Discussion in 'Article Discussion' started by bit-tech, 11 Jun 2019.
Anyone wanna sell me some self control?
That 3950x is going to make waiting for Zen2 Ripper way too hard
They are too expensive IMO.
They are just as pricey as Nvidia tax, once you remove the features the Nvidia cards have over the AMD ones. Just far too much money for what is apparently shrunken Polaris.
They also have an 8 pin and a 6 pin and very high clocks, and I will assume power consumption also and therefore heat. That coupled with a blower is a bad idea.
Ed. I am, of course, talking about the two new GPUs.
The 3950X has me in something of a pickle. I want one, largely because I don't want a "crippled" chip... but dang if that isn't a jump in price.
I'm sadly tied to nVidia GPUs, until I can convince someone to pay me for a year or two to rewrite a shedload of CUDA code into OpenCL... if it's possible to maintain reasonable levels of performance in the process...
...that said, I could get an AMD card for a pure-Windows/gaming rig...
Absolutely not shrunken Polaris - same performance as an RTX 2070 with the same amount of CU's as a 580/590, would need more than a 400Mhz clock speed boost to make up that difference.
I'm not expecting AMD to beat Nvidia on performance-per-watt with Navi, but they look to have closed the gap IMMENSELY. If the slides are true and it's, on average, 5-10% faster than a stcok 2070, for 20-30W more, i'd call that a win.
Those Vega 56 cards at ~£250 still look pretty good to me!
They're priced at the current price : performance brackets, AMD would be mad to do so otherwise, but things might start getting interesting once nvidia release their refreshed 2070 super ti ultra cheesecake, rumored to being brought out at current 2070 prices, with the original recipe 2070 getting a price drop.
+1, but then this generation is too expensive overall. Mid tier has taken a hammering. I think everyone was hoping AMD would bring down the mid tier back to what it was in price, instead they've just priced it the same as Nvidia, actually worse looking at the 2060.
Cheapest 2060 on OCUK, £299.99. Sure it's got less VRAM, but, love it or hate it, it also has RTX. If AMD $/£ swap, the 5700 will be £380, for 2GB more VRAM, no RTX, slightly more performance (presumably), and more power usage (presumably). And that's probably just for the reference one.
XT is a bit better, 2070 is also £450 at it's cheapest on OCUK. Same VRAM, slightly worse performance (presumably), no RTX, and less power usage (presumably).
The tin foil hat in me is wondering if this is all a way to push people towards the streaming services that are coming out.
More likely to be just a consequence of whatever TSMC (who manufacture GPUs for both team green and red) charges AMD and Nvidia.
Maybe, but foil has always been good at blocking out sensible waves
What you're not taking into account is that Nvidia is already pre-empting the disruption to the performance hierarchy with their "Super" line - rumours are they will be direct replacements for current, non-super, 2060 and 2070's, with those older cards receiving a price cut. Nvidia cutting prices means AMD cutting prices to match. I wouldn't be surprised to see $50 knocked off the MSRP for both Navi cards once everything is out and the dust has settled. This is good for everyone.
Edit: In AMDs case, ignoring the switch, streaming or consoles all help AMD if I'm not mistaken. So it's win win, pay more for GPUs, or pay for consoles/streaming services.
Yep, I wasn't. Mostly because I don't think there's much info out there. Have we got any more info than just "something super is coming"? Rumours are nice, but Nvidia might just decide to look at the competition and push the prices up because they can tbh. Maybe not linearly, we might get say 10% more performance for 5% more cost, so long as they beat AMDs parts they can turn round and say but we're faster and have RTX, and price/performance has improved since the originals so the consumer is getting a super deal!
Sure, they could just price it the same, or they could price it lower. We won't know until they do it. In all honesty, I was expecting Ti versions of the cards, but I don't know if they have the performance space to do it or not.
$50 would be a nice price cut (what price cut isn't?), but it'll still leave the mid range at £330 for AMDs part for example, and I'd imagine the 2060S would be around the same.
Yarp - which is a bit embarrassing for Nvidia, after positioning its GeForce GRID platform as the go-to solution for cloud gaming for years prior.
Well, there are three big players in streaming at the moment: GeForce Now, PS Now, and (maybe, assuming it is a smash hit or Google will just let it die like they do with other services that don't do well) Stadia. GeForce Now is based on Nvidia cards, PS Now is basically a farm of stripped down PS4s (AMD semi-custom), and Stadia is a bunch of cut-down Radeon Instinct MI50s (+ Intel CPUS). Though of course, GRID is mostly hand-me-downs from the enterprise market which Nvida has pretty much cornered for remote GPU (and GPGPU) for thin clients. If they can also sell it for gaming outside of Geforce Now great, but I don't think they'll be losing much sleep over it.
With cost/transistor continuing to go up and up with each process shrink, GPUs are only going to get more and more expensive per unit area, and therefore per unit performance: to go above ~2GHz (which everyone has been hitting for a while now) is going to need area 'wasted' on doubling or tripling up switching transistors (as is done in CPUs) which is not a good tradeoff vs. using those same transistors to go wider, and going wider means more transistors which means more cost. The extremely high parallellism makes splitting them into chiplets a much more daunting task than with CPUs (forget dealing with corralling ~16 threads across different dies, try several thousand threads most of which are not independent and want to access each others' data).
You forgot MS' xCloud which is basically PSNow... only with xbones...
And inside every Xbone is an AMD SoC. Chalk up another cloud win for Team Red.
I wonder how the 'win' works out in component volume sales when capacity factor is taken into account: if your cloud hardware is the same APU as the console hardware (i.e. you take the 'cut down consoles into racks' approach rather than the virtualised GPU approach) then each cloud hardware sale could potentially have been a console sale. But each 'cloud console' can serve multiple end users, whilst a console spends most of its time idle (e.g. if your average user plays 2 hours a day, a cloud console could serve up to 12 users in a 24h period, depending on capacity factor, local timezones, and max concurrency design). This couldlead to a drop in "now I don't need a console!" users that would need to be offset by an increase in total users in order for total number of APU sales not to drop. Dropping number of sold consoles is in the favour of MS (or Sony) because it sustains - or increases - software sales while decreasing hardware outlay, but this is not in favour of component suppliers. On the other hand, streaming using dedicated hardware and/or exclusively targeting users outside your existing offerings (e.g. Stadia) does not cannibalise hardware sales, and the higher margins for virtualised GPU cards may serve to offset the lost sales of commodity hardware.
tl;dr: If it works out that all your streaming users are people who would never have bought a console, then its a net increase in component sales. If your streaming users are people who have taken it as an opportunity to avoid purchasing a console, then it could be a net reduction in component sales.
With so many games announcing ray tracing support, and the next gen consoles having it, and almost certainly Navi 2 having it, then spending that much on a new card without it seems a bit short sighted. If AMD don't have hardware support for it yet they'd have been better of releasing cards to compete with the 1660 up to 2060 range where that doesn't matter so much.
Separate names with a comma.