Hi www
I'd like to share a few thoughts.. I originally came up with these ideas for Intel.. however, they could be applied to make the *perfect* graphics card..
Why does data need to travel around so much?.. storage to RAM.. to CPU to VRAM to..Why??
What if we had the entire system on a card? A card that could be plugged into any current PC?
Imagine with me..
Intel brings out a graphics card, the card has a 6c12t CPU.. a co-processor to compliment your Desktop's CPU. The card has an astonishing 2TB of storage which is also it's RAM/VRAM.. this is not just fantasy, I'm talking about permanence of data.. you may not know it but.. PCIE Drives are equally as fast as RAM. But I admit.. 1/3rd (15GB/s) the speed of conventional VRAM..but we must also consider that, we are reading direct from source..
It also invites the idea.. one protocol to rule them all.. PCIE. (yes yes USB too but.. not for this context).
By PCIE 5.0 the protocol will over take the mighty 45GB/s VRAM.. potentially being 128GB/s for PCIE 5.0.
Limited PCIE Lanes might have been a problem, however due to us having everything on the card.. its not a problem, the card can be designed specifically with internal PCIE lanes in mind.
For Intel this strategy yields great advantages.. a closed ecosystem (the card itself).. the Apple Mac is a great example of what you can do if you control the factors available, Apple make their operating system lighter than the hardware, and so they know how the system will perform.
The same could be true for Intel.. firstly they would have to give up those airs and graces on price to get mass adoption... however this is the perfect storm we are dreaming up so let's say they do it..
Intel make a higher end graphics processor, they include a decent CPU, they pressure chip manufacturers for PCIE 5.0, they pressure everyone to pretty much skip 4.0.. and price it around £700/$850/€850
Thinking on how they control their own card ecosystem, they would have to adhere to the most common API.. which is DirectX. There is no sense in trying to make your own API.. as we seen with Vulcan and AMD.. sure OpenGL was great for a while and even outstripped DX but that was while Microsoft was dragging its heels for Xbox 360 and DX9. The smartest way to do business would be just to accept DX as~is.. and take every line of its code and make hardware that runs it so smoothly.. it becomes more than hardware running code... it becomes a lover reciting poetry..I'm going a bit left field here with the refs.. but that's why I left blogging on OCN.. well, one of the reasons.
Think of the card like this.. DX is like it's OS, as its CPU would be specifically designed to run DX code, it *can* run normal code just like any co-processor but its best at running DX.. as it's also a storage device, it has all the data ready for gaming.
It struck me that the system I have described isn't entirely a closed system.. as DX is by Microsoft and Intel couldn't change the code to suit their hardware, but it's not all one way, as Intel could lobby Microsoft for features and code that they wished in the long run.
Intel *could* with MS's permission make a DX derivative, but that would mean convincing publishers.. and then you are back in a situation like AMD and Vulcan.
No what is clean is.. knowing DX 'chapter & small print' and making custom hardware solution solely to run that code like butter. Like loose leafs of paper are caught by the wind.. think of each one with a drawing upon it, these are the textures.
CPU & GPU & Storage; in perfect harmony.
Dava
-----------
Refs:
Speed
*http://blog.logicalincrements.com/2018/08/data-transfer-rates-bandwidth-cpu-ram-pcie-m-2-sata-usb-hdmi/
Speed
*
https://www.pcworld.com/article/3400176/pcie-40-everything-you-need-to-know-specs-compatibility.html
Tuesday, 21 April 2020
Dava's Tenets of Basic Common Sense in game design.. or too many no-no's don't make a yes-yes!
Hi WWW
Here are some Cardinal Rules of game design:
Rule 1:
Don't mess with people's saves. EVER.
Rule 2:
Don't put an invisible walls anywhere inside the sandbox. to cage us inside is fine. But once inside; it is OUR PLAYGROUND.. no longer yours. .. stop picking at it.
Rule 3:
Intuitive control.. beg, borrow.. steal if you have to from other Devs and Dev publishers.. the middle tool is so vastly unimportant, that you, Dev, sweating about it, isn't helping! If a control method is popular use that instead.
Rule 4:
Let us skip EVERYTHING.. cutscenes, cinematics, intro movies (I am aware I own a Nvidia graphics card, no need to waste my time showing me a mini movie about it).
Rule 5:
Don't put things in the game I could mistake as a 'glitch'. Seriously.
Rule 6:
If I have near completed a quest and the final part fails .. not due to me misunderstanding what to do.. but the game did not allow me to do it; ensure you have checkpoints. If the game failed and not me.. that was your problem not mine.. so to stop any arguments between player and Dev.. have checkpoints while the mission is active.. if the player quits; they must *then* and *only then*.. have to restart the whole mission if it is not completed.
Rule 7:
All bugs MUST GO. Even if your game is 20 years old.. if you are making coin off the franchise and a bug is reported. fix it!
Break any of these rules and you are a BAD DEVELOPER.. it's like being a bad parent but instead of disappointing a handful of kids.. disappointing 10's of thousands of people in one go.
@Publishers
Have a 'Gamma Team'.. who are much like Beta Testers in reverse. Gamma 'Testers'.. small team that covers all franchised games, that make coin for you as a publisher. Team of 5 to 10 people.. like a dead letter office of games publishing. You send annoyed or concerned players to them instead of your media team having to deal with an old game.
Dava
-----
I *may* come back and update this with more.. but that's enough for now.
I would like to thank Ubisoft for inspiring this list.. FLIP!! It's just basic common courtesy.. don't mess with peoples save games! FC5-DLC. Whoever thought it was a good idea over at Ubisoft.. well they probably deserve all the twitter hate the Ubi media team had to deal with and I hope they passed that along to that person.. as 'feedback/suggestions'. I am sure other players had some suggestions what that Dev could do with the game!
Here are some Cardinal Rules of game design:
Rule 1:
Don't mess with people's saves. EVER.
Rule 2:
Don't put an invisible walls anywhere inside the sandbox. to cage us inside is fine. But once inside; it is OUR PLAYGROUND.. no longer yours. .. stop picking at it.
Rule 3:
Intuitive control.. beg, borrow.. steal if you have to from other Devs and Dev publishers.. the middle tool is so vastly unimportant, that you, Dev, sweating about it, isn't helping! If a control method is popular use that instead.
Rule 4:
Let us skip EVERYTHING.. cutscenes, cinematics, intro movies (I am aware I own a Nvidia graphics card, no need to waste my time showing me a mini movie about it).
Rule 5:
Don't put things in the game I could mistake as a 'glitch'. Seriously.
Rule 6:
If I have near completed a quest and the final part fails .. not due to me misunderstanding what to do.. but the game did not allow me to do it; ensure you have checkpoints. If the game failed and not me.. that was your problem not mine.. so to stop any arguments between player and Dev.. have checkpoints while the mission is active.. if the player quits; they must *then* and *only then*.. have to restart the whole mission if it is not completed.
Rule 7:
All bugs MUST GO. Even if your game is 20 years old.. if you are making coin off the franchise and a bug is reported. fix it!
Break any of these rules and you are a BAD DEVELOPER.. it's like being a bad parent but instead of disappointing a handful of kids.. disappointing 10's of thousands of people in one go.
@Publishers
Have a 'Gamma Team'.. who are much like Beta Testers in reverse. Gamma 'Testers'.. small team that covers all franchised games, that make coin for you as a publisher. Team of 5 to 10 people.. like a dead letter office of games publishing. You send annoyed or concerned players to them instead of your media team having to deal with an old game.
Dava
-----
I *may* come back and update this with more.. but that's enough for now.
I would like to thank Ubisoft for inspiring this list.. FLIP!! It's just basic common courtesy.. don't mess with peoples save games! FC5-DLC. Whoever thought it was a good idea over at Ubisoft.. well they probably deserve all the twitter hate the Ubi media team had to deal with and I hope they passed that along to that person.. as 'feedback/suggestions'. I am sure other players had some suggestions what that Dev could do with the game!
Monday, 9 March 2020
The PCIE Lanes famine..or High Performance vs Drastically Reduced Functionality
Hi www
Today an amazing deal for a Ryzen came up. Less than half price!.. wow.. and then I thought.. 'wait don't newer CPU's have like no lanes?? (20 lanes) What does that mean for my many PCIE drives??..no, I don't have enough reason spend a fortune on a new rig, the rig I have (HP workstation) is more than capable and isn't PCIE gimped.'
Going forward we should even MORE lanes not less.. my dual X5650.. give 56 lanes EACH. Admittedly they are PCIE gen 2.. but.. 112 lanes of gen 2 vs 20 of gen3 or even 4.. nope.
This encourages me to do the unthinkable.. to recommend to people old tech.
I don't need the thing to be a server but I need the ability to have as many drives as I would like in the machine.
Let's face it.. if it's not PCIE its not the fastest possible speed. So it follows that all internal components should be PCIE to give us the fast possible performance.
Why do we need other busses in this day and age? What advantage does SATA have over PCIE?.. and even USB.. I DO think we should keep USB devices.. but couldn't the next new standard be a reworking of PCIE? For speed purposes only.. not the 75w requirement!
What can I get from on board busses that I couldn't get from a separate card of the same type?
To do as I suggest..we need 60 to 100+ lanes and 20 physical slots (at least 20 physical slots) each of the first four double spaced away from each other so no overlap. After that as tight as you like. These would be.. as I would dub them.. EXL-ATX boards and be near double the length of a ATX.. They would need fasteners at the side of the board for the lower 16 physical slots.. as the peripherals would be expected to be internal *only*..
I digress..
I don't why this isn't happening already.. but again I am bemused when case manufacturers don't work out a way to do vertical cases and insist on horizontal. The air wants to escape upwards anyway, why push it elsewhere?.. you would need a splash guard yes, but any decent fans would be able to be a splash guard anyway.
I digress .. again.. sorry..
We don't NEED all this new fancy hardware.. we have already reached a comfortable stage, however if the is .. IF THERE is to be new hardware, we should be designing it with the future in mind, not building in absurd obsolescence.
Dava
Today an amazing deal for a Ryzen came up. Less than half price!.. wow.. and then I thought.. 'wait don't newer CPU's have like no lanes?? (20 lanes) What does that mean for my many PCIE drives??..no, I don't have enough reason spend a fortune on a new rig, the rig I have (HP workstation) is more than capable and isn't PCIE gimped.'
Going forward we should even MORE lanes not less.. my dual X5650.. give 56 lanes EACH. Admittedly they are PCIE gen 2.. but.. 112 lanes of gen 2 vs 20 of gen3 or even 4.. nope.
This encourages me to do the unthinkable.. to recommend to people old tech.
I don't need the thing to be a server but I need the ability to have as many drives as I would like in the machine.
Let's face it.. if it's not PCIE its not the fastest possible speed. So it follows that all internal components should be PCIE to give us the fast possible performance.
Why do we need other busses in this day and age? What advantage does SATA have over PCIE?.. and even USB.. I DO think we should keep USB devices.. but couldn't the next new standard be a reworking of PCIE? For speed purposes only.. not the 75w requirement!
What can I get from on board busses that I couldn't get from a separate card of the same type?
To do as I suggest..we need 60 to 100+ lanes and 20 physical slots (at least 20 physical slots) each of the first four double spaced away from each other so no overlap. After that as tight as you like. These would be.. as I would dub them.. EXL-ATX boards and be near double the length of a ATX.. They would need fasteners at the side of the board for the lower 16 physical slots.. as the peripherals would be expected to be internal *only*..
I digress..
I don't why this isn't happening already.. but again I am bemused when case manufacturers don't work out a way to do vertical cases and insist on horizontal. The air wants to escape upwards anyway, why push it elsewhere?.. you would need a splash guard yes, but any decent fans would be able to be a splash guard anyway.
I digress .. again.. sorry..
We don't NEED all this new fancy hardware.. we have already reached a comfortable stage, however if the is .. IF THERE is to be new hardware, we should be designing it with the future in mind, not building in absurd obsolescence.
Dava
Subscribe to:
Posts (Atom)