If bench showed are real it will compete with Nvidia at all levels…
now let’s wait for reviews…
Looks like my next PC will be completely AMD. I plan on a new rig sometime next spring.
Now if only pytorch ran on ROCm without having to compile from source…
On the other hand, i can run my ML workloads at the company as well.
If combined with a zen3 cpu it can do some memory mumbo jumbo and it might be even faster.
Looks like AMD is tackling gamers hard! PS5 and Xbox are also full of AMD. Good times.
If those numbers translate to actual performance, AMD has really put their best foot forward this year, considering how far off they were with their previous generation.
When I saw the pricing of the 3080, I was very hopeful that meant that AMD was going to be coming with some really good cards.
And boy does this announcement look like a good reason to rush the 30xx release!
This may be the first time I’m ever getting a GPU (or any pc part) as soon as it releases.
I’m long overdue for a bunch of upgrades, running a single-fanned second hand GTX 980 Ti (backed by i5-4690K) to push VR to my eyes has been long enough.
I don’t even care about the details of the cards vs Nvidia, I’m sure they’ll be similar enough in performance and I really don’t want to make a big payment to Nvidia because I think they’re run by careless scumbags who will do anything they can get away with (FreeSync vs G-Sync, ESG rating, etc).
Try to talk me out of it
Seriously though, wait for independent benchmarks, then decide. It does look very good though.
Yep , i’m waiting on benchmarks , teardown vids and board component analysis before deciding , but AMD is looking good , and i’d sure like to do what i can to spank Nvidia some for all the games they’re playing .
The AMD cards’ listed results are all ** with things that make it not apples-apples for the most part.
Ok, so are they talking about a Ryzen 5/RX6000 vs Intel i5/RTX match? Or is it both Ryzen 5? Did they do any Intel/RX6000 benching?
Apparently an all-AMD system can enable some tricks that a hybrid blue/red PC can’t, so while a Ryzen 5/RX6000 might beat an Intel/RTX, would a Ryzen/RTX lose to an Intel/RX6000?
I don’t envy the benchers who are going to need to run these tests now to demo how it stands since AMD went the “works better with our new CPU” route.
As for “I never buy from company X because I don’t like blah blah blah”, I could say that about literally every company that has ever existed. So it does not factor into my equations on what/who to buy from in any circumstance. Maybe the company you listed isn’t as good at it and got caught. Maybe the other company did not do it in that area, but was nasty to its employees, or backed the wrong political candidate, or does business with some shady overseas country, or whatever. It’s all a wash. They will all sell their family for $1 if it will get their execs those bonuses.
I’ve bought red for CPUs and GPUs in the past, just not lately where I’ve been blue/green for several years. I will switch back without batting an eye if I thought I could get better performance for the money, but the times I’ve upgraded recently that hasn’t been the case.
rDNA2 + Zen 3 = no 256mb BAR Limit in Memory Access…
Critical for VR/4K/8K.
I’ve committed to the AMD path - I know one of their officers in Austin.
I WILL be buying at least one Ryzen 5x series with a 6-series video card. I already have Crosshair 8’s in both machines.
I’ll keep y’all posted - It’ll be roughly New Year’s when I get it done on on machine - probably the one with the 2700x and 1070ti first.
Can you elaborate on this more?
Can you explain better?
Edit @Wes beated me to it!
Looks like 2021 will be new PC year.
PCIe GPU Base Address Register, or the Amount of Memory that the PCI Bus Can access.
For compatibility with 32bit OSes, discrete GPUs typically claim a 256MB I/O region for their frame buffers and this is how typical firmware configures them.
w/ Ryzen 3 and rDNA2, the AMD Processor has direct access to all 16GB of the GPU’s Memory.
Something not Possible w/ Intel nor nVidia Products.
So you could see faster use of VRAM as a result?
Still I gather more is better as trying to change ram means getting to disk IO which is slower.
Would we potentially see Intel / nVidia firmware updates to match this or does it get too close to hardware and we’d have to wait for a new platform?
As 32-bit OS are pretty much dead in 2020, I can’t imagine they’re spending any time worrying about it at this point. Not Intel, nvidia, MS, or any of the other big players.
The issue is how much work it would be to change this. Can it be done in sw? Would BIOS/fw updates cut it? Or is it fundamental to the design?
A lot of the RX6xxxx reveal covered how their 128MB infinity cache was important because it minimizes the need for VRAM access by the GPU, which takes more time and power. Basically L3 cache for the GPU.
Framebuffer is where your picture is. Not the textures, just the final rendered image that is sent via DP to you monitor.
In 24 bit color depth every pixel is 3 bytes. Times 4k is 3 x 4096 x 2160 = 26,5 megabytes for one frame.
Do triple buffering and you need 3 x 26,5 MB = 80 MB memory.
Now do the math with 8K resolution and 10 bit color depth (for HDR) and you will see the current limit of 256 MB being exceeded eventually at some point in the future.
Intel and Nvidia will come up with a plan, too. But AMD was faster. They also defined Freesync and other good stuff. Gamers are a focus for them. Now we begin to see what that means.
AMD was faster because they designed the CPU and GPU to do this from the get go.
if I had to make a prediction, Intel will be next to give their Xe Discrete GPUs Direct Access w/ their next CPU Revision.
nVidia is pretty much out of options unless AMD and Intel allow their CPUs to interface w/ nVidia’s GPU Directly.
You Also need to remember, the PCIe Bus and Northbridge have to all support this too.
So Ryzen 3, 500 Series Northbridge and rDNA2 are required.
The 256MB Limit is a PCIe Bus Limit, it applies to all the memory, not just the frame buffer.
having a CPU only being able to access 256MB at a time to process vs having a CPU being able to address the entire 16 GB at any time.
Ok, that could be substantial!