Question 'Ampere'/Next-gen gaming uarch speculation thread

[Ottonomous]

How much is the Samsung 7nm EUV process expected to provide in terms of gains?
How will the RTX components be scaled/developed?
Any major architectural enhancements expected?
Will VRAM be bumped to 16/12/12 for the top three?
Will there be further fragmentation in the lineup? (Keeping turing at cheaper prices, while offering 'beefed up RTX' options at the top?)
Will the top card be capable of >4K60, at least 90?
Would Nvidia ever consider an HBM implementation in the gaming lineup?
Will Nvidia introduce new proprietary technologies again?

Sorry if imprudent/uncalled for, just interested in the forum member's thoughts.

 

[JasonLD]

Do the math and you will see the Series X is delivering 12 TF at 140-150w compared to Radeon 5700XT's 9 TF at 225w. Thats 2x the perf/watt.

That perf/watt figure can be very misleading considering Series X GPU is running right at perf/watt sweet spot while 5700XT is stepped a bit out of it in order to be competitive against its Nvidia counterparts.
Who knows if only adding 100mhz to series X GPU could jack up the power figure above 200w?

 

[AtenRa]

That perf/watt figure can be very misleading considering Series X GPU is running right at perf/watt sweet spot while 5700XT is stepped a bit out of it in order to be competitive against its Nvidia counterparts.
Who knows if only adding 100mhz to series X GPU could jack up the power figure above 200w?

I dont believe you will save 70-80W by lowering RX5700X by 100mhz

 

[JasonLD]

I dont believe you will save 70-80W by lowering RX5700X by 100mhz

Probably not by just dropping clockspeed alone, but undervolting it may. Perf/Watt just heavily favors higher CU count/lower clockspeed against lower CU count/higher clockspeed. Perf/watt improvement wouldn't look as impressive if PS5 GPU was compared against 5700XT.

 

[DeathReborn]

I dont believe you will save 70-80W by lowering RX5700X by 100mhz

70-80W probably not, but 5700XT-5700 does show you can save 40W+ (TPU Peak Gaming, 5700 180W, 5700XT 227W).

 

[AtenRa]

Perf/Watt just heavily favors higher CU count/lower clockspeed against lower CU count/higher clockspeed.

Thats not always the case, its not simple black and white. You can have way higher perf/watt from small GPUs at high clocks vs bigger GPUs with lower clocks.
Just remember, measuring perf/watt without a context is meaningless for Gaming GPUs.

For example, in the latest Techpowerup review, the ASUS TUF RX5600XT has the highest perf/watt at 4K resolution topping even the RTX2080Ti.
But that is a useless metric at 4K because RX5600XT doesnt have the power for 4K gaming.
But if you take the performance of the RX5600XT at 1080p as a context and compare its perf/watt against the same performance in that resolution then you get a meaningful result out of that perf/watt. Because that shows that RX5600XT has ~10% higher perf/watt at the same performance as the competition the RTX2060.

Perf/watt improvement wouldn't look as impressive if PS5 GPU was compared against 5700XT.

Why not??? you know something we dont ??

 

[Gideon]

5700 is a very interesting comparison. It has the same CU count and Memory bus width as PS5 @180W and yet is clocked nearly 600 mhz lower (game clock, if Cerney is to be believed). The only unknown is TDP, but I really doubt it's more than Xbox

 

[JasonLD]

Thats not always the case, its not simple black and white. You can have way higher perf/watt from small GPUs at high clocks vs bigger GPUs with lower clocks.
Just remember, measuring perf/watt without a context is meaningless for Gaming GPUs.

For example, in the latest Techpowerup review, the ASUS TUF RX5600XT has the highest perf/watt at 4K resolution topping even the RTX2080Ti.
But that is a useless metric at 4K because RX5600XT doesnt have the power for 4K gaming.
But if you take the performance of the RX5600XT at 1080p as a context and compare its perf/watt against the same performance in that resolution then you get a meaningful result out of that perf/watt. Because that shows that RX5600XT has ~10% higher perf/watt at the same performance as the competition the RTX2060.

Oh, I meant that it is easier to manipulate the perf/watt in favor toward bigger GPUs with lower clocks vs smaller GPUs with higher clocks. perf/watt can be all over the place with gaming GPUs since product segmentation is ultimately based on performance, not just perf/watt.

Why not??? you know something we dont ??

I expect PS5 and Series X will be very close in terms of TDP. PS5 seems like it went a bit out of comfort zone in order to close the performance gap against Series X.

 

[raghu78]

That perf/watt figure can be very misleading considering Series X GPU is running right at perf/watt sweet spot while 5700XT is stepped a bit out of it in order to be competitive against its Nvidia counterparts.
Who knows if only adding 100mhz to series X GPU could jack up the power figure above 200w?

Even if you take mobile GPUs like RX 5500M (22CU, 1408sp, Game clock - 1448 Mhz, 4.08TF, 85w) the Xbox Series X is 1.7x the perf/watt of mobile RDNA. So thats a fair comparison as you can consider Series X to be a mobile GPU which is optimized for sweet spot on the V/F curve.

Sony's Mark Cerny made a statement that reducing the GPU freq by 10% cuts power by 27% when asked whether the PS5 GPU will actually be able to run most of the time at 2.23 Ghz.

PlayStation 5 uncovered: the Mark Cerny tech deep dive

On March 18th, Sony finally broke cover with in-depth information on the technical make-up of PlayStation 5. Expanding …

www.eurogamer.net

Put simply, with race to idle out of the equation and both CPU and GPU fully used, the boost clock system should still see both components running near to or at peak frequency most of the time. Cerny also stresses that power consumption and clock speeds don't have a linear relationship. Dropping frequency by 10 per cent reduces power consumption by around 27 per cent. "In general, a 10 per cent power reduction is just a few per cent reduction in frequency," Cerny emphasises.

 

[badb0y]

Why exactly? I thought a big reason for Samsung was no lack of wafer supply. I actually expect the opposite this time. If AMD has a true competitor this round, then it's in Nvidia's best interest to flood the market with cards.

Corona + high demand. I think these cards will be hard to get for at least a few months.

 

[JasonLD]

Even if you take mobile GPUs like RX 5500M (22CU, 1408sp, Game clock - 1448 Mhz, 4.08TF, 85w) the Xbox Series X is 1.7x the perf/watt of mobile RDNA. So thats a fair comparison as you can consider Series X to be a mobile GPU which is optimized for sweet spot on the V/F curve.

Also, I wouldn't solely rely on TF numbers and perf/watt to estimate possible performance figure for top of the line Navi. All the TFlops numbers from RDNA2 might not translate to actual performance figures in terms of traditional rasterization performance.

 

[raghu78]

Also, I wouldn't solely rely on TF numbers and perf/watt to estimate possible performance figure for top of the line Navi. All the TFlops numbers from RDNA2 might not translate to actual performance figures in terms of traditional rasterization performance.

This is what AMD has said about RDNA2. We will see how RDNA2 vs Ampere plays out. Till then I disagree with you.

 

[DiogoDX]

We don't know the XBOX power and people are already extrapolating the Big Navi power from it. Only thing we know is that it has a 315W PSU. Actual power consumption could be as high as 250W. Just because the ONE X peaks at 175-180W doesn't mean that the Series X will be the same.

 

[ozzy702]

Can't wait for the 3000 series to come out so I can replace the temporary (sold my 2080TI) AMD 580 I'm using that keeps black screening and freezing up my system. Hopefully there's good competition and more reasonable pricing, we'll see.

Seems like every AMD GPU launch is hype hype hype and then a poor execution. Hopefully more $$$ is being thrown into that division and AMD regains some of it's former glory in the space.

 

[raghu78]

We don't know the XBOX power and people are already extrapolating the Big Navi power from it. Only thing we know is that it has a 315W PSU. Actual power consumption could be as high as 250W. Just because the ONE X peaks at 175-180W doesn't mean that the Series X will be the same.

First if you did see the Series X teardown and did some basic analysis you would come up with similar conclusion. Dual board design. Dual PSU rails. +12v 21.5A, +12v 5A. Mainboard with SoC, 16 GB GDDR6 memory, onboard 1TB NVMe, external NVMe connector and HDMI 2.1 connects to the +12v 21.5A rail. The daughter I/O board connects to the other +12v 5A rail.

90% efficiency of PSU and assuming a load of 90% (worst case) gives 206w.
8C/16T at 3.66 Ghz draws 54w from Renoir measurements.

“核弹”降临——R7 4800H简单测试

自2017年发布以来，AMD的Ryzen锐龙系列CPU一路高歌，着实给了英特尔不小的压力。特别是ZEN 2架构的3900X、3950X、3960X、3970X、3990X“五虎将”，以强大的性能和极具性价比的价格让不少人直呼“AMD YES！” 然而…

zhuanlan.zhihu.com

4w onboard 1 TB NVMe + 4w for external NVMe + 2w HDMI.

Roughly 150w for GPU + 16 GB GDDR6 (asymmetric setup with 10GB at 560 GB/s and 6 GB at 336 GB/s). The GPU on its own is drawing roughly 120w.

 

[MrTeal]

First if you did see the Series X teardown and did some basic analysis you would come up with similar conclusion. Dual board design. Dual PSU rails. +12v 21.5A, +12v 5A. Mainboard with SoC, 16 GB GDDR6 memory, onboard 1TB NVMe, external NVMe connector and HDMI 2.1 connects to the +12v 21.5A rail. The daughter I/O board connects to the other +12v 5A rail.

90% efficiency of PSU and assuming a load of 90% (worst case) gives 206w.
8C/16T at 3.66 Ghz draws 54w from Renoir measurements.

“核弹”降临——R7 4800H简单测试

自2017年发布以来，AMD的Ryzen锐龙系列CPU一路高歌，着实给了英特尔不小的压力。特别是ZEN 2架构的3900X、3950X、3960X、3970X、3990X“五虎将”，以强大的性能和极具性价比的价格让不少人直呼“AMD YES！” 然而…

zhuanlan.zhihu.com

4w onboard 1 TB NVMe + 4w for external NVMe + 2w HDMI.

Roughly 150w for GPU + 16 GB GDDR6 (asymmetric setup with 10GB at 560 GB/s and 6 GB at 336 GB/s). The GPU on its own is drawing roughly 120w.

If the output of the PSU is 21.5A on the rail it doesn't matter what the PSU efficiency is, there's a rated 258W available to the motherboard.

 

[Glo.]

Here we go again... After all the hype, NV killer guerilla marketing, Glo's raving (I do miss RussianSensation) Nvidia are going to announce and launch their cards before AMD has even gotten a finalized design down (for all we know).

Execution machines, those guys. It will be even sadder if they actually are 100% on the Samsung 8/10nm train for their consumer parts. Working with a brand new partner for large dies and still getting the stack out the door before we even hear a peep from AMD.

If you have inferior product it is very important for you to release it faster than your competition.

 

[Konan]

If you have inferior product it is very important for you to release it faster than your competition.

First movers attract eager buyers, secondary movers usually go when demand begins to grow and they have an inferior product trying to build off the success of the first.
In this case, with Nvidia we're already seeing smart marketing with the 21 days for 21 years ago approach. A 3 week build up to an announcement followed by a launch has all been planned out. Product segregation over 30-60-90 day plans. No fear. The market is most anxious for a product right now and that is when you launch. Funny it is usually this time that happens, back to school leads to thanksgiving leads to black Friday etc all way till holiday season. This year we have consoles to deal with so better to get out in front of all that.

Bet you that AMD wishes they were launching right now. But no, their board design was late, drivers are late, launch is late and no AIB partners. That smells of poor planning and execution. That said the advantage they have is pricing but the more it is left late the more damage is done.

 

[Thala]

If you have inferior product it is very important for you to release it faster than your competition.

I guess thats too bad for AMD then, they should have been faster...

 

[raghu78]

If the output of the PSU is 21.5A on the rail it doesn't matter what the PSU efficiency is, there's a rated 258W available to the motherboard.

Even if we go with that 258w number the max power draw on the Series X SoC is expected to be around 70-75% of the max available power draw. These PSUs are never loaded upto their max rated limit to account for degradation in rated wattage over expected lifespan. These consoles are expected to last 7-10 years.

The Xbox One X Review: Putting A Spotlight On Gaming

www.anandtech.com

Xbox One X drew a max of 180w (172w in Gears of War) with a 245w rated PSU. I would say 220w is a worst case scenario for the Series X mainboard. More realistically 200-210w.

 

[FaaR]

Why not??? you know something we dont ??

Sony is clocking the hell out of their GPU, right up to the bleeding edge of reliability, according to Cerny. That's going to bump power draw significantly, I'm really curious how much it'll end up being. PS5 is a very large console from what we've seen, could be they needed to engineer a big, sophisticated cooling system to keep the noise level down while still dissipating a significant wattage. Sort of like with launch PS3 units.

90% efficiency of PSU and assuming a load of 90% (worst case) gives 206w.

PSU (in)efficiency goes on top of rated output. It doesn't subtract from it...

 

[DiogoDX]

If you have inferior product it is very important for you to release it faster than your competition.

1080Ti that lauched 5 months before Vega says hello.

 

[Glo.]

Bet you that AMD wishes they were launching right now. But no, their board design was late, drivers are late, launch is late and no AIB partners. That smells of poor planning and execution. That said the advantage they have is pricing but the more it is left late the more damage is done.

Are you sure about that?

Then why Nvidia is RUSHING the launch of next gen GPUs? There is ONE reason why we see those GPUs so early.

Ask yourself, why is Nvidia rushing this launch, if AMD has inferior products? Do they have inferior products?

 

[JujuFish]

Are you sure about that?

Then why Nvidia is RUSHING the launch of next gen GPUs? There is ONE reason why we see those GPUs so early.

Ask yourself, why is Nvidia rushing this launch, if AMD has inferior products? Do they have inferior products?

What makes you think they're rushing the launch?

 

[BenSkywalker]

5700xt came out how long ago?

Rushing? NVidia is very late.

 

[Konan]

Are you sure about that?

Then why Nvidia is RUSHING the launch of next gen GPUs? There is ONE reason why we see those GPUs so early.

Ask yourself, why is Nvidia rushing this launch, if AMD has inferior products? Do they have inferior products?

I don’t think that they are worried about AMD as much as you think they are. You’re making some massive assumptions to fit your conscious narrative.

I believe that they will not have inferior products. I think that they will be competitive and have a great looking stack. It will be progress over their last iteration, but it won’t be perfect and that’s ok.

I think that they will have inferior software. I also think that if they could choose to launch with AIB partners they would.

Both companies are dealing with global issues across their lines of business. The sooner the better.

I don’t believe for a second that it is coincidence with the 21 days 21st year etc. that stuff is planned out months prior (once the product is cooked)