AMD Acquired HiAlgo. Bye bye stuttering.

[Prysin]

Source: http://www.amd.com/en-us/press-releases/Pages/amd-acquires-software-2016jun29.aspx

 

Quote

AMD (NASDAQ: AMD) today announced the acquisition of software company HiAlgo Inc., a developer of unique PC gaming technologies designed to help Radeon™ RX Series GPUs transform gaming experience, increase GPU efficiency and improve the overall consistency of gaming experiences. The acquisition lays the groundwork for future gaming innovation in Radeon Software that will benefit owners of Radeon™ RX Series GPUs.

“Software is an integral part of advancing the science of graphics, enabling us to best harness the silicon of the GPU to 
maximize performance and deliver outstanding experiences in games and applications,” said Raja Koduri, senior vice president and chief architect, Radeon Technologies Group, AMD. “HiAlgo embodies our spirit of passion, persistence and play by delivering a number of creative approaches to software that improve gamers’ experiences, and helps future-proof¹ the GPU.”

 

 

Now what is HiAlgo? It is a company, most known for a mod for Skyrim, that smoothed out frametimes to get rid of stuttering. They did so by applying a wrapper over the DX9 API.

 

Now, Raja Koduri, the VP of Radeon Technologies Group explains this pretty darn well here:

 

 

incase link didnt work: he starts explaining at 43:45

 

 

 

Personally, i think this is DAMN nice. If AMD is able to improve frame latencies with this,  then even if it doesnt match up in FPS, if the frame delivery is 100% perfect, aka NO STUTTERING, then that is perhaps even more important then a few FPS less (assuming you are over 60 FPS avg).

One of the big selling points of Maxwell vs say the Fury and Nano, was that the 980 was remarkebly stable in its frame delivery, while most AMD cards, due to driver overhead and whatnot, always have a little bit stutter.

 

Scott Wasson, another AMD employee and former Editor in Chief of The Tech Report (The review site that alongside Nvidia Pioneered frame latency testing) posted this image on Twitter the other day:
https://mobile.twitter.com/scottwasson/status/748162402265403392/photo/1

 

If this graph is true, that DAMN that is good. The 970 is showing REALLY bad microstuttering in that graph. 

 

[Shoob]

Great news. I'm really looking forward to the results of this acquisition. If the latency is greatly reduced even in multi-GPU configurations, I'll be very impressed.

[Notional]

Sounds interesting, but it does mean Freesync is only useful for tearing then.

 

To be fair, the 970 has massive stutter, because of the weak 3.5+05GB vram, compared to the 8GB vram. Games simply use more than 4GB today with highest texture resolution, so it will stutter heavily as it's reading system RAM.

[Citadelen]

I'm looking forward for the fruits of this purchase, I can see it having a great effect on VR as well.

[sof006]

YES! AMD is doing everything right at the moment, please keep it this way AMD!

[MrDynamicMan]

Huh. I'd be interested how heavily they market this tech, and what kind of impact it will have. 

[mikegray]

HiAlgo doesn't have a particularly good reputation in the world of Skyrim modding. It works by lowering resolution when you're in motion (or panning the screen), then raising it again afterward, and a lot of people seem to feel that it really messes up your pretties. Also, it tends to have issues with ENBs - HiAlgo and ENB development seem to take turns breaking each other.

 

I recently tested it on a very heavily modded Skyrim installation (including an ENB) on my current, fairly high end rig (Skylake i7, 2x 290x in CF), and it DID seem to smooth framerate in some places, but it also caused some weird visual effects. The strengths didn't outweigh the weaknesses.

 

OTOH, a couple years back (pre ENB), I had some good experiences with it, like a week I spent at my mother-in-law's place: thanks to HiAlgo I has able to run my modded Skyrim install on a weak laptop that would otherwise NOT have been usable. And I have to say that I, personally, hardly noticed the graphical degradation - and REALLY noticed the steady FPS. 

 

I'll be particularly interested to see how it does when implemented through drivers instead as an add-on .dll.

[Prysin]

29 minutes ago, mikegray said:

HiAlgo doesn't have a particularly good reputation in the world of Skyrim modding. It works by lowering resolution when you're in motion (or panning the screen), then raising it again afterward, and a lot of people seem to feel that it really messes up your pretties. Also, it tends to have issues with ENBs - HiAlgo and ENB development seem to take turns breaking each other.

 

I recently tested it on a very heavily modded Skyrim installation (including an ENB) on my current, fairly high end rig (Skylake i7, 2x 290x in CF), and it DID seem to smooth framerate in some places, but it also caused some weird visual effects. The strengths didn't outweigh the weaknesses.

 

OTOH, a couple years back (pre ENB), I had some good experiences with it, like a week I spent at my mother-in-law's place: thanks to HiAlgo I has able to run my modded Skyrim install on a weak laptop that would otherwise NOT have been usable. And I have to say that I, personally, hardly noticed the graphical degradation - and REALLY noticed the steady FPS. 

 

I'll be particularly interested to see how it does when implemented through drivers instead as an add-on .dll.

ENBs and HiAlgo both use wrappers. The reason they break eachother is that the wrapper is a software sitting outside the API and driver. It can only apply things AFTER the fact.

 

What AMD/RTG is intending to do, is implement the tech at a driver level. This would offer MUCH more granular control then a simple overlay would.

[DocSwag]

Cool... Perhaps in the future this could make multi GPU setups worth it again....

[Thony]

3 hours ago, Notional said:

Sounds interesting, but it does mean Freesync is only useful for tearing then.

 

To be fair, the 970 has massive stutter, because of the weak 3.5+05GB vram, compared to the 8GB vram. Games simply use more than 4GB today with highest texture resolution, so it will stutter heavily as it's reading system RAM.

My gtx 760 stutters before it reaches its vram limit. So I dont agree with your theory.

[themaniac]

does AMD/RTG really have the money to acquire, even if their really small

[Briggsy]

rise of the tomb raider is a horrible implementation of dx12. I'm surprised the graph looks coherent with the 970 at all.

[Notional]

1 hour ago, Thony said:

My gtx 760 stutters before it reaches its vram limit. So I dont agree with your theory.

I didn't state VRAM limitation is the only thing that causes stutter, but running out of VRAM will ALWAYS cause stutter, due to huge latencies from reading from system RAM instead of VRAM.

Case and point:

50 minutes ago, Briggsy said:

rise of the tomb raider is a horrible implementation of dx12. I'm surprised the graph looks coherent with the 970 at all.

Indeed, however DX12 in RotTR does seem to raise the minimum fps, so you should generally get a more consistent framerate.

[spartaman64]

1 hour ago, themaniac said:

does AMD/RTG really have the money to acquire, even if their really small

amd is still a pretty big company

[themaniac]

1 minute ago, spartaman64 said:

amd is still a pretty big company

that is true but afaik they are still losing money so why are they buying companies even if they are small if they cant even make a profit

[spartaman64]

4 minutes ago, themaniac said:

that is true but afaik they are still losing money so why are they buying companies even if they are small if they cant even make a profit

if amd doesnt innovate then there is no way for them to make back market share. obviously that company has something that amd sees as valuable and if amd dont do things like this they will be left behind

[Misanthrope]

4 hours ago, sof006 said:

YES! AMD is doing everything right at the moment, please keep it this way AMD!

Everything except complying with the motherboard pci-e specifications:

 

https://www.reddit.com/r/Amd/comments/4qfwd4/rx480_fails_pcie_specification/

 

So yeah, not everything. Not by a long shot.

[AlTech]

7 minutes ago, themaniac said:

that is true but afaik they are still losing money so why are they buying companies even if they are small if they cant even make a profit

They're expecting higher sales soon.

[raphidy]

This is a good move. They would develop, incorporate this tech in drivers when dx12, vulkan are maturing. We'll see some results maybe in a year?

[don_svetlio]

Even though the 480 ended up between the 970 and 980 this might make CF 480s a viable option to compete with 1070s.

[Colonel_Gerdauf]

8 minutes ago, don_svetlio said:

Even though the 480 ended up between the 970 and 980 this might make CF 480s a viable option to compete with 1070s.

I really want to know what you have been reading, but most reviews (video or not) report it performing slightly worse than the 970 in quite a number of theoretical and real-world scenarios.

[don_svetlio]

5 minutes ago, Colonel_Gerdauf said:

I really want to know what you have been reading, but most reviews (video or not) report it performing slightly worse than the 970 in quite a number of theoretical and real-world scenarios.

 

[Bouzoo]

53 minutes ago, Colonel_Gerdauf said:

I really want to know what you have been reading, but most reviews (video or not) report it performing slightly worse than the 970 in quite a number of theoretical and real-world scenarios.

But then again it's even faster than a 980 in "a number of theoretical and real-world scenarios".

Here is the thing, as always, the card has day 1 drivers and the performance will increase over time, same as Fury X, 330X, NV cards and every card ever. I could say "expect 20% better performance in future", but more realistically I'm hoping for 10-15%. 

And even with current performance and all combined, it performs better than the 970 according to TechPowerUp and Guru3D. Those are the ones that take everything into account that I can think off and that do proper testing. I'm not taking LTT and such mediocre reviews into account. You could say I'm cherry picking it, that some sites like Tom's HW etc. show it being slower than 970 in many scenarios and that's absolutely correct, so do the sites that I've mentioned, but then again, as I've said, it's on pair and even faster than a 980 in some tests and you can see that for yourself. All in all it's faster than a 970, but it's not that consistent, and not by a huge margin. 

And I can't say that it's bad performing because it isn't... to a degree. The reference cooler is hands down horrible, and I've been saying that for few weeks now. Other things like the power delivery, OCing and temps are being improved with AIB cards that are showing an 8 pin connector (like from Sapphire, the fact that it draws more than advertised is till there, but the latest drivers have lowered, for instance, idle draw from 16 to 11W, so I'll bet something can be done there, even if by a small margin), OCs from 1480-1600 MHz, according to HardOCP chief editor, and we've seen 63C on cards that run at 1350, with moded cards, not even proper AIB ones. 

So it definitely should not a bad card for the price, which is stupid at the moment and way higher in most parts than it should be. I'll bet that AIB cards that will run at 1500 MHz with 8 pins will be great. And who knows, those maybe will be on pair with the 980, seeing how Guru3D got 1380MHz from their card and got ~10% better performance. I personally don't care about power delivery, if it's 150W or 200W. 

[don_svetlio]

5 minutes ago, Bouzoo said:

But then again it's even faster than a 980 in "quite a number of theoretical and real-world scenarios".

Here is the thing, as always, the card has day 1 drivers and the performance will increase over time, same as Fury X, 330X, NV cards and every card ever. I could say "expect 20% better performance in future", but more realistically I'm hoping for 10-15%. 

And even with current performance and all combined, it performs better than the 970 according to TechPowerUp and Guru3D. Those are the ones that take everything into account that I can think off and that do proper testing. I'm not taking LTT and such mediocre reviews into account. You could say I'm cherry picking it, that some sites like Tom's HW etc. show it being slower than 970 in many scenarios and that's absolutely correct, so do the sites that I've mentioned, but then again, as I've said, it's on pair and even faster than a 980 in some tests and you can see that for yourself. All in all it's faster than a 970, but it's not that consistent.

And I can't say that it's bad performing because it isn't... to a degree. The reference cooler is hands down horrible, and I've been saying that for few weeks now. Other things like the power delivery, OCing and temps are being improved with AIB cards that are showing an 8 pin connector (like from Sapphire, the fact is that it draws more than advertised is till there, but the latest drivers have lowered, for instance, idle draw from 16 to 11W, so I'll bet something can be done there, even if by a small margin), OCs from 1480-1600 MHz, according to HardOCP chief editor, and we've seen 63C on cards that run at 1350, with moded cards, not even proper AIB ones. 

So it's definitely should not a bad card for the price, which is stupid at the moment and way higher in most parts than it should be. I'll bet that AIB cards that will run at 1500 MHz with 8 pins will be great. And who knows, those, maybe will be on pair with the 980, seeing how Guru3D got 1380MHz from their card and got ~10% better performance. I personally don't care about power delivery, if it's 150W or 200W. 

Exactly - DigitalFoundry and Science Studio show it beating the 970 and in SS' case it ran at 78*C at reasonable fan levels. For some reason engineering samples are wildly inconsistent.

[Bouzoo]

1 minute ago, don_svetlio said:

Exactly - DigitalFoundry and Science Studio show it beating the 970 and in SS' case it ran at 78*C at reasonable fan levels. For some reason engineering samples are wildly inconsistent.

I've seen that it runs in load at 80+ C and I think that's more than realistic, since that is wildly consistent to all that I've seen, from sites like Guru3D, TPUP and Tom's HW. And the noise is... well nothing worse than a 290X or 980Ti from their tests.