Depends how you measure.
Vega mobile is actually good. Blows intel out of the water. They aren’t an nvidia competitor… but where they compete, they are good.
Gaming laptops? Sure, i’ll give you that, but gaming laptops are a shit idea anyway as far as I’m concerned. AMD have all the console design wins, and they also have stadia, so expect there to be a lot of optimisation for AMD cards moving forward.
Woah, woah, woah.
I wonder if that’s a mistake.
That can’t be.
Read somewhere this morning that write bandwidth was a deliberate trade-off for improved read bandwidth due to the nature of client applications being more read heavy.
I’d suggest to hold off with “the sky is falling” based on a single synthetic benchmark just yet as per my assertions above…
Definitely. Warrants more investigation.
laptops for Sci Comp and parallel compute. Im not convinced on vega mobile throwing down with a MAX Q variant of a desktop nvidia gpu? I do things both mobile and desktop… Im that strange being where my laptop is as powerful as yalls gaming desktops… and my desktop is well a monster in and of itself…
LOL but I digress I suppose thats very accurate… If I was looking at an Intel CPU/iGPU combo the amd would take the cake
I suspect this is why/how AMD are able to stuff 16 cores in AM4, and still get performance.
Most end user apps more read heavy on memory IO = we allocate more pin bandwidth to read, so we can make do with less pins (than say, threadripper).
Whilst it sucks if you’re desperate for write bandwidth, given AM4 compatibility it is perhaps a clever hack (thinking outside the box somewhat) to make do with the limitations of the existing AM4 socket.
Ditto for the trade-offs they made with prior gen Ryzen regarding AVX workloads. They aren’t common, so add half the hardware and do them in 2 cycles instead of 1.
Yeah, that’s true.
I’m interested to see if we get something similar on EPYC/TR though. They too go through an IO die.
Same here.
It will be a different IO die, but maybe that will give them a differentiator between ThreadRipper and EPYC. i.e., threadripper will have half the write bandwidth of EPYC (in addition to half the pcie lanes and memory channels). Will still have more than Ryzen due to the bigger IO die, but presumably still “optimised for client workloads” (or if you’re a pessimist; “crippled for server workloads”).
Because without something like that, there’s a lot of overlap.
Definitely. I’m eager to see what happens.
Realistically, I’m running a 1950x as a workstation/server. It works great for running my company’s application stack, gaming and doing desktop stuff all at once. I really can’t fathom the performance of epyc if people are saying it’s crippled.
Also, i think its probably why “AM4 is guaranteed compatible until 2020”.
They know that next core doubling, they’re out of memory bandwidth on the socket… and this time around is for 3000/4000 with same core count, as a compatibility compromise. Ryzen 4000 will be a tweak to 3000 (at a guess, with similar core count) and then new socket for 5000.
yeah, I think they’re definitely going to need to go DDR5 or quad channel. (maybe both? Please!?)
I though this might be the place to ask.
When i’m upgrading to, say, a 3600X and don’t care for PCIe 4.0, i can go with a 400 Series MB, correct? Will those be compatible out of the box? I don’t have another AMD chip to potentially flash the Bios.
Also, any input on 3600X vs 9600K for Gaming with parallel Browsing and productivity stuff?
I don’t do nVidia. I don’t like the company, I want my FOSS drivers mainlined into my kernel, I want wayland support…I actually need wayland support. I develop a lot of wayland software, nVidia is a no go. That being said I’m inclined to agree that the Radeon VII might not be the best choice when compared to navi. Might just wait for the partner boards.
I would say you will need to verify whether or not the board is “Ryzen 3000 ready” out of the box. And - right now - i would suggest that most on store shelves might not be.
Last time around, a UEFI update was required, and this is a more significant change. I have seen several BIOS revisions to support new CPUs on my X470 Taichi (presumably for Ryzen 3000 series).
So - if you’re planning X470/B450 - you will need to likely check with the vendor - last time around i think they were putting stickers on boxes (on 300 series chipsets) if they were Ryzen 2000 series ready.
i.e.
Also, as i understand it, the BIOS/UEFI contains the AGESA for the processor. So you’ll need a UEFI that is recent enough to contain a compatible AGESA for Ryzen 3000.
Yeah, MB will be a challenge. I’m limited to mATX. So choices are slim. And non existent for x570 atm. Also no clue how pricing will look for x570 boards.
Add to that, that i could probably reuse my 2133 RAM with the 9600K and should probably upgrade that for Ryzen, i’m not so sure the Ryzen 5 is a great deal anymore…
I’m also not sure if i want an actively cooled mainboard, which all x570s seem to be.
I’d hold off a bit and see whether vendors start listing whether or not their x470 boards and b450 boards have compatible UEFI.
The X470 taichi for example got the UEFI update for Ryzen 3000 back in May. So it depends on how long it has been on a shelf i guess?
So its possible that if it is recent stock then it will have the update from the factory. But not guaranteed. I think the boards usually have the shipping BIOS listed on the box. So you can maybe check that by looking up the motherboard vendor’s BIOS page and confirm which bios version includes support for the combo AGESA (that supports Ryzen 3k).
But you’d need to have the box in hand, or confirm with the reseller whether or not that is the case.
edit:
i’m not keen on active cooling on the chipset either and have no pcie4 devices yet, so not keen on the current x570 boards either yet.
Yeah, i’m also holding off for a direct comparison for the CPU’s. Also, how RAM will influence Performance. DDR4 3000 is decently cheap, but above that, prices are ridiculous. And until now, Ryzen has been a lot more susceptible to RAM speeds.
With the 9600K being 50 bucks cheaper at the moment (without intel potentially changing prices in face of Ryzen) I’m not sure the Ryzen 5 is such a great deal. If you’re looking at the high-end (9900K and 3900X) this is more clear, but mid-range it’s a tough call.
If i can then reuse my 2133 RAM with Intel, i could upgrade to a 9700K and still spend less than for a new 3600X combo…)
I wouldn’t worry overly much about RAM speeds - i’m running DDR4-2866 on a 2700x (on non-qvl ddr4-3000, dual-rank 16 GB sticks) and performance is great.
Ryzen 3000 has huge (double size of 2000 series) caches ,so performance on that config with 3000 will be better.
Yeah, Jayz2cents also tested with 2133 Mhz RAM and it didn’t seem to impact performance too much compared to intel.
I also forgot that i’d need a Z Series MB for Intel, which are notoriously expensive in mATX.
Yup, that’s the kicker with Intel and why i find so many cpu comparisons badly skewed in intel’s favour when comparing CPU prices - you need to take board price into account as well, and even a B450 chipset board will run an 8 core Ryzen just fine. Probably even a 12 core too if you’re not planning on going crazy with overclocking.
the only big ticket feature (IMHO) you miss with B450 is SLI (crossfire still works) and dual-gpu in 2019 is such a waste of time as to be almost worthless.