AdoredTV attempts to justify his position on Pascal

[fourletterfame]

Basically the dude makes enjoyable videos for nerds, and in a previous video that came after the announcement of the 1080 and 1070, he noted that the new architecture hadn't changed much in the node shrink, it was just a more densely packed maxwell with higher clocks.

Apparently he got a lot of flack for it and decided to test it.

The results are interesting and in line with the speculation.

It doesn't mean much, it was just an interesting video for anyone who geeks out about this stuff.

 

[basco]

thx+ i love his accent

 

[rtwjunkie]

Thanks! It seems he validated his position. It's nice that GPU-Z got such a big plug in there, too!

 

[R-T-B]

Honestly, "if it ain't broke don't fix it" comes to mind.

Still interesting, of course.

 

[newtekie1]

I like how he just throws in the memory issue as quickly as possible and then says its "basically nothing". The 980Ti has a good 5% memory bandwidth advantage, so hardly nothing.

 

[fourletterfame]

Honestly, "if it ain't broke don't fix it" comes to mind.

Still interesting, of course.

I agree completely, and I suspect most people would as well. It was a sound strategy, but there are so many "experts" commenting on youtube videos I can't blame him for putting the brakes to all the speculation and testing it. Interestingly I suspected the same things he did when I found a record breaking gtx 980 ln2 overclock at 2ghz that performed pretty closely to the 1080, that was enough for me to suggest most if not all of their gains were from clockspeed and not iterative changes to the architecture.

 

[Adored]

I like how he just throws in the memory issue as quickly as possible and then says its "basically nothing". The 980Ti has a good 5% memory bandwidth advantage, so hardly nothing.

I like how you ignore that by downclocking the 1080 rather than overclocking the 980 Ti like most others did, bandwidth issues would be far less of an issue.

I especially liked how you ignored Pascal's superior compression technology, which Nvidia themselves claim gives an 11%+ benefit to bandwidth over Maxwell. Also 2GB extra VRAM.

By downplaying the memory issue I basically did a favour to the 1080. This stuff contributed <1% difference, same with ROPs and all that.

 

[P4-630]

I like how you ignore that by downclocking the 1080 rather than overclocking the 980 Ti like most others did, bandwidth issues would be far less of an issue.

I especially liked how you ignored Pascal's superior compression technology, which Nvidia themselves claim gives an 11%+ benefit to bandwidth over Maxwell. Also 2GB extra VRAM.

By downplaying the memory issue I basically did a favour to the 1080. This stuff contributed <1% difference, same with ROPs and all that.

Aaand....You signed up to just say that?

 

[Adored]

Aaand....You signed up to just say that?

Yes, correcting bad assumptions on tech is one of my favourite hobbies, so why not?

 

[RejZoR]

Hey Adored For some reason I read the above post with your accent. This is contagious... XD

Btw, GTX 1080 also uses GDDR5X at a lot higher clock than the one found on GTX 980Ti...

 

[Adored]

Hey Adored For some reason I read the above post with your accent. This is contagious... XD

Btw, GTX 1080 also uses GDDR5X at a lot higher clock than the one found on GTX 980Ti...

Cheers. The 1080 I had also had a 10MHz increase in memory clock over the base too, so another small gain there. It's too difficult to say exactly what difference this stuff makes but given the final result of 0.6% difference in FPS and Anandtech's recent article on the shaders, it seems not a lot. Avoiding bandwidth issues was the main reason for underclocking the 1080 though, that was why I chose to go down to the 1200MHz range.

 

[rtwjunkie]

Yes, correcting bad assumptions on tech is one of my favourite hobbies, so why not?

Thanks! Glad to have you. I thought you did an excellent breakdown of it all.

 

[the54thvoid]

@Adored - wtf is that accent? I'm from East Kilbride and now live in Glasgow - your accent sounds...'forcibly poshed up'....? Is that for the internet audience? I thought I had a soft posh accent until I heard that.. no offence.

@Finners can back me up.

 

[Adored]

Thanks! Glad to have you. I thought you did an excellent breakdown of it all.

Was a tough video tbh, it's amazing how almost nothing works as you expect it to when doing these things (the shadowplay being worse on the 1080 is an example).

@Adored - wtf is that accent? I'm from East Kilbride and now live in Glasgow - your accent sounds...'forcibly poshed up'....? Is that for the internet audience? I thought I had a soft posh accent until I heard that.. no offence.

@Finners can back me up.

Yeah it's a combination of things, including the Swedish gf and living here but obviously my actual accent is a bit different lol. Like another language talking to friends and family from Ayrshire, the gf can't understand us after we've had a few drinks.

 

[D007]

Was a tough video tbh, it's amazing how almost nothing works as you expect it to when doing these things (the shadowplay being worse on the 1080 is an example).



Yeah it's a combination of things, including the Swedish gf and living here but obviously my actual accent is a bit different lol. Like another language talking to friends and family from Ayrshire, the gf can't understand us after we've had a few drinks.

Shadowplay is worse? Ahh ffs, I love using that.. Well, I hope they fix it..

 

[Finners]

@Adored - wtf is that accent? I'm from East Kilbride and now live in Glasgow - your accent sounds...'forcibly poshed up'....? Is that for the internet audience? I thought I had a soft posh accent until I heard that.. no offence.

@Finners can back me up.

Haha, you have lost the title of the poshest Scotsmen I've ever heard.

Not offence Adored

 

[Adored]

Shadowplay is worse? Ahh ffs, I love using that.. Well, I hope they fix it..

It's still mostly better than trying to get video done using AMD's stuff. They really need to fix Raptr or make their own prog. I can't really understand why a company that can create Mantle can't write a simple prog to make full use of their video encoding, especially with the new hardware in Polaris.

 

[cadaveca]

It's still mostly better than trying to get video done using AMD's stuff. They really need to fix Raptr or make their own prog. I can't really understand why a company that can create Mantle can't write a simple prog to make full use of their video encoding, especially with the new hardware in Polaris.

They don't have the staff that NVidia does, making the reason why quite obvious. I mean, after all, they have similar numbers of staff, but NVidia doesn't have a CPU on the market, so that staff dedicated to CPU production at AMD is not staff that can help with GPU things in a big way, whereas NVidia has nothing but GPU staff. Although the two companies are competitors in some ways, in many ways they are very different from each other, and as such will never reach parity with each other. I think AMD is smart to focus on lesser-performing cards right now, that are much more affordable for the end user, since these cards tend to sell in higher numbers anyway. All AMD has to do here is sell every GPU they make, and they will do well. There is no real need for them to expand their market. Meanwhile NVidia has much faster products, and a much wider user base ($250-$1500+).

It was quite interesting to see both NVidia and AMD trade blows in the performance arena when both were using TSMC to make their chips. Now they are using different silicon, and could not be more different from each other.

 

[Adored]

Nvidia has a lot of products outside of consumer GPU also remember and I think both companies have a fair amount of convergence between their CPU and GPU departments these days. I think AMD's biggest problem before Su took over is that they had their finger in too many pies, and a lot of them were a bit off.

Tbh though it's their ability to market their own stuff that is bad. The number of people telling me that they'd love to go AMD but "need ShadowPlay" is pretty mind-boggling.

 

[cadaveca]

Tbh though it's their ability to market their own stuff that is bad. The number of people telling me that they'd love to go AMD but "need ShadowPlay" is pretty mind-boggling.

I agree. Yet I was rather pleased to hear that FRAPS is getting an update... its like Beepa noticed that need there and is trying to capitalize on it.

I talk to a lot of people that stream or make gameplay videos, and ShadowPlay makes this very easy, while there area quite a few other alternatives that end up being much more complicated to use is comparison. That's what FRAPS always offered - ease of use.

 

[Adored]

Yeah I heard Fraps is getting a DX12 update. My personal favourite is OBS but it's a hassle to set up and always ran pretty badly on AMD cards. I had quite a few issues with ShadowPlay though actually - Witcher 3 just wouldn't run with it switched on. I normally use Action but it can be quite random as well. Looking at getting a capture card for this stuff and I wouldn't mind using FCAT but the cost of the card is really prohibitive.

 

[cadaveca]

Looking at getting a capture card for this stuff and I wouldn't mind using FCAT but the cost of the card is really prohibitive.

It is, but you know, I complained about rendering latency for years before the whole frame-pacing thing came to light, and people told me I was mistaken, or that I was crazy. That just pushed me even more since I knew I wasn't wrong, and that those that doubted my claims or outright refuted them were the wrong ones. Yeah, I was driven crazy by poor rendering quality on graphics solutions that I spent tonnes of money on and the tools that would let me show the truth of the situation were at the time not even available. It took half a decade and NVidia to develop proper tools before undeniable evidence was possible. So the cost of FCAT systems seems minor to me. NVidia spent more money in the right way on that than I ever could, even if I had endless cashflow. I think the cost is high for all the wrong reasons.


So, in comparing GPU designs, you do need to look at more than just frame count in order to see the truth about what's going on. You need to analyze frame latency, color quality, power consumption, etc, etc, etc. Features offered by drivers need some thought too...

That's not to say I disagree with your video's synopsis, but I do kind of feel that maybe we're overlooking something here when it comes to comparing Pascal and Kepler.

There was even a time when certain people were touting the idea that gamers didn't mind artifacting in games...

 

[newtekie1]

I like how you ignore that by downclocking the 1080 rather than overclocking the 980 Ti like most others did, bandwidth issues would be far less of an issue.

I especially liked how you ignored Pascal's superior compression technology, which Nvidia themselves claim gives an 11%+ benefit to bandwidth over Maxwell. Also 2GB extra VRAM.

By downplaying the memory issue I basically did a favour to the 1080. This stuff contributed <1% difference, same with ROPs and all that.

I don't know why you think changing the core clock would affect the memory bandwidth at all. But whatever...

And you talk out of one side of your mouth saying Pascal is just a clocked higher Maxwell, then out of the other saying Pascal has significantly better memory compression. I'm sorry, but that is an architectural improvement!

You can make the argument that image processing wise Pascal isn't really any better than Maxwell. Yep, it turns out if you make them computer wise at the same TFLOP level, they perform at the same compute level. Big surprise there! But they aren't the same. There have been improvements, in the memory side. They didn't just take Maxwell, shrink it down, and shit it out the door as you claim.

 

[Adored]

I don't know why you think changing the core clock would affect the memory bandwidth at all. But whatever...

And you talk out of one side of your mouth saying Pascal is just a clocked higher Maxwell, then out of the other saying Pascal has significantly better memory compression. I'm sorry, but that is an architectural improvement!

Why do I get the feeling that you'd be one of the first people claiming that AMD didn't do anything except rebrand for 4 years, even though they added more tessellation every GCN revision, better memory compression and finally HBM?

Downclocking the GPUs reduces their requirement for more bandwidth, that one shouldn't be hard to figure out. It's why weaker GPUs have less bandwidth to start with, they don't need it.

You can make the argument that image processing wise Pascal isn't really any better than Maxwell. Yep, it turns out if you make them computer wise at the same TFLOP level, they perform at the same compute level. Big surprise there! But they aren't the same. There have been improvements, in the memory side. They didn't just take Maxwell, shrink it down, and shit it out the door as you claim.

How can you talk about big memory improvements in Pascal after suggesting that my dismissal of bandwidth was favouring Maxwell?

Which one is it...was I favouring Maxwell by downplaying the effect of the memory subsystem or was I favouring Pascal by downplaying the effect of the memory subsystem?

 

[GreiverBlade]

sooo then the 1070 is a overclocked 970 ? i see your point but ... a 1070 that has Titan X performances in general and GDDR5 all the same as the 970 (sans the 3.5+0.5 and with 1gbps more) even on a refined 16nm FinFet node would not give that even with a substantial OC ...

i fail to see Pascal as a "die shrink" of Maxwell nonetheless ... as even a Skylake to Haswell can have the same performances if you downclock it ... i can't recall (nor i want to review the vids right now) did you test both at exact same clock and get that result or did you just lowered the 1080 clocks until it reached the 980Ti level (btw that's wrong to compare the 980Ti to the 1080 as the predecessor is the 980 even if they have better performances, well the 1070 is ... "on par or slightly above" the 980Ti nonetheless )