Monday, September 3rd 2018

NVIDIA GeForce RTX 2080 Ti Benchmarks Allegedly Leaked- Twice

Caveat emptor, take this with a grain of salt, and the usual warnings when dealing with rumors about hardware performance come to mind here foremost. That said, a Turkish YouTuber, PC Hocasi TV, put up and then quickly took back down a video going through his benchmark results for the new NVIDIA GPU flagship, the GeForce RTX 2080 Ti across a plethora of game titles. The results, which you can see by clicking to read the whole story, are not out of line but some of the game titles involve a beta stage (Battlefield 5) and an online shooter (PUBG) so there is a second grain of salt needed to season this gravy.

As it stands, 3DCenter.org put together a nice summary of the relative performance of the RTX 2080 Ti compared to the GeForce GTX 1080 Ti from last generation. Based on these results, the RTX 20T0 Ti is approximately 37.5% better than the GTX 1080 Ti as far as average FPS goes and ~30% better on minimum FPS. These are in line with expectations from hardware analysts and the timing of these results tying in to when the GPU launches does lead some credence to the numbers. Adding to this leak is yet another, this time based off a 3DMark Time Spy benchmark, which we will see past the break.
The second leak in question is from an anonymous source to VideoCardz.com that sent a photograph of a monitor displaying a 3DMark Time Spy result for a generic NVIDIA graphics device with code name 1E07 and 11 GB of VRAM on board. With a graphics score of 12, 825, this is approximately 35% higher than the average score of ~9500 for the GeForce GTX 1080 Ti Founders Edition. This increase in performance matches up closely to the average increase in game benchmarks seen before and, if these stand with release drivers as well, then the RTX 2080 Ti brings with it a decent but not overwhelming performance increase compared to the previous generation in titles that do not make use of in-game real-time ray tracing. As always, look out for a detailed review on TechPowerUp before making up your minds on whether this is the GPU for you.
For those interested, screenshots of the first set of benchmarks are attached below (taken from Joker Productions on YouTube):
Add your own comment

86 Comments on NVIDIA GeForce RTX 2080 Ti Benchmarks Allegedly Leaked- Twice

#76
Captain_Tom
cucker tarlson, post: 3897767, member: 173472"
Point is, your point is miselading. No 7nm Vega for gamers this year or in early 19. How much will 7nm and 4 stack hbm2 bring to vega anyway, it's 1080 performance. If 2080ti is 1.35-1.4x of 1080ti, that's 1.35-1.4 times 1.3x, so 1.8x of vega 64. 80% faster - don't think so.... 7nm Vega can probably beat 16nm 1080Ti,though not by much. 2080 performance in other words and 10-15% faster than 2070. worth waiting for months ? If you got money for a gpu now then I don't think so,might as well pick up 1080ti when they drop price, won't be much difference from 7nm Vega.
Techpowerup just had an article, nothing misleading about it lol:

https://www.techpowerup.com/247305/amd-fast-tracks-7nm-navi-gpu-to-late-2018-alongside-zen-2-cpu


I really don't think you (or most people I see commenting on 7nm Vega/Navi) fully comprehend what it means to have a node advantage. 7nmFF allows for 3x the density, 40% higher clocks, and 65% less power usage compared to 16nm (And AMD's 14nm is worse than TSMC's 16nm!).

A 250mm^2 Navi chip could have as many as 6320-SP's while only using 150w. A 336mm^2 Vega would likely have up to 8000 SP's and over a TB/s of bandwidth. Again - Nobody should write off 7nm Radeon. It would be like saying "There is no way the Vega 64 can beat the 980 Ti lol).
Posted on Reply
#77
cucker tarlson
lol as if they were gonna just pack 8000 sp in it and blow everything away, their 4096 cards are already demonstrating diminishing performance returns and efficiency over 3854 sp ones. It'll only scale this way in compute oriented work,which is exactly what Vega is,a workstation card. They'll pack more sp but lower the clocks, they will get dimisnishing returns in gaming. In order for Vega to compete with top of the line geforce cards,it'd have to be designed in that way. Look at Titan V, 5120 sp and huge bandwidth, 2080 OC with 2944 and half its bandwidth will probably match it.
Read my comment,amd explicitly said 7nm is for HPC
Posted on Reply
#78
Captain_Tom
cucker tarlson, post: 3897777, member: 173472"
lol as if they were gonna just pack 8000 sp in it and blow everything away, their 4096 cards are already demonstrating diminishing performance returns and efficiency over 3854 sp ones.
read my comment,amd explicitly said 7nm is for HPC
The most credible rumors point to 88 or 96 compute units - that's up to 6144 Cores. However they will likely be spread out the SP's for higher clocks, and have 50-100% more ROP's so there is less bottle necking this time around. Then keep in mind they will have over double the bandwidth.

Vega 10 only had diminishing returns with SP's because HBM2 yields fell below the expected performance, and thus AMD was forced to launch cards with 10-30% less bandwidth than intended.
Posted on Reply
#80
Captain_Tom
cucker tarlson, post: 3897793, member: 173472"
We'll see what happens,vega 20 will not be gaming oriented,neither in design nor in its availability.

https://www.tweaktown.com/news/62491/amd-vega-20-7nm-32gb-hbm2-up-tflops-400w-tdp/index.html



I have much more faith in navi, the question is, can amd create an architecture that will scale with die size and sp count from low-end to high-end as well or better than nvidia's.
GCN is definitely due to be replaced with a brand new architecture - no doubt it is showing its age! However the "scaling issue" is no mystery at all.

AMD hasn't increased ROP count since the bloody 290X, and it painfully shows. Then again Vega 10 is a compute-first architecture, and AMD did increase ROP's per CU by a factor of 2 with their Vega M lineup. It shows too, with a 75w Vega card competing with the 120w 1060. My "shooting from the hip" guess is that Navi 10 will consist of:

-200-300mm^2 7nmFF Die
-4096-5120 SP's @ 1.8-2.5GHz
-96-128 ROP's
-240-320 TMU's
-No HBCC, hopefully Tiled Rasterization has Improved
-256-384-bit GDDR6 (512GB/s+)
-100-200w power usage


There are many ways AMD can skin this cat, but my expectation is that "Navi" is effectively Vega M scaled up with some extra tweaks. It really needs to be emphasized that there are competing rumors too, with some pointing to a very cheap 90-100w card that's meant to replace Vega 64, and another rumor stating that AMD wishes they made Polaris 10 a little bigger (and thus might consider a 300mm^2 card). I am hoping it's the latter considering how well the XBX turned out, and Nvidia is going for a cash grab this round.
Posted on Reply
#81
Berfs1
Captain_Tom, post: 3897755, member: 155434"
AMD announced they are launching 7nm Vega this year for gamers. Don't count out the possibility that it could blow away Turing...


That's definitely true. I noticed a major disadvantage with iron sights when I gamed on my laptop recently (900p).
It’s not going to compare to Turing, because its practically a die shrink... yes I have insider info on that...
Posted on Reply
#82
notb
Berfs1, post: 3898611, member: 163173"
It’s not going to compare to Turing, because its practically a die shrink... yes I have insider info on that...
Of course you do. :-)

Did I miss something? We're having a significant growth of active people here in last few days (both old and new accounts). What's going on?
Posted on Reply
#83
Captain_Tom
Berfs1, post: 3898611, member: 163173"
It’s not going to compare to Turing, because its practically a die shrink... yes I have insider info on that...
You don't need insider information to know it's a die shrink of Vega 10 (Or from what I understand Vega 10 was almost a Beta version on 14nm).

But are you saying it's still on a 4096-SP's? Even with the still fairly large 3336mm^2 die? I heard that it's 88-96CU's....
Posted on Reply
#84
cucker tarlson
Captain_Tom, post: 3898823, member: 155434"
You don't need insider information to know it's a die shrink of Vega 10 (Or from what I understand Vega 10 was almost a Beta version on 14nm).

But are you saying it's still on a 4096-SP's? Even with the still fairly large 3336mm^2 die? I heard that it's 88-96CU's....
It's impossible to know what gaming vega will be like at this point.
Posted on Reply
#85
Berfs1
Captain_Tom, post: 3898823, member: 155434"
You don't need insider information to know it's a die shrink of Vega 10 (Or from what I understand Vega 10 was almost a Beta version on 14nm).

But are you saying it's still on a 4096-SP's? Even with the still fairly large 3336mm^2 die? I heard that it's 88-96CU's....
Yes, 4096 as of the information I currently have. That would be Vega 20. Maybe another architecture (codenamed something else) may have more than that.

cucker tarlson, post: 3898829, member: 173472"
It's impossible to know what gaming vega will be like at this point.
It’s not going to be BAD, can do BF5 at I think Ultra settings (forgot what settings my friend used, but his setup is a 7700K slight OC, and heavily OC V56) and at least 1080p60 assuming you have a current gen cpu with 6 or more hardware threads. EDIT: it only gets better from what I mentioned because that was my friend’s results with a Vega 56 and a 7700K, with Ryzen 5 and 8th gen i5 CPUs being competitive contenders as well)
Posted on Reply
#86
cucker tarlson
If they deliver 1080Ti performance at 250W that's going to be fairly good. In the enthusiast segment Nvidia has two series that coexist cause AMD can only match 1080 at higher price and power draw. This can't continue.
Posted on Reply
Add your own comment