AMD: Speculation, Rumors, and Discussion (Archive)

Status
Not open for further replies.
call us pessimistic, but when have you seen a launch with so many leaks of benchmarks :)?

for the 200 bucks this card is what it is, a performance performance, coming to the midrange segment, nothing particularly special. It is where it should be, there is no redefining the segment with increased performance. It should do well until the 1060 comes out (depending on how well the 1060 performs and power usage of course, but at least one of those are probably going to be better than the rx480 from what we have so far seen with Pascal)
 
Last edited:
If you're talking about EarlyZ then it's part of hidden surface culling. Construction of pixel shader warps after EarlyZ seems like a sane choice to reduce divergence. But don't TBDRs already do this? For compute I don't think you can apply culling before running the shader as there is no concept of EarlyZ in compute. It's just data in -> compute shader -> data out; can't make any assumptions about the data itself.
Not quote EarlyZ, more like LateZ or other situations that arise after invoking the shader. Cases where you decide the pixel is no longer needed in the middle of the shader execution. Fully transparent pixels or an edge detecting post processing effect for example. Situations where the pixel shader doesn't actually output anything.

Doing stuff in shader certainly gives you more programmability but it's not so good for power efficiency. Primitive discard in all GPUs has been fixed function since ages. It's a fixed function task meant to be done more efficiently by fixed function units.
Generally I'd agree, but a scalar processor could theoretically have instructions to accelerate it and the ability to go parallel if required. A shader bouncing between scalar and vector operations. I was also laying out an additional culling opportunity in addition to the typical fixed function usage. it would likely apply more to the wave level instructions that are currently being added to HLSL.
 
You are a lil pessimistic...
I think it is a solid card, and looks to be a big improvement over 380/380X at a price that will appeal to quite a few people.
But the problem is expectations, same was seen with the 1080 where some were disappointed a heavy OC 980ti may be equal/better than an OC 1080, forgetting they are different architectures in terms of GPC and SM and with more flexibility from an optimal performance envelope for previous gen (previous gen was hitting 1.2V to 1.25V without special bios while seems limit is around 1.1V to date with GP104 even though there still seems to be the same internal 1.25V limit in place identified by extreme 'DIY' HW OC setups)

Cheers
 
Last edited:
call us pessimistic, but when have you seen a launch with so many leaks of benchmarks :)?

for the 200 bucks this card is what it is, a performance performance, coming to the midrange segment, nothing particularly special. It is where it should be, there is no redefining the segment with increased performance. It should do well until the 1060 comes out (depending on how well the 1060 performs and power usage of course, but at least one of those are probably going to be better than the rx480 from what we have so far seen with Pascal)

I think it is a solid card, and looks to be a big improvement over 380/380X at a price that will appeal to quite a few people.
But the problem is expectations, same was seen with the 1080 where some were disappointed a heavy OC 980ti may be equal/better than an OC 1080, forgetting they are different architectures in terms of GPC and SM and with more flexibility from an optimal performance envelope for previous gen (previous gen was hitting 1.2V to 1.25V without special bios while seems limit is around 1.1V to date with GP104 even though there still seems to be the same internal 1.25V limit in place identified by extreme 'DIY' HW OC setups)

Cheers
From 380x to 390 there is a good range. IMO what will make the difference is the over-clock potentials.. This last will of course have to deal with the cooling system. But custom coolers are coming D:
 
I think it is a solid card, and looks to be a big improvement over 380/380X at a price that will appeal to quite a few people.
But the problem is expectations, same was seen with the 1080 where some were disappointed a heavy OC 980ti may be equal/better than an OC 1080, forgetting they are different architectures in terms of GPC and SM and with more flexibility from an optimal performance envelope for previous gen (previous gen was hitting 1.2V to 1.25V without special bios while seems limit is around 1.1V to date with GP104 even though there still seems to be the same internal 1.25V limit in place identified by extreme 'DIY' HW OC setups)

Cheers
It may still be too early to really judge its performance. It's entirely possible some of the architectural improvements are partially or heavily software based and not necessarily implemented or optimized in the drivers yet. They may also benefit DX12/Vulkan in ways that we haven't seen either with the upcoming HLSL/SPIRV. Regardless, I'm really hoping they release a whitepaper and some architectural details later this week.
 
anyone with eyes can see that the ref 480 isn't going to OC well. But I think that's fine. Its the play NV ran with for 970/980/980ti. let the stock be the marketing TDP number, let OEM's ramp up the volts/clocks/power and watch as everyone quotes the ref TDP :) . One thing to be said depending on how much voltage control WATMAN allows, the OC control panel looks awesome!
 
The 7870 was faster than any previous gen GPU while using much less power, so aside from price ($200 vs $350) I'm finding it hard to be impressed by the 480. Price is certainly important, and I will likely never spend more than $400US on a GPU - but with 28nm we at least had the option for not only something faster and cheaper than last gen, but something even faster still at a similar price. Nvidia are failing here also because of the inflated FE prices creating an artificial price floor for the time being.

I guess the issue with my expectations is the inefficiency of the 40nm generation vs the improvments made with the 28nm architectures more so than 16/14nm chips failing to deliver. I will wait for benchmarks, but unfortunately I really want something in the 980TI performance range for $300 or 20% faster for $350. I can't see such a product on the horizon.
 
a "factory OC" ~1.4-5 ghz 480 not good enough for you ?
Remember we had OEM's asking customers if they would be interested in AIO water coolers for 480's, my guess is they wouldn't bother even asking if there is no room/capabilities for clocking higher.
 
Last edited:
From 380x to 390 there is a good range. IMO what will make the difference is the over-clock potentials.. This last will of course have to deal with the cooling system. But custom coolers are coming D:


Over all its a good card for where its a placed and the price its coming in at, I'm just hoping the OC versions aren't power guzzlers, no need or a hummer h2 to be put into the upper midrange card section, if nV had half a brain *which they do, they would just make a gtx 1060 ti, and it would be over. I think this is exactly why AMD doesn't seem to be keen on binning these chips themselves and making a higher clocked 480 out of the gate. Just have to see of those leaks are real about the power consumption when overclocked.....
 
a "factory OC" ~1.4-5 ghz 480 not good enough for you ?
Remember we had OEM's asking customers if they would be interested in AIO water coolers for 480's, my guess is they wouldn't bother even asking if there is no room/capabilities for clocking higher.
Of course they'll bother. Their goal is to create demand for their flavor of the card over someone else's. If people prefer flashy water coolers or 88 phase VRMs, or color LED laser shows, or quad xacto-blade fans or dual 8-pin PCIE connectors, OEMs will make them. This doesn't mean they meaningfully help overclocking. Case in point: Stock GTX 1080 FE overclocks reasonably well on air, to around 2.0 and 2.1 GHz. There's dozens of "boosted" designs with a wild variety of coolers and pins and heatsinks. These cards with the fancy additions overclock reasonably well to around 2.0 and 2.1 GHz. Hey, wait a minute...
 
I wouldn't bet on the R480 having a significantly higher peak(!) performance than the 390, if at all. Even considering only the publicly known/leaked details, it's safe to assume that the worst case performance has improved a lot.

Polaris mostly revolves around resolving old inefficiencies and bottlenecks. Not so much on raw performance.

Oh, and I think it's safe to say the base for Polaris isn't Tonga. It's clearly a Fiji successor.
 
in europe, checking my importer prices:

RX480 8G - 235€ + tax
R9 390 8G - 284€ + tax
GTX 970 - 256€ + tax


Sure, the new gen *80 class jumped one performance class above the previous 380. But I cant feel anything but disappointment knowing that i do not need the 235€ RX480 to have the same performance that a previous gen 256€ card can achieve today. 20€ separate the two.
 
c4plDlh.jpg
Edit: Link provided.
The first gaming performance results of AMD’s Radeon RX 480 graphics card have been leaked by Videocardz. The performance results were originally posted by CD-Action (A Polish magazine) but have since been removed due to NDA.
http://wccftech.com/radeon-rx-480-gaming-performance-review-leaked/?utm_source=wccftechtwitterfeed&utm_medium=wccftechtwitter&utm_campaign=Feed:+Wccftechcom+(WCCFtech.com)

Polish magazine called CD-Action shares some information about RX 480 performance in three games, Metro Light Light Redux, Witcher 3 and World of Tanks.
Średnie detale = Medium details
Bardzo wysokie detale = Very high details

According to the review, 5% overclocking causes severe artifices in 3DMark, while 7% overclock causes system to freeze. Not a word about drivers uses though.
http://videocardz.com/61512/first-radeon-rx-480-gaming-benchmarks-hit-the-web
 
Last edited:
Where exactly are you going with this? Tonga and Fiji are pretty much the same except for the size and memory controllers?
I think Fiji also improved upon the geometry/Tessellation throughput, which is better than 285/380, although how well that translates to games could be debated when one looks at peformance of a 390X (Hawaii) to FuryX at times.
It does sort of raise a technical debate whether 480 is derived more from 380x or Fury depending upons one view/context, probably part of both.
Cheers
 
Last edited:
Status
Not open for further replies.
Back
Top