AMD Says RDNA3 50% Perf/Watt increase and mentions RDNA4

because they don't sell online, is that correct ?

also,cherry picking is stupid.
the best choice is xfx and in that case it's 145eur for speedster quick

so, on average and cheapest vs cheapest, it's not 40 dollars. it's twice as much and 100 in eur

 

I'm only partially aware of how it works, they do have a website listing in-store prices with the option to reserve and pick up.

It can also be like with UK etailers, CCL, Overclockers, Scan etc blocking web crawlers from scraping prices to list on pcpartpicker or skinflint etc (in an effort to block bots bulk buying).

Otherwise, it's also probable that there are indeed retail only stores with no web presence or web purchasing etc but that's speculation.

 

yup
but you can't compare best amd sku at microcenter against any nvidia 3070 online retail
or just forget about 6700xt,which is just plenty at the moment
i'm trying to have some objective criteria here, he's not

you mean you can't buy 6700xt in north america now
mkay
I guess partpicker is wrong
also
https://www.anandtech.com/show/17381/amd-releases-radeon-rx-6950-xt-6750-xt-and-6650-xt
have you not seen that news too ?

sad, cause in eu there are 10x more than 6750xt to buy at 100eur less

 

And 6800 is 60% faster than 5700 XT at same ASIC power consumption with both GPUs UVed. I know that because I did that test myself.

 

wasn't really taking uv as part of the equasion
since it depends on silicon lottery just like oc
and requires tweaking

but it's an incredible result
what power though ?
and asic power ?

 

Not so much when you don't hit clock wall. I slightly reduced clocks of the 5700 XT to go down further with voltage, whereas I didn't with the 6800. So probably the latter could be UVed stronger if I wanted to.
It's also part of AMD's optimization that RDNA2 requires much less voltage to achieve high clocks. 5700 XT GPUs usually needed ~1020mV to be stable at ~1.9GHz, whereas RDNA2 GPUs usually hit 2.3GHz at 0.9V (or perhaps a little bit higher) without issues.

Both slightly power limited at 180W ASIC in Witcher 3 1440p forest scene.

 

I really did want 6800 for this very reason
it's vastly superior in efficiency over any other rx6000/3000 card
in gaming that is

 

Newegg USA Gigabyte Gaming OC (Newegg not partners)

6700xt - $549
6750xt - $549
RTX3070 - $599

Yes, they have both XT's at the same price.

 

Well, I had it and I'm cured. I don't think AMD is mentally able to provide a "high-end experience", but they still charge high-end prices. But your mileage may vary.

 

what do you mean ?

why this exact model at this exact shop ?
no other options ?

 

The cheapest on 3070 (dual fan Asus probably crap edition) on Microcenter is the same price as the Red Devil 6750XT

And the high end 3070 skus on there are very expensive

 

I mean "AMD bugs". Things that should be easy to fix or to improve, but they usually just don't do that, because their Windows support is a wasteland. Instead, there's new random breakage with new drivers every few weeks or months. It is still old ATi inside and I don't expect them to change anymore. I could describe a lot of individual issues that are unbearable and beyond stupid, but that's not really the right place.
Like I said, your mileage may vary, but I don't want AMD GPUs anymore, apart from cheap APUs in notebooks (as Intel is even far, far worse).

 

ah
I was a little afraid of that but I thought it's past them
what bugs are you referring to
it's not like rtx30 is perfect.
or has been.
they're really rather minor inconveniences rather than bugs, but what are they on amd ?

 

Spoiler: wall of rant

-I couldn't move my hardware mouse cursor with vsync (games or desktop) without causing stutter (AMD Linux devs managed to fix it, Windows team didn't, even though SpajdrEX and I spent lots of time reporting this). Terrible issue.
-driver shader cache for D3D12 didn't work at all, first loading time of lots of games like Fortnite always took ~5 minutes. This was not a temporary issue, it's an "AMD forever" issue. And it's the first loading time every time the game is run, not just total first loading time.
-unreliable Vulkan driver shader cache, creates duplicate caches and ignores old ones (who doesn't like shader compile waiting and stutter)
-both voltage control and also fan controls in Wattman for RDNA2 are garbage and require MPT instead
-bad D3D11 driver (ok, seems they've finally fixed that one)
-even worse D3D9 driver
-their Vulkan driver can't do proper triple buffered fifo vsync
-two crappy fps limiters in driver (Chill & FRTC) instead of a single proper one
-driver overlay caused stutter with VRR in some titles
-you can't force AF with D3D12
-driver control panel often is botched in many driver versions and crashes repeatedly, often "silently" in background and causing stutter
-more visual corruption issues than with Nvidia, e.g. not even screen space reflection in (once used to be) important titles like BFV look correct for years
-you can't force vsync via driver
-stutter in Firefox VP9 with hardware decoding, not there with Nvidia (at least not when forcing max clocks)
-no way to force max clocks in case power saving feature screws up
-more performance regressions in games than with Nvidia, be it linked to some feature like vsync or just in general
-corruption issues with RT, be it missing reflections (WD Legion) or else (Control Astral dimension)
-bad frame time variance (stutter or micro-stutter) with RT, be it in WD Legion or Doom Eternal (much better/perfect with Nvidia)
-windowing issues with Vulkan on Windows, DXVK apps or mpv might get stuck on switching fullscreen (another eternal AMD issues)
-calling FSR 1.0 atrocity "Super Resolution" to fool public and customers
(-well, now initial FSR 2.0 version also seems to be a bit too bugged for my taste.)

Thanks for reminding me of all this crap. Please do so again in the future before RDNA4 release, so I won't be tempted.
Yes, Nvidia has issues too. But they feel like going on vacation in comparison...

And to sound like a total d*ck: Yes, lots of AMD users will dispute those issues. But as an Arch Linux user who patched and compiled drivers to provide devs with feedback etc., I don't need to care about that mortal coil.
Chances are not low I know better. Sorry, guys.

 

whoa
at least it wasn't boring

when I was referring to nvidia I meant more like adding g-sync compatible displays that work to the official list

 

Despite some warnings in few games still havent noticed anything but its only matter of time.

 

That's not how MCM works though. Its nothing like Crossfire or SLI where the GPU's are treated as two pieces of hardware and the rendering is done mainly using methods like Alternate Frame Rendering where one GPU would render the odd numbered frames and the second GPU would render the even numbered frames. Or Scissor Mode where GPU one renders the top half of the screen and GPU two renders the bottom half and many more methods too but AFR seemed the most common. Not to forget but the memory pools of each GPU are mirrored as well for these older methods to work.

In an MCM design the GPU chiplets are seen as only one GPU. So the frames data is sent to the GPU and is only returned to the screen when the rendering is finished. The GPU will handle all the data by it self and split the data how it is programmed to do so by a hardware scheduler on the I/O die. So there should be no induced microstutter that plagued older types of multiple GPU's. Remember AMD has had hardware based schedulers inside their GPU's since GCN1. So they have had a long time to get to gripes with them and further improve them. Its why GCN suffered in DX11 games as AMD bet big on low level API GPU access which never came with DX11 and Nvidia added a software scheduler inside their driver which improved performance in DX11 games. This is also why older GCN cards perform so well for how old they are in newer DX12 and VK games.

Rumours are also pointing towards only the top end AMD 7000 series GPU being MCM. The rest are supposed to be monolithic. The big MCM design is probably going to be AMD's version of a TITAN class card with a price to reflect that.

 

A year later and he is still being an nvidia fangirl up in here. Glad I still have the ***** blocked.

 

RDNA3 are not MCM design, still monolithic with Cache chiplets.

 

wasn't the point of mcm to bring the cost down ?