So is the TekSyndicate guy going to off himself at this news? What next? Zen in 2016 being comparable to an i3?
Damn, and this was supposed to be their big day to announce the Radeon Fury X. At the last second Nvidia comes in and releases a better card for $200 less.
This is AMD's 6
Just to further dance on AMD's grave, now Broadwell destroys all the AMD APUs for gaming on an IGP
I think it's over for AMD.
They'll be ok. Between the console hardware and the fact NVidia shouldn't want them to die and get anti-trust scrutiny, they'll survive.
They do need to do better.
I think Cry Havoc is the TekSyndicate guy tbh
I meant I think it's over for AMD being a viable performance option in GPUs like it is in CPUs. Nvidia won't kill them outright for the same reason Intel won't, like you said, so they're not legally a monopoly. But I guess Intel is eventually going to end up as Nvidia's real compe ion.
Last edited by baseline bum; 06-02-2015 at 09:57 PM.
Nah, CH would never say the FX-8350 was a better gaming CPU than the i5-3570k man. And man, Cry is ducking this thread like cuck was with your mspaint one.
Cry Havoc man, you still alive? AMD fan is still much higher than console peasant, we still got much love.
Yeah, a new era with a bunch of ing old cards
Maybe I'm writing their epitaph too early. That is a giant ing die there:
Supposedly that Fiji card can draw up to 375W:
http://www.hardwareluxx.com/index.ph...tx-980-ti.html
That's ing crazy. When the entire industry is optimizing performance per watt, these s are going the other way around
I bet you the thermals are terrible too. Heck, I was checking the 295x2, and couldn't believe those things come with liquid coolers from the factory. Ridiculous.
If that's really 375W with lower performance than a 980 Ti while using lower power HBM instead of GDDR5 that will be the epic fail of epic fails. But 375W just refers to using two 8-pin PCIE connectors (150W each) and the 75 W it can get from the PCIE slot. I have been reading it's a 300W card, which isn't a big deal since 980Ti isn't exactly efficient at 250W either.Supposedly that Fiji card can draw up to 375W:
http://www.hardwareluxx.com/index.ph...tx-980-ti.html
That's ing crazy. When the entire industry is optimizing performance per watt, these s are going the other way around
I bet you the thermals are terrible too. Heck, I was checking the 295x2, and couldn't believe those things come with liquid coolers from the factory. Ridiculous.
Last edited by baseline bum; 06-03-2015 at 01:26 AM.
Then again, drawing power usage conclusions from the pin layout doesn't make sense. The R9 295x2 goes way out of spec with only two 8-pin connectors considering it draws significantly more than 375W. The anandtech review has a 4960x system drawing almost 700W with a 295x2, and I doubt more than 250W of that is from the overclocked CPU.
You have to to think a lot of AMD going with a liquid cooler is them not wanting a repeat of the fiasco with the 290/290x reference coolers. The reference cooler was amazingly bad and ridiculously loud. Not FX-5800 loud, but still pretty awful. ElNono man, remember the good days at your old job with the FX-5800?
Back on topic though, the reference cooler on the 290/290x was so bad the core shot up to 94C immediately and then had to constantly throttle. And then Sapphire came out with the Tri-X cooler that keeps temperatures in the low 70C range with room temperature ambients.
The reference fail 290/290x
The awesome Tri-X 290/290x
Because of that awful reference cooler the 290/290x got reputations as being ridiculously hot and power hungry cards even though Nvidia 700 series were also. It's just they had a reasonable reference cooler. But I don't think there was a lot of difference when using quality aftermarket coolers.
There's a reason to optimize energy efficiency (and thus, thermals), it's not just style points. The biggest roadblock to reduce nanometer process further is current leakage. If you're using inordinate amounts of power, you're not going to be able shrink your die, your thermals are only going to get worse (as you add more transistors to the same process). It basically doesn't scale.
Intel figured this out a long time ago, and starting with Sandy Bridge, it started to reduce power usage while shrinking the die (because that was the only way to go, raising the clock was no longer feasible). That's how you end up with a 4-core, hyperthreaded Core i7 consuming the same amount of power as a single core Pentium 4 from yesteryear. NVidia started with Maxwell, same reason.
AMD is going to have to head the same way, because making bulkier GPUs that cost more to cool down and don't outperform the compe ion doesn't sound like a winning strategy.
nm
I actually caught your post before you edited it. Didn't know about the foundry issues, but makes sense. They really need to scale the process down.
What do you think of the FinFET transistors AMD is trying to use at 16 nm for 2016? What advantage does it have over a traditional MOSFET? I don't understand how you can keep shrinking and slow current leakage at the same time, you're scaling the gate oxide thickness down too which makes quantum tunneling all the worse.
Last edited by baseline bum; 06-03-2015 at 02:13 AM.
FinFet deals, as you said, with the fact that if you shrink the MOSFET transistor gate electrode too much, the transistor efficiency just drops too much, and if you can't electrically control the gate with enough precision, you have to overshoot, which then causes current leak. So what they did is redesign the source, drain and electrode with a thin film to improve electrical control. The improved control can be built at a smaller scale and reduce current leak.
Here's a video that compares them and tries to put it in somewhat layman terms:
IIRC, Samsung already can build FinFet at 14nm. As usual, the question with these new designs and processes is what kind of yield they can get. The design itself is somewhat proven as it's based on a 90's single-gate transistor design (depleted lean channel transistor).
EDIT: FWIW, I read NVidia will be shipping cards with 16nm FinFet by the end of the year... they could also wait for Pascal..
You're just figuring that out?
http://www.spurstalk.com/forums/showthread.php?t=241539
Also, my main concern from that thread looks like it's justified
DJR210, you getting this ?
Let's hit Best Buy my nigger.
I'd rather slice my urethra 50 times, and shoot lime juice and salt down the shaft than get that POS.
There are currently 1 users browsing this thread. (0 members and 1 guests)
Posting Permissions
Reply With Quote
