend that much on arch development, nowhere near as much as Nvidia and given they also make (really competitive, might i add) CPUs, its even more compounded. But GCN is on its last legs and lets hope the next arch is super awesome

 

[Ibotibo01]

Navi killer is already here. XD

1660 Ti with 1536 CUDA and GDDR6 only just matches 1070. this iwll be 15% slower I think. But honestly for $219 it will be a fantastic 1080p gaming card and great value. I may get one if they land at or below £200 here.


Ehm, Navi isn't even here. For all we know, Navi could be 2070 performance and $249 which would make all cards under 2080 useless for most gaming.

I don't think so. Nvidia knows Navi's power and They develop GTX 16 series. I think that Navi will match to GTX 16 Series. Anyways we will see.

 

[ArbitraryAffection]

I don't think so. Nvidia knows Navi's power and They develop GTX 16 series. I think that Navi will match to GTX 16 Series. Anyways we will see.

Yep and i agre. I was just making a deliberate exaggeration to stress my point - Navi is still a massive unknown at this point.

 

[efikkan]

I don't think so. Nvidia knows Navi's power and They develop GTX 16 series. I think that Navi will match to GTX 16 Series. Anyways we will see.

I didn't know Nvidia used psychics to foresee the future.

GPUs are designed on assumption on where the market will be in 3-5 years time. Nvidia don't know how good Navi will be, as AMD is struggling even producing fully working engineering samples…

 

[Ibotibo01]

I didn't know Nvidia used psychics to foresee the future.

GPUs are designed on assumption on where the market will be in 3-5 years time. Nvidia don't know how good Navi will be, as AMD is struggling even producing fully working engineering samples…

Huang said VEGA 7 isn't wisely. This card didn't reveal at the moment. Companies know other companies technology.

 

[ArbitraryAffection]

I didn't know Nvidia used psychics to foresee the future.

GPUs are designed on assumption on where the market will be in 3-5 years time. Nvidia don't know how good Navi will be, as AMD is struggling even producing fully working engineering samples…

Yeah but looking at the last 5 years of AMD GPU releases, aside from gems like 570, Nvidia probably isn't worried that much....

 

[efikkan]

Huang said VEGA 7 isn't wisely. This card didn't reveal at the moment. Companies know other companies technology.

Competent people can do qualified guesses, but Nvidia doesn't know how Navi will perform. You can't know a fact that doesn't exist yet, well unless you "know" the future…

 

[cucker tarlson]

Huang said VEGA 7 isn't wisely. This card didn't reveal at the moment. Companies know other companies technology.

he certainly knew more of vega than they know of navi.

Yeah but looking at the last 5 years of AMD GPU releases, aside from gems like 570, Nvidia probably isn't worried that much....

especially in the high-end segment,where having a vastly more power efficient architecture gives you the edge.

 

[Ruru]

If the prices won't have that "Finland extra", this could be an awesome card by price/performance ratio.

 

[ArbitraryAffection]

If the prices won't have that "Finland extra", this could be an awesome card by price/performance ratio.

Damn you guys have 24% VAT that is high, i thought UK 20% VAT was bad enough. yeah this card is gonna be great value. I was thinking maybe if it land around £200 i will get one

you know whats funny though right? Cake here is 0% VAT. because it is considered an essential food item. Thats right, cake is an essential food item. )

Was a thing about whether Jaffa Cakes were cakes, or luxury chocolate biscuits here. If they were considered luxury chocolate biscuits they would need to take higher VAT. Of course they are cakes IMO and the company defended their cake status. Cake FTW

 

[Ruru]

1660 Ti with 1536 CUDA and GDDR6 only just matches 1070. this iwll be 15% slower I think. But honestly for $219 it will be a fantastic 1080p gaming card and great value. I may get one if they land at or below £200 here.

In some scenarios 1660 Ti gets between 1070 and 1070 Ti. 1408 shaders and GDDR5 isn't THAT huge difference, this could trade blows with 1070 in some scenarios.

e: Yeah, 24% VAT. It wasn't even that long when it was 22%

 

[MEC-777]

Age becomes relevant at some point, especially since AMD tend to drop driver support after a few years, in a couple of cases even products that are still for sale.

Yes, but not after 2-3 years or so for the past few generations.

The tale of the continuously improving driver updates from AMD is a myth. Nvidia also work on their drivers, and have made numerous improvements over the last decade. The argument that a lesser card from AMD will overcome a "better" card from the competitor over time has basically never panned out.

Not a myth. I used to run AMD GPUs and actually saw the improvements. Overcoming a better card from Nvidia? No. But better over time? Yes.

 

[efikkan]

Not a myth. I used to run AMD GPUs and actually saw the improvements. Overcoming a better card from Nvidia? No. But better over time? Yes.

The point is, all drivers generally improve. This is not unique to AMD.

 

[ghazi]

Previous rumors said 1280 SP, but 1408 SP is quite believable to me. Why? 1280 SP with GDDR5 might not pull ahead of RX 590.

 

[Ruru]

Previous rumors said 1280 SP, but 1408 SP is quite believable to me. Why? 1280 SP with GDDR5 might not pull ahead of RX 590.

I doubt that, "RX 480 v3" isn't that special. But of course 1408 shaders is great.

 

[ArbitraryAffection]

Previous rumors said 1280 SP, but 1408 SP is quite believable to me. Why? 1280 SP with GDDR5 might not pull ahead of RX 590.

Pascal CUDA cores, yeah, as in 1060, it is slower. But with Turing improvements i think 1280 CUDA would have still beaten 590 in all honesty.

I mean 2060 has 1920 CUDA, just like 1070. And in fact when you consider it has fewer ROPs it is still not insignificantly faster than 1070. That said it does have 31% more memory bandwidth, too. Huh I would like to see a 2060 with memory downclocked to 10.6Gbps~ so it matches 1070 bandwidth of ~256GB/s and then test performance. Also clock for clock, would be interesting

But yeah i see what you are saying. NV want to make sure it is a killing blow and the extra CUDA's should do that.

 

[ShurikN]

Lol, what? I favour Radeon but this comment makes no sense. GTX 1660 at $219 makes a lot of sense. It will be 15-20% faster than RX 590 but 12% cheaper. It is also going to be over twice as efficient. At its current price, the only card that is useless and redundant, is the RX 590.

Considering it will have 10% more Cuda cores and everything the same compared to 1060 (bus, mem type, mem amount), the performance will definitely not be 15-20% higher than 590. That's 1660Ti territory.

 

[Nihilus]

lol,he's apparently hurt that rx590 is a sku that's been the laughing stock for people all around tech sites and channels lately.can't blame them tho.it's like amd said "let's sell overclocked polaris under a higher sku name,but with a bigger price premium than last time"
a slightly cut die with same amount of memory but it's ddr5 this time,it'll end up at 10% slower than 1660Ti. Sorta like 1070Ti was to 1080. meanwhile,a $60 price difference is pretty noticeable for people buying in this segment.

The performance difference will most likely be bigger than 10%. The bandwidth drop from 256 bit GDDR5X to GDDR5 was usually unnoticed at that performance level.

However, going from 192 bit GDDR6 to 192 bit GDDR5 will have a significant impact beyond the shader count difference.

 

[cucker tarlson]

Th

The performance difference will most likely be bigger than 10%. The bandwidth drop from 256 bit GDDR5X to GDDR5 was usually unnoticed at that performance level.

However, going from 192 bit GDDR6 to 192 bit GDDR5 will have a significant impact beyond the shader count difference.

Yeah but not by much,1080 was like 7% faster than 70ti, this might end up 10% slower.

Edit:it was actually 4% at 1080p.

 

[95Viper]

Stay on topic.
Stop posting nonsense for the pure joy of it

Thank You.

 

[ArbitraryAffection]

Stay on topic.
Stop posting nonsense for the pure joy of it

Thank You.

Oof, sorry. I am master digressor. t_t . didn't mean to post nonsense. i will stfu :x

 

[Prima.Vera]

The leather jacket dude lost his mind....
So many cards, series...; confusion at it's finest. Good job!

 

[kings]

I don´t think he has lost his mind, I think he knows perfectly what he is doing, flooding the market with his cards.

If people only talk about Nvidia, if only Nvidia launches products, if people only listen and read Nvidia everywhere, they end up buying Nvidia. The 81% market share is proof of that!

AMD turns out to be lost and forgotten in the middle of so many Nvidia cards.