Read more.These next gen APUs are tipped to feature a mix of Zen 3 CPU cores and Vega graphics CUs.
Read more.These next gen APUs are tipped to feature a mix of Zen 3 CPU cores and Vega graphics CUs.
Hopefully they make the switch to RDNA for the Zen 4 SKUs, Vega has been great and has held its own but if RDNA is as good as some/AMD are making it out to be, needs to be used more.
Rembrandt is supposedly the RDNA2 APU for higher end purposes.
Van Gogh in the lower-end as well (can't have dGPU attached, has to rely in iGPU for everything, so this is a low-power, wide-GPU, APU for mobile, embedded, etc).
Given how AMD's been cascading new tech desktop first, this is well out of sequence, isn't it?
Even weirder that they're prepping Zen3/Vega APUs for mobile when the two consoles are Zen2/RDNA2 APUs this year; surely you'd want to utilise what you've learnt on those.
Actually, AMD created 'enhanced Vega' for Renoir.
Efficiency and performance-wise it's actually comparable to existing Navi GPU's.
When you think about it, AMD managed to enhance Vega cores by 56%. About 16% of this performance came from clock increases (roughly 30% increase in clocks), whereas the rest (40%) came from architectural improvements to Vega (and some of these enhancements will be used in RDNA 2).
If AMD transitions to RDNA for Zen 4, they will need to use RDNA 2 (which comes with 50% more performance per watt) to see a higher performance uplift (that, and more CU's).
To be fair existing Vega would also be excellent if AMD decided not to remove CU's... but its possible Zen 3 will feature enhanced Vega with more CU's (who knows).
Except the consoles have been in development for some time now, whereas Zen 3 is being finalized as we speak.
Consoles tend to use HW that's out on the market... but I wouldn't worry because the difference in iGPu-s is negligible... the main major difference will be in the CPU (ZEN2 vs Zen3).
Zen 3 is going to be a major architectural change featuring about 17% more IPC (possibly more) and some minor frequency improvements.
To be fair, consoles using Zen 2 instead of Zen 3 isn't a big deal.
They're still getting a massive uplift on both CPU and GPU ends (enough for what they need).
Which is why I don't expect to see a moving target already being integrated onto an APU when it takes 3 months after tape out to get silicon back.
Honestly, I suspect this is just some 4700G variant given the timing, level of performance and the fact it says "Renoir" in the platform name.
Not sure what the point in this is, they haven't even dropped the 4000 series APU's properly yet, why are we looking at 5000 series...
Geek stuff aside, it seems the sweet spot is a mobile workhorse apu that games at 1080p on the integrated gpu.
It seems well within grasp on a 6 core zen3 apu w/ pcie 4 nvme & fast 32GB.
There seems a flawed assumption that amd must do it alone... that games are a static problem they must conquer.
Games are what they are because hardware is what it is - they can adapt if hardware changes.
the apu cant be ignored, nor do they want to.
the apu has deep roots in consoles. If a a game plays on consoles, & consoles are apuS, & mobile is just more advanced apuS, then what is the problem?
e.g. they may not have as fast gpu cache, but they have many times more memory as cache.
a dgpu is all very well, but there goes a big chunk of mobile power envelope, cost, .. & at some point, even latency - the apu keeps both processors & the memory controller within the bounds of a tiny monolithic chip.
that sweetspot mobile would sweep that market like the 3600 has desktops.
It's fairly pedestrian even for DDR3
https://hexus.net/tech/reviews/cpu/8...adwell/?page=4
If I were AMD speccing a chip to produce to test zen3, surely we'd see a chiplet talking to an off-the-shelf zen2 IO die? That'd be the simplest way to get zen3 silicon that you can test
WOW, they still can't figure out how to sell these better. 16 CU's and add $50-100 to each model. IF you're going to sell me the same crap, at least make MORE OF THAT CRAP in there. With Intel at over 100w on desktops, WTH are you constraining your gpu for? A 100w apu will not encroach on your gpu sales at all. You are sacrificing no matter what on these, turning crap down per game. At least with double the gpu you might be able to get out of some of that, but you won't be turning everything on in anything but minecraft. Not sure why they so resist more gpu on the apu and keep this 8cu crap going. CHARGE MORE, make it better. If you are worried about sales of $50-100 cards, just charge that much more for the APU. People want ONE LESS FAN/FAIL point, especially in HTPC. I'd sell all the way up to 140w (16/24 cu), and charge $50 more for 16, and $100 extra for 24cu. Whatever, you get the point. MORE margin to you, rather than low end card sales that add noise usually and more parts to fail.
Because the number of CUs isn't the thing constraining performance, memory bandwidth is?
If you want to yell for them to slap some onboard HBM on these things have at it but if that was ever going to happen this (or last) generation would have been the moment. In the absense of that these APUs will only meaningfully move forward when they have access to DDR5 and even then with new consoles pushing more demanding graphics engines they probably won't be all that hot either.
Still, if you wanna throw $500 at them I'm sure they accept cheques in the post.
If you want a 100W APU that badly then just get an intel one and enjoy your rebadged HD530 making full use of the 100W headroom.
I actually think it's more likely next gen. Current intel desktop APUs are really CPUs with a display out, so the next launch (hopefully with shiny new Xe in desktop chips) will give some actual competition. AMD will probably still have the DDR4 bottleneck, so HBM cache (or even a separate die of normal cache) is more likely.
Intel are also talking about some new "rambo" cache for Xe HPC, so they might repeat the 5th gen iris plus graphics in their APUs.
There are currently 1 users browsing this thread. (0 members and 1 guests)