My RX 5700 XT uses GDDR6 btw. It’s already being done with their latest released GPU’s in RX 5K series.
Quite positive most of the RDNA 2 cards will use GDDR6 Memory since the first series did as well. Even Nvidia was using it in their GPU’s in RTX 2K series. Should be interesting to say the least.
Yep but the memory controller is a bit sensitive and there’s two manufacturers Samsung and Hynix with the only way to know which being to remove the cooler and check the chips.
Samsung’s robust but the timings differ compared to Hynix and they use one less strap around 1870 - 1900 so overclocking can be a bit touchy.
Hynix is a bit more varied but cited as a better overclock although I believe they also use more lax timings for their respective memory straps for different speeds.
Third party coolers make a bit of a mess with the sensors and junction temps both the GPU core and memory core and the fitting is kinda finicky though AMD seems to be find with around a 90 memory junction temperature in Celsius (110 for the GPU core.) though I am not sure how the memory behaves in regards to high temperature such as reducing timings or error rates.
(VRAM has error correction too so since voltage and downclocking is normally locked only overclocking remains and it can be tricky to figure out where this starts kicking in and performance decreases even if it’s subtle.)
The 5600 initially having 12 GB/s modules I think is how that was rated and then AMD started giving out bios files with 14 GB/s likely also meant some difference in binning and quality but that’s pretty common to avoid wasting the chips or wafers by not discarding still usable dies.
With the new GPU’s I hope they’ve improved some of this but we’ll see what the RAM chips actually are and then how AMD configures them and their clock speeds.
NVIDIA overclocked theirs to 15 GB/s for the 2080Ti (I think it was?) though it won’t match the 19GB/s GDDR6X modules but the usage of infinity fabric and the cache here even with a lower bus width could make for a interesting performance improvement but we’ll just have to see.
(If it’s linked in some way to clock speeds and other factors like power draw or voltage overclocking and undervolting might get yet more complicated though.)
Wonder what the clock speeds will be as well, NVIDIA is going with ~ 1750 usual boost speeds for the 3080 up to 1900 on well binned well cooled chips and AMD might be trying to push into 2.1+ properly with RDNA2 possibly even above 2200 but the effective clocks versus boost could be quite dynamic.
There’s a ton of sensors and all kinds of data and factors for how this is handled now on both AMD’s GPU’s and CPU’s and also NVIDIA’s GPU’s and Boost V3 and this I guess V4 on Ampere.
EDIT: Although a direct comparison between the two in flops or gigahertz isn’t really indicative of much, they do a few things fairly differently in hardware after all. 
Resulting performance and things like power draw could be though. NVIDIA’s going with a lot of core power and increase in stuff like ROPs and also the big push on memory performance again.
Some of this leaked early through Twitter though and seemed to be about as expected although with a surprise uplift on RTX performance with rasterization performance being as assumed though still pretty good. (Up to I think the number here was 1.8x so maybe not a full 2.0x the 2080 but it got close.)
AMD might be able to have something close to 2x the hardware on the 5700XT going by data but that’s not a flat 2x performance much as that’s a common comparison and also now there’s comparisons with the custom console hardware which I don’t think is a good measure but could be a glimpse into RDNA 1.5 / 2 and whatever it gets called.
(Bit of a difference with the PS5 GPU versus the XSX one it seems.)
EDIT: Up to is a good one as usual, DOOM Eternal as it scales well and just about a 2.0x result here with some difference in reviews and then the average I think is 1.6 to 1.7 so 60 - 70% faster. 
And of course some games or games engines scale terribly but could gain more from newer drivers and fixes or improvements here reducing results like I think it was Borderlands 3 at a 40% gain around the lowest and Ghost Recon Breakpoint might have been busted or they used D3D11 instead of Vulkan as it barely scaled at all hitting CPU limits or something.
EDIT: Think that’s about it, good to see a 50% increase and up.
AMD being competitive in the GPU segment would be nice but there’s still a month or so until there’s some hopefully more detailed info available.
IF I run my GPU on 4.0 PCI-E Bus Speed it hits 16GT/s But its hovering at 8GT/s For PCI-E Link Speed. Not sure if its really due to Card running in PCI-E 3.0 Slot. Though its memory is clocked at 14GB/s
This is my Card actually: ASUS ROG STRIX RX 5700 XT GAMING OC
How did that work again.
It splits it a bit depending on if it’s 3.0 or 4.0 mode for these lanes and how it drops from 16x to 8x or lower.
Might be more for the 5600 running at a 8x PCI Express connector so dropping to 3.0 shows a more significant performance drop-off as a result since it’s a bigger cut in overall bandwidth.
And for AMD the 570 has 4.0 CPU and chipset whereas the 550 uses 4.0 CPU and 3.0 chipset.
Usually the topmost PCI Express slot covers the 16 lanes 4.0 and then the first M.2 slot under that has 4.0 mode both to the CPU.
Then it gets a bit tricky for what the motherboard vendors do with the additional chipset lanes.
(Gigabyte throws on some additional M.2 ports for example but filling these in disables the SATA ports in turn.)
Usually the manual has a better overview for how this is resolved for each motherboard.
(Plus the CPU used and how many lanes it can deliver.)
EDIT: Seems I have missed a lot of these later hardware advancements and features even when reading up on a bit of it to try and keep updated, well at least it should be less tricky than the USB situation if that’s something.
3.0, 3.1 and 3.2 with A, B or C connectivity I think is how that works and cabling too for A (Box one) and C (Rounded one but you never have to rotate it half around first before it fits.) then two generations and separate up to speeds here as a short of how it’s done.
4.0 and Thunderbolt parity and once that’s out throw it all on the motherboards and toss out 3.x entirely.
(Or not, 2 - 4 USB ports might be a bit less than some enthusiasts prefer. 
)
(Particularly when the input stuff has a extra cable for USB pass-through functionality or fancy LED’s or both thus that split keyboard connector that’s common enough nowadays.)
You know what’s weird?
That’s been a feature of Windows since Windows 8.1 and still virtually no graphics engines for games use it. It’s a really neat feature that lets you draw your game’s UI at a rate and resolution independent from the rest. So for example, you could draw natively at 8K/60 for your UI while drawing the game at 4K/30.
I am adapting Special K to use Multiplane Overlays whenever possible, because it represents a better way of handling API interop than injecting code straight into Vulkan or D3D12. The features for running the UI at a different resolution / refresh rate than the rest of the game are a tad complicated and while I might get around to experimenting with that in the future, just getting MPO-based rendering up and running will be an impressive feat.
Would that be possible in exclusive fullscreen as well?
Which part?
I mean, technically, engines have been hacking their way toward something like this forever simply by using different resolution rendertargets for the HUD and main scene. That’s actually not identical, multiplane overlays are capable of two different VSYNC behaviors in the same application, as well as the overlays can mix and match pixel formats and color spaces.
It’s actually possible for lazy developers who don’t want to learn how HDR works to scribble their SDR overlay on top of an HDR game and have multiplane overlays handle the format conversion. Don’t tell anyone this, they will be taking the easy way out and prolonging the status quo of truly terrible support for everything HDR on PC
The use of MPOs themselves. I assumed that it was the same type of overlay that the Windows game bar makes use of, which is only accessible in non-exclusive mode.
So I basically mistook it for a type of overlay created by a separate process on the system, outside of the game — not one created within the same context of the game itself.
@JonasBeckman Seen the state of the RTX 3080 stock situation in Sweden? Seems multiple retailers now lists January 2021 as the earliest when the GPU will be in stock again.
I… I don’t think I’ve ever seen that long of a delay before after a new GPU release…
Yep, October, November and now a listing of 2021 I assume is due to a ton of back orders needing to be filled and a lower than expected quantity of GPU’s which even NVIDIA acknowledged recently doing their apology for the shortage of GPU’s.
3090’s are stocked now at least but it’s the same situation far as I can tell where they unlock for purchase at a set time tomorrow and the quantity seem to be about one third of what some of the 3080’ were so 5 cards instead of 15 for what I can see.
Pricing around 1700 to 2000 Euro too roughly.
EDIT:
Webhallen.
Zotac Trinity - 5 GPU’s, 1800 EUR (Well it’s not quite 10 SEK = 1 EUR.)
MSI Ventus OC - 1 GPU, campaign price (With a stock of one!?) 1700 EUR
MSI Gaming X - 2 GPU’s, 1850 EUR
Gigabyte Eagle OC - 2 GPU’s, 1700 EUR
Gigabyte Gaming OC - 4 GPU’s, 1850 EUR
Palit Gaming Pro - 1 GPU, 1860 EUR
Palit Gaming Pro OC - 2 GPU’s, 1900 EUR
Gainward Phoenix GS - 3 GPU’s waiting to be shipped but none in stock, 1900 EUR
I haven’t seen this before… Apparently Uplay’s overlay warns you when RAM is almost filled out (my system is at 95%).
And on that note, I need to pick up a new 16 GB stick so that I can go back to 32 GB… At the time I didn’t think I actually ever used that much RAM but since going back to 16 GB I seem to max it out every now and then… 
Uplay did just update earlier today but that might be part of some earlier update though it’s not something I’ve seen before.
Went with 2x 16 GB sticks for this system but I don’t remember ever getting close to filling it up normally for the prior 16 GB RAM in the old system short of leakage which is usually very uncommon.
Watch_Dogs 2 (?) might hit some issues depending on settings though but it was mainly CPU and GPU limiting due to the extended draw distance slider and mostly just when going above 50% where much of the smaller city props are already showing up in the distance.
Never tried MSAA though (Costly and ineffective after all.) though TXAA utilizing MSAA+TAA could perhaps add a bit extra to memory requirements.
A later update also introduced SMAA T2X which is solid enough but temporally unstable (Tons of ghosting.) if you can’t get a stable framerate ideally 60 or higher.
(Cry Engine 3 Crysis 2 example.)
EDIT: Gah staring at that is almost as bad as chromatic aberration.
Game has some sorta TAA mode too other than T2X but like many earlier methods it’s pretty poor quality wise although like many other TAA modes it is fast with better coverage than FXAA or SMAA on it’s own.
(Or MLAA if using AMD and forcing that but I’m of the opinion it should have been replaced with display driver implemented SMAA ages ago since it’s kinda a improvement of MLAA anyway.)
(Alternatively the highest quality level FXAA.)
On that note 6 GB RAM for the game 2 - 4 GB for the OS and 6 for the browser yeah that’d do it.
Tabbed browsing with Firefox or Chromium can take a good amount of memory after all. (Chromium meaning something like all the others at this point besides Firefox? )
EDIT: Well it’s more like 2 - 3 GB I suppose and it’d be compressed and ready to unload after a while too plus inactive tabs and what not for conversing memory usage.
~ Solid good quality ad-blocker or script blocker to really reduce RAM usage for some reason.
I believe this is what’s called having your cake and eating it too
Special K has always had insanely smooth framerate limiting, but here we have a substantial boost to minimum framerate combined with a 10 ms reduction in input latency…
What’s scary about this, though, is the performance with no framerate limiter. It’s up in the 100 ms range without SK or RTSS, and it’s a VR-optional game. I really hope that is not indicative of how that game performs in VR mode.
What’s up with the weird 0.1% percentile FPS of Unlimited and RTSS ? Is Special K just pacing the frames better to prevent such weird drops?
That’s the magic of aligning your framerate limiter’s clock to VBLANK And some pretty nifty clock drift compensation code I designed that sneaks in occasional clock re-syncs only when re-syncing the clock can be done without inducing a hitch.
I basically always stop delaying frames at precisely the same time that the DWM compositor has swapped the front-buffer out and is ready to accept a new frame. Thus when I tell the engine to continue, it can do so with zero back-pressure from pending VSYNC.
We’re not even playing the same game here, RTSS may try to count the frames and limit them but it’s not taking into account the traffic cop that is the DWM. RTSS releases the GPU into oncoming traffic and causes accidents frequently because its clock is not sync’d up with the DWM
You can basically look at any framerate limiting done by RTSS and you will always find this problem. It can count frames, just probably shouldn’t be directing traffic since it’s not paying attention to all the collisions caused by its bad clock.
@Aemony does some frame-limiter page exist in the PCGW? I feel like it’s the sort of thing that would make sense, to explain the different frame-limiting options, partially because most users don’t know where to go besides the PCGW if looking for the best frame-limiting tool.
Also don’t take this as me asking you to create the page 
There’s been a discussion about this if you search PCGW… I would advise you don’t. Evidently I am
“opinionated” and “believe that borderless window mode is superior to fullscreen exclusive.” That’s not my opinion
My opinions are not published on MIcrosoft’s website. I just happen to be familiar with this subject where evidently nobody else is? 
And for what it’s worth, if pigs fly I might be able to get my limiter design into RTSS and then “best framerate limiting tool” doesn’t really matter. Everyone can go back to using RTSS and there would be no difference between SK and RTSS. But I have a hard time believing I could get Unwinder to go along with this unless I have some super amazing backing from like 5 or 6 journalists
Did DF ever get back to you about Special K?
I’m wondering if you’re better off prioritising publicity (aka a DF video) over implementing your frame-limiting in RTSS, which could result in SK being overlooked by those who care for frame-times. DF in particular are obsessive when it comes to frame-times, so your frame-limiter would seriously win them if they actually check it out.
That’s not to say don’t get your frame-limiter working in RTSS, that’s something you can still do after SK becomes more widespread.
