AMD Unveils Vega GPU Architecture With 512 Terabytes of Memory Address Space (hothardware.com)

MojoKid writes: AMD lifted the veil on its next generation GPU architecture, codenamed Vega, this morning. One of the underlying forces behind Vega's design is that conventional GPU architectures have not been scaling well for diverse data types. Gaming and graphics workloads have shown steady progress, but today's GPUs are used for much more than just graphics. In addition, the compute capability of GPUs may have been increasing at a good pace, but memory capacity has not kept up. Vega aims to improve both compute performance and addressable memory capacity, however, through some new technologies not available on any previous-gen architecture. First, is that Vega has the most scalable GPU memory architecture built to date with 512TB of address space. It also has a new geometry pipeline tuned for more performance and better efficiency with over 2X peak throughput per clock, a new Compute Unit design, and a revamped pixel engine. The pixel engine features a new draw stream binning rasterizer (DSBR), which reportedly improves performance and saves power. All told, Vega should offer significant improvements in terms of performance and efficiency when products based on the architecture begin shipping in a few months.

Comments

512 should be enough for anybody....

[Anonymous Coward]

Wait, TERABYTES? Holy..

Re:512 should be enough for anybody....

[fbobraga]

WTF ho made AMD PR is smoking?!?!

512TB of address space means nothing

[Lisandro]

Most high end GPU cards available have 8Gb, a large number of budget versions settle for 4Gb, and only a few offer 16Gb. Marketing this as a stand out point is iffy.

Re:512TB of address space means nothing

[iXam]

GB, not Gb.

Re:512TB of address space means nothing

[Lisandro]

Oh, I know. That is still only 64Tb...

Re:512TB of address space means nothing

[BigBuckHunter]

Most high end GPU cards available have 8Gb, a large number of budget versions settle for 4Gb, and only a few offer 16Gb. Marketing this as a stand out point is iffy.

What you will find is that most cards have only a fraction of their RAM as addressable, so a 16GB card either 4 or 8 gigs addressable. The increase to 512GB is a godsend to AI researchers and other fields with large datasets.

Re:512TB of address space means nothing

[Anonymous Coward]

Lisandro you COULD RTFA, you know? It's even an effing meme around here.

The HBCC gives the GPU access to 512TB (half a petabyte) of virtual address space and gives the GPU fine-grained control, for adaptable and programmable data movement. Often, more memory is allocated for a particular workload than is necessary; the HBCC will allow the GPU to better manage disparities like this for more efficient use of memory. The huge address space will also allow the GPU to better handle datasets that exceed the size of the GPU’s local cache. AMD showed a dataset being rendered in real-time on Vega using its ProRender technology, consisting of hundreds of gigabytes of data. Each frame with this dataset takes hours to render on a CPU, but Vega handled it in real-time.

Re:512TB of address space means nothing

[Lisandro]

512Gb GPUs? Doubt you'll see that anytime soon. The very largest memory on any commercial GPU is AMD's own FirePro W9100, at 32gb. It is more of a cost issue than a limitation on addresable space; 64Gb is right around the corner though.

Re:512TB of address space means nothing

[Lisandro]

Wow, marketing lingo is eerily efficient on you.

Re:512TB of address space means nothing

[Anonymous Coward]

yea thats 32 gig's in banks of 4 bankswitching

even if you do not use it all just the wasted time and overhead saved is quite a bit in serious applications (and maybe even a few fps in gamez!)

Re:512TB of address space means nothing

[K. S. Kyosuke]

Except for AMD cards with massive storage for (semi)fixed datasets... Kind of the reason for this, in fact.

Re:512TB of address space means nothing

[Anonymous Coward]

You are assuming the target use case is games and graphics - which it clearly is not. GPUs are used in a lot supercomputing type calculations including fluid dynamics and particle physics. But even as a game engine, I can see how I may want to have one (or few) giant cards calculating the entire scene for a multiplayer game in the cloud and then sending each user his personal PoV rendition.

Re:512TB of address space means nothing

[Lisandro]

But this is not new at all. IIRC Nvidia's CUDA 5 already gives you 49 bits of unified address space. Don't really know the addressing limitations on previous AMD architectures, but I doubt it was substantially lower.

Realistically, large address spaces when you can only practically fill 0.05-0.1% means little for performance. I don't want to attack AMD with this, who usually manufacture really good GPU hardware, but this sounds like a marketing gimmick and nothing more. I particularly enjoyed the "hours to real-time" comparison... against a CPU.

Re:512TB of address space means nothing

[Anonymous Coward]

professional GPU cards normally ship with just 16Gb.

Re:512TB of address space means nothing

[Anonymous Coward]

I half-wonder if he meant "Vega's huge memory space lets you overcommit your memory, just like sane, rational OSs do.".

The other half of me wonders how many zeros are on his paycheck.

Re:512TB of address space means nothing

[dfghjk]

"The increase to 512GB is a godsend to AI researchers and other fields with large datasets."

References?

While there's no doubt that there is SOME application that could use that amount of physical addressability, it would seem extraordinarily unlikely that a single GPU would be sufficient for such an application and, even so, it's absurd to refer to such a niche as a "godsend". Meaningless hyperbole, most likely without any supporting insight.

Re:512TB of address space means nothing

[sexconker]

What you will find is that most cards have only a fraction of their RAM as addressable, so a 16GB card either 4 or 8 gigs addressable. The increase to 512GB is a godsend to AI researchers and other fields with large datasets.

Nope.

1: The GPU addresses the whole damn pool.

2: We're talking about 512 TB, not GB.

3: They're not planning to release a card with 512 TB of RAM, but they are releasing professional cards with lots of RAM (8 GB, 16 GB, or more) AND onboard connections for flash storage (SSDs). Vega will likely continue and extend this. By having a huge address space, you simply have the ability to keep the entire dataset in your cache on the card. The memory controller then decides what needs to live in the fast HBM2 chips at any given moment. You don't need to use PCIe bandwidth, go through the CPU, or (gasp) go to disk storage to get your dataset onto the card for processing after the initial load. You don't need to manually load pieces into or out of the GPU's memory. You just load your shit once and tell it to fucking go.

Re: 512TB of address space means nothing

[Anonymous Coward]

Thank you, was about to post the same comment. AMD probably had like 44b addressing support for a while, and maybe even more. Plumbing some more bits isn't free or trivial, but let's not pretend this is really groundbreaking... Nvidia was already there, this was underwhelmingly inevitable.

Re:512TB of address space means nothing

[Anonymous Coward]

Which part exactly in there is "marketing lingo" since you called it that? I'm not saying you're obfuscating and not giving details to support your position.

I'm implying it and calling you out implicitly. BTW, how many Nvidia cards do you need to heat your place anyway?

And have you stopped beating your wife, true scotsman?

Re:512TB of address space means nothing

[Cmdln Daco]

You just spend two hours loading everything into the GPU's memory. Then you start managing it, updating what parts of it change, etc.

That PCIe bandwidth you mentioned? It's pretty scanty when you're shuttling 512 TB of data through it.

Re:512TB of address space means nothing

[unixisc]

2^64 is 1.8x10^19. 1TiB is 2^40. So any 64 bit addressing scheme should at least cover 2^63 or 1,048,576TiB, assuming 1 address bit is sacrificed for the firmware or whatever else is needed.

So 512TiB is nothing, given what a flat 63 bit address space is capable of achieving. Also, supporting it on the address ain't difficult, given that there have been both data-address multiplexed lines as well as address-multiplexed lines.

Re:512TB of address space means nothing

[K. S. Kyosuke]

Address space is nice, physical cells even better. Unlike nVidia's CUDA, APUs at least give you *actual* unified address space, including memory protection, and SSG Radeons push the cells closer to the chip on the other side of the PCIe bus.

Re:512TB of address space means nothing

[K. S. Kyosuke]

I think he meant GPU/CPU memory moves without remapping pointers, as long as the GPU also has some MMU hardware available. Just because all the data doesn't actually fit into GPU's local memory at once doesn't mean you can't successfully pretend that it does.

Re:512TB of address space means nothing

[Anonymous Coward]

They're actually using NVMe drives as the extra "memory". This works out well for huge datasets where you take a performance hit streaming it from the host. Load up the data on one of those 4GiB/s NVMe SSDs. They already have a product out that does this and it makes certain workloads much faster. Just waiting to see an 8x PCIe 4.0 NVMe XPoint SSD. Will be wicked fast for what they use it.

Re: 512TB of address space means nothing

[Anonymous Coward]

There is already technology available to feed this monster. Things like the EMC DSSD can have 1/2 PB of NVMe flash connected via a PCIe bridge, and presented as a single shared memory mapped space to an entire rack if servers. I assume that is the use case for these cards, mostly in the supercomputing space.

Re: 512TB of address space means nothing

[Anonymous Coward]

Nothing in the article mentions memory bandwidth, unless I missed it.

Re: 512TB of address space means nothing

[Anonymous Coward]

It's even scantier if you need to load it more than once. The point is this card lets you not have to do that. A battery backup on the card, like with raid, means it will persists across reboots too. In case you wanted to throw that second red herring out too.

Re:512TB of address space means nothing

[Anonymous Coward]

So I'm guessing you slept through all those lessons on computer architecture, OS design and the like? Oh well, I won't bother mentioning all the cool stuff you can do with all of that "wasted" memory mapping. Guess you would have sided with Bill back when he was making proclamations about 640k being enough for anyone.

At the very least you must admit that they've neatly future-proofed their memory address space.

Re:512TB of address space means nothing

[Lisandro]

Oh, it is nice, don't get me wrong :) I'm just saying that promoting this as a dealbreaker is insane.

Re:512TB of address space means nothing

[Lisandro]

Pretty much all of it. I'm having a hard time finding out the number of addressing bits supported by, say, an Arctic Lake (4xx) GPU, but considering that AMD's GCN offered unified memory on the entire 64 bit space since 2011 and nVidia offers 49 bits of unified address space since CUDA 5 it surprises me that someone tried to make a selling point out of this feature.

They went as far as comparing a CPU render against their new GPU. WTF.

The only reason i can imagine someone would try to push this feature is that 512TB sounds like a huge number. There's no practical application for it in the near future, and any benefit of such a large addressing space you already got on previous architectures, both from AMD and its competition.

Re:512TB of address space means nothing

[Lisandro]

Don't be a moron. You already got virtual memory mapping over huge address spaces on previous-gen GPU products, both from AMD and nVidia. Vega looks like a nice architecture but all these hyperbolic performance claims based solely on having 512TB of addressable space are utter bullshit. I'm actually surprised that most people here can't tell the difference.

Ah, and i'm pretty sure that "future-proofing" a GPU architecture that will be obsolete 5 years from now was certainly not a consideration for AMD engineers working on this.

Re:512TB of address space means nothing

[K. S. Kyosuke]

It can be once you get used to the convenient programming model. The same thing happened a quarter century ago regarding the segmented memory model.

Re:512TB of address space means nothing

[Anonymous Coward]

TFS summary is talking about "512 TB", which would be 4096 Tb. I think OP was at least originally referring to bytes, not bits. TFA is referring to bytes.

Anyway, it doesn't mean the cards themselves would need to have such huge amount of memory, it means they are now able to handle a memory address space that vastly exceeds their physical memory. Graphics cards don't need to have everything in local memory.

Re:512TB of address space means nothing

[Anonymous Coward]

It has some meaning for the architecture and implementations of that, not so much for every single consumer graphics card of today. As a marketing device it is only relevant for people building clusters, which enable them to enter the "church of the early day saints of the holy Nvidia."

Re:512TB of address space means nothing

[Anonymous Coward]

More meaning than it had when AMD presented it as GCN back in the early 2010s?

Re:512TB of address space means nothing

[Anonymous Coward]

It's all about the software and the ecosystem, so it depends, doesn't it?

Re:512TB of address space means nothing

[Anonymous Coward]

Please stop spreading misinformation, as you obviously do not understand what this technology is about. We're talking about memory address space, not physical on-board memory.

Re:512TB of address space means nothing

[Lisandro]

You really need to re-read my parent post.

Re:512TB of address space means nothing

[Anonymous Coward]

" and any benefit of such a large addressing space you already got on previous architectures " - CITATION NEEDED.

Re:512TB of address space means nothing

[Lisandro]

Anyway, it doesn't mean the cards themselves would need to have such huge amount of memory, it means they are now able to handle a memory address space that vastly exceeds their physical memory. Graphics cards don't need to have everything in local memory.

I've replied this below a good number of times below, but for completeness sake: this is not new. Not even for AMD, which i'm unable to find the exact number of addressable bits for each GPU family but they all support unified virtual memory with 64 bits CPUs since the days of the HD7700. Hell, Linux has support for this feature since 3.20.

nVidia? CUDA 5 has you covered as well.

My point is: current GPU offerings can already address way, way more memory than they usually physically carry. No idea why someone would push this as a selling point other than 512GB sounds like a large number to people who don't know better.

Re:512TB of address space means nothing

[Lisandro]

512TB, sorry.

Re:512TB of address space means nothing

[Lisandro]

AMD has supported this for years.

Re:512TB of address space means nothing

[sexconker]

You don't understand.

The architecture can address that much, but the actual product will only address what's available.
There will be on-package HBM2 and the ability to connect to on-board (but off-package) storage in the form of fast flash.

512 TB of addressable space is just future proofing to allow for seamless work with a dataset regardless of whether it's on the 16 GB of ball-smackingly fast HBM2, on the SSD on your RadeonPro card, or in your system memory (or potential even abstracted out to disk storage). The drivers and GPU's memory manager handle moving the data around as you work on it. You shouldn't have to explicitly manage the dataset and thus not have to load it twice. In a worst-case scenario you'll be moving stuff back and forth over the PICe bus, just as you do now. But that'll only happen if you're trying to be stupid or your dataset exceeds the capacity of the on-board memory and SSD.

Also keep in mind that AMD will almost certainly be releasing APUs (their CPU+GPU combos) with Vega cores and HBM2 memory. This is all an extension of their previous push for HSA (heterogeneous system architecture) - essentially tying CPU and GPU and memory and everything else closely together and letting everyone talk to each other in the most efficient way possible.

Re:512TB of address space means nothing

[Anonymous Coward]

There's nothing "hyperbolic" about it. It's a single point of talk.

Re:512TB of address space means nothing

[Cealestis]

This isn't so the GPU can address just vram, it allows their new GPU HBM2 cache to directly address data in memory and disk as well. For big simulations on a cluster utilizing GPUs this can be a huge benefit.

Re:512TB of address space means nothing

[Anonymous Coward]

Did you read TFA? There's an entire paragraph singing the praises of the 512TB address space.

Re:512TB of address space means nothing

[WorBlux]

Or maybe custom super-computers, a lat HP's "The Machine", with gobs and gobs on non-volatile system memory (persistent RAM), or some other configuration where the Vega chip can access a lot of other memory over a dedicated fast fabric.

Re:512TB of address space means nothing

[K. S. Kyosuke]

On the programming side, yes. Of course they have. But even in AMD's APU hardware, proper memory architecture was delayed at least until Carizzo. You still need proper hardware to properly take advantage of it.

Re:512TB of address space means nothing

[Anonymous Coward]

Good call on the latter part. If they do get to releasing their CPU + GPU + HBM2 product on server socket in particular, then it will be mundane to have 512GB DRAM or more playing the part of the secondary storage, and it'll be eight channel of DDR4 so quite fast, if still much slower than graphics memory.

Re:512TB of address space means nothing

[Anonymous Coward]

Your parent post is just as silly. This feature has nothing to do with gaming, which doesn't need more than 8GB currently, and everything to do with machine learning, modeling, and simulation applications.

You know, for Gamers!

[Anonymous Coward]

Simulate an entire universe in your rig, with realistic physics all the way down to Planck scale.

Re:You know, for Gamers!

[Anonymous Coward]

at .0000000000000000000000000000012 fps

Re:You know, for Gamers!

[Anonymous Coward]

Framerate on the outside doesn't matter to the lifeforms inside the simulation.

Re: You know, for Gamers!

[dfeifer]

I want my holodeck..

Re:You know, for Gamers!

[Cmdln Daco]

And we are completely safe from said lifeforms, since nothing they do will ever emerge.

Person A: "Is it bricked, or just really really busy?"

Person B: "Hmmm, maybe if we wait a little longer..."

Re:You know, for Gamers!

[MancunianMaskMan]

yes, after all, what happens in Vega, stays in Vega.

Re:You know, for Gamers!

[fbobraga]

the wait really worth-ed: my own "The Matrix"!

Enough for

[Anonymous Coward]

512 TB ought to be enough for anybody.

O.. wait..

But, but, but...

[Anonymous Coward]

Bill Megabucks Gates said: "640 kilobytes is more than enough for anyone."
As the richest pig in the sty, he must be right.

Re:But, but, but...

[Anonymous Coward]

That's an urban legend. Bile Gates may be one of the greediest, most dishonest, and most unethical men in the world -- but he's not stupid to say something like that.

(Really, anyone believes his "foundation" is anything but a tax avoidance scheme? Don't be stupid.)

Re:But, but, but...

[__aaclcg7560]

I had a computer instructor in the early 1990's who said that 4GB on a 32-bit processor was enough RAM for anyone to use. For the most part, he was correct. I had a 4GB system for nine years before I upgraded to a new motherboard with 8GB. The only time I ever use more than 4GB is when I'm playing a videogame or encoding video.

Re:But, but, but...

[91degrees]

(Really, anyone believes his "foundation" is anything but a tax avoidance scheme? Don't be stupid.)

How would that even work?

Re:But, but, but...

[Anonymous Coward]

4GB is still sufficient if it is used efficiently.
1GB would probably be more than enough if it weren't for all the bloat.
I mean, you run countless of different scripted languages where everything both has an original text representation, possibly with 16 bit per char, and a "compiled" representation that computations are made from. On top of that even the simplest datatype is probably linked in a list and/or a reference stored in a hashtable.
We are way beyond the "Oh, perhaps there is 10% memory overhead." into a world where perhaps 10% is the actual data if we are really lucky.

2017 might be AMD's year

[ITRambo]

With Rizen coming out soon and a new GPU design that looks very advanced, AMD is set to make substantial progress in market share, as long as they don't screw up. I'm rooting for them. I had switched all of our shops new PC's to Intel when they released their 6th gen Core series as AMD was just too far behind. Teh consumer PC's were all AMD for the past five years or so. I wanna go back to AMD, as long as the new stuff performs. Don't let us down AMD!

Re:2017 might be AMD's year

[fisted]

But 2017 is already the year of the Linux desktop...

Re:2017 might be AMD's year

[Highdude702]

on AMD Hardware.

Re:2017 might be AMD's year

[Cmdln Daco]

1982 was the year of AMD hardware.

Re:2017 might be AMD's year

[Z80a]

Linux will never be on the desktop.
10 years from now, the linux powered terminators being commanded by the linux powered skynet, riding their flying linux powered bikes will still use Windows on their desktop computers, while performing experiments on installing linux on living people.

Re:2017 might be AMD's year

[Anonymous Coward]

Please. Let's see how they answer the 1080ti.

Re: 2017 might be AMD's year

[Anonymous Coward]

They answered last year, with two r480s at a fraction of the price. That you could buy at MSRP.

Re:2017 might be AMD's year

[Anonymous Coward]

Any news on when either might be a viable alternative to nVidia?

Re:2017 might be AMD's year

[Anonymous Coward]

too late, I bought Intel and Nvidia. next rig update 2020.

Re:2017 might be AMD's year

[Anonymous Coward]

when will you amd fanboys learn?

they've produced absolutely nothing but disappointment since their initial dual core chips, which predate intel's "core" architecture.

graphics? forget that, too. nvidia's had the lead there nearly as long as intel has had it in processors... and so long that radeon is absolutely irrelevant to gamers, just as amd processors are.

if you want to game, you have intel and nvidia. if you have no money, even, for hardware, you still have intel and nvidia because at the lower end of cost, they still kick amd's ass.

Re:2017 might be AMD's year

[Anonymous Coward]

strange, I have been gaming fine on AMD since the K6. I'm not that pleased with ATI parts, but I've happily let the radeon in my family (since the 5950 maybe?).

I have never had issues associated with AMD parts and I'm fairly certain performance per cost has exceeded equivalent nvidia and intel parts too. What is your problem? These days I mostly stream stuff from my rig to my silent AMD APU in my living room. A great combination. Most of the titles I play these days even run linux..

Re:2017 might be AMD's year

[The-Ixian]

I have only built AMD systems for nearly 20 years. My current AMD system is almost 4 years old now but still drives 5760x1080 pixels in most modern games with reasonable quality and frame rate (30fps or more). It also boots in less than 10 seconds and is an all-around snappy computer. I have not yet found a reason to upgrade.

Re:2017 might be AMD's year

[Anonymous Coward]

10 years from now, the linux powered terminators being commanded by the linux powered skynet, riding their flying linux powered bikes will still use Windows on their desktop computers, while performing experiments on installing linux on living people.

After conquering the world and subjugating humanity, the only real challenge left is keeping Windows 10 from trying to install itself on the remaining Windows 7 boxes.

Re: 2017 might be AMD's year

[ezelkow1]

Sure as long as you only played their one bland boring dx12 game, ashes of the singularity. Anything else and dual r480's still usually get smoked by a standard 1080, let alone whatever the ti version might be. Hell they just about match a 1070

http://arstechnica.com/gadgets...

Re:2017 might be AMD's year

[Anonymous Coward]

with an Nvidia graphics card and Intel networking.

News for nerds?

[Orgasmatron]

The "news for nerds" version of this story's headline is "AMD Unveils Vega GPU Architecture With 49 bits of Memory Address Space"

Re:News for nerds?

[Luthair]

What we really need to know is if they tested a Beowulf cluster of them.

Re:News for nerds?

[Kjella]

The "news for cynics" version of this story's headline is "AMD unveils yet another set of powerpoints". Where is (Ry)Zen? Where is Vega? Every month is another month Intel and nVidia rule unchallenged on the high end. We need actual product on the shelf, not more tech demos. And I bet so does AMDs financials, you have to actually hard launch it before you get any revenue. I'm a bit hyped out, now it's more like hoping for a miracle.

Re:News for nerds?

[lcllam]

And whether it runs Crysis(tm).

Re:News for nerds?

[Cmdln Daco]

I think he meant SKUs in retail outlets and in eShops that can be ordered, not vapor-hardware at a trade show.

Re:News for nerds?

[Anonymous Coward]

This one should be able to run Windows leaks !

Re:News for nerds?

[CanadianMacFan]

Did they blend a Beowulf cluster of them?

Re:News for nerds?

[Anonymous Coward]

Strange vapor-ware[sic]. Did you even read the article? The CPU exists, the motherboards exists, the coolers exists. That's some pretty solid vapours you've got right there.

OP whining about not being able to go to an outlet and buy them like, now, now, now doesn't make them "vapour". It makes him an self-entitled whiny little bitch who wants what he wants when he wants and if people doesn't make that happen, he'll got to the Internetz and whine about what massive failures they are, dammit!

Re:News for nerds?

[Areyoukiddingme]

Strange vapor-ware[sic]. Did you even read the article? The CPU exists, the motherboards exists, the coolers exists.

Well one of them does, at least...

Good Video Analysis of the data so far

[Anonymous Coward]

Here is a good analysis of the data released so far: https://www.youtube.com/watch?v=E4BMhquFvec (Funky channel name 'RedGamingTech', but very high quality analysis of hardware, specifically high end PC components & video game consoles).

512 terabyte addr space should B enough 4 anybody

[youn]

Is that how the saying goes? I am not sure :p

Probably's not how many memory chips you can fit

[Anonymous Coward]

Remember how how CPUs which had a memory management unit seemed dull to end users? Why would I want a 386SX above a 286?

It sounds to me that AMD is on a whirl, which is what we need to stop Intel and Nvidea getting complacent.

Roll on the APUs with the new graphics as well as their new cores to shake things up (until Intel manages to make process size do things noone can compete with).

Messing up the meme.

[hackwrench]

The meme around here is that nobody reads the articles.

Re:Messing up the meme.

[ArylAkamov]

Why bother reading when you can shitpost?

Re:Probably's not how many memory chips you can fi

[Lisandro]

I really want to see AMD releasing a CPU competitive with Intel's latest offerings. I love my 8-core FX, but the real reason I bought it is that it costs almost 1/3rd compared to the competition.

We been here before...

[__aaclcg7560]

The early 2000's was the last time had a Nvidia Geforce video card (256MB) with more RAM than my PC (192MB).

Re:We been here before...

[Anonymous Coward]

Well, that shouldn't have happened under normal conditions. You probably bought a card way out the league of your PC.

I remember having 256Mb on my PC when my top of the line videocard had 64Mb.

Re:We been here before...

[__aaclcg7560]

Well, that shouldn't have happened under normal conditions.

I was a professional videogame tester at the time.

You probably bought a card way out the league of your PC.

Back then I was probably upgrading my video cards every year. After I rebuilt my gaming PC for Windows Vista in 2007, I upgraded my video cards every three years.

Re:We been here before...

[Anonymous Coward]

I remember scavenging a PC/AT and replacing the MDA non-graphics card with an old Trident SVGA. 512KB RAM, 512KB graphics RAM. Ran some old games nicely.

Good Video Analysis of the data so far-vGPU.

[Anonymous Coward]

Unfortunately nothing on their technology for GPUs in a virtualization environment.

Re:Good Video Analysis of the data so far-vGPU.

[Khyber]

Well, don't bother looking at any Microsoft Benchmarks. They killed RemoteFX performance by half in Server 2012. I still run my stuff on 2008R2 for this reason.

Forget 512TB of address space

[Anonymous Coward]

You should read that as supporting 49 bit address. Woo. They added some bits. Will get back with you on this in 2025 when it actually means something.

Re:Forget 512TB of address space

[Cmdln Daco]

Are they really allocating 49 pins on the CPU package to the address bus?

Will there be Vega-SX parts that are in much cheaper packages and only put out 40 pins?

Re:trash posting

[hackwrench]

Much of the time the summaries give enough additional information to what some commenters have learned from other sources for them to make useful comments. Some commenters aren't most commenters though, and sometimes metacomments like these are in order.

Re:Probably's not how many memory chips you can fi

[Anonymous Coward]

and gets its' ass handed to it by a core i3

Re:Probably's not how many memory chips you can fi

[tginouye]

That's why my FX 8350 stays at home in it's "Safe space". Those mean intel CPUs are always out picking fights.

Re:Probably's not how many memory chips you can fi

[K. S. Kyosuke]

...in any antiquated single-threaded application, sure.

Re:trash posting

[ArylAkamov]

I wasn't talking about you, if you were thinking that. Besides, meta comments are best comments.

512TB? Why?

[Khyber]

What's the point? By the time we hit that amount of memory on a GPU, we're looking at this architecture being entirely obsolete.

Should've just said "We're slapping 1TB on this bitch!" and been done with it. No point in fussing about the scalability of the architecture when we're likely never going to see it hit full potential until long after its deprecated (AGP slot, anyone? When PCI-E cards came out, we'd barely even thought of saturating a 4X AGP slot.)

Re:512TB? Why?

[smallfries]

If you have a small address space then you need to write code that manually pages / caches the working set for an algorithm from storage. If you have a large address space then you use an interface similar to mmap and address the large dataset directly. It makes the code easier to write, and means that the paging / caching can be handled in hardware, where there are opportunities to speed it up.

Re:512TB? Why?

[Anonymous Coward]

Exactly. Consumer video memory increased 128x in the last 16 years... 512TB? I say Vega's chips will be at museums long before video cards even come near a tiny fraction of this...

Re:512TB? Why?

[Lisandro]

Keep in mind large address spaces were here long before Vega. Hell, AMDs own "Graphics Core Next" architecture already supports flat 64-bit addressing, an that's been out since 2011.

Re:512TB? Why?

[squiggleslash]

The Intel i386 had a 32 bit address bus back in the late eighties. Nobody could afford 4Gb back then for the type of machine that would have one. I had an Amiga with 5Mb of RAM and people ooh'd and aah'd about that.

But it didn't matter. Early 32 bit machines didn't use the top bits to support more RAM, they used it to support more functionality. Flat address spaces, with VM used to locate memory exactly where it needed to be.

I suspect the aim is similar here too.

Re:512TB? Why?

[Bob the Super Hamste]

AGP, try VESA Local Bus when talking about a limited use bus that was deprecated quickly. Add in that VLB cards were a real bitch to get in to the slots, or at least that is my most vivid memory of them.

Re:512TB? Why?

[Anonymous Coward]

VLB was a total hack and a real pain in the ass.. But it sure was fast!

Considering it was literally just a pin extension of the 486 processor bus, it had completely unrestricted and direct access to the processor and main memory. It was directly connected to both! (Also one of the reasons it was a pain. (Physically and electrically it was sketchy proposition and it was locked to the bus clock. Compatibility was all over the place. The best system builders did lots of testing and stuck with known good combinations of processors/motherboards/videocards/controller cards)

It's one of the reasons later 486s were faster than early pentiums in practice. Early PCI implementations were not great and the operating systems at the time did not take great advantage of PCI either. Most things you ran were single task and ran pretty close to bare metal (Dos/3.11). Had little need for bus mastering or greater system bandwidth until serious multitasking was more common. (Like running a TCP/IP stack, music player, a browser, and office suite all at once)

Re:Probably's not how many memory chips you can fi

[Lisandro]

Not really, sadly enough. Intel's single core performance is so above AMDs that for most computational tasks it makes little impact.

Let's get real here

[zifn4b]

Someone check me on my logic here. The way I read this article is that AMD has created a new architecture with a memory controller that can address 512TB of memory address space. That's great and all but are we going to see cards any time in the near future with 512TB of GDDR on them? Not likely. How many years away are we? Who knows. It seems to me this is highly theoretical and possibly to put pressure on the memory industry to innovate on even more dense memory to push graphics even farther to the limit. It could also be to get some investor interest in the next "big thing".

Side question: How did AMD validate that their architecture works without actually being able to fabricate an actual board in practice, simulation?

Re:Let's get real here

[Lisandro]

Side question: How did AMD validate that their architecture works without actually being able to fabricate an actual board in practice, simulation?

You don't need to actually hook up memory to see if a memory bus works correctly. I used to test addressing on 8-bit CPUs using a Tektronix logic analyzer back in college.

Re:Let's get real here

[Anonymous Coward]

AMD can actually connect SSDs directly to a GPU. So you can have your 32GB of HBM2 memory with ridiculous bandwidth, and a multi-TB SSD on the other side of the board, and you have a self contained parallel supercomputer on a PCB, that just happens to also run Crysis. Really fucking well.

Re:Let's get real here

[zifn4b]

AMD can actually connect SSDs directly to a GPU. So you can have your 32GB of HBM2 memory with ridiculous bandwidth, and a multi-TB SSD on the other side of the board

That's a neat idea. It's like you turn a PCI-E SSD into a modern day Voodoo 2 card

Re:Probably's not how many memory chips you can fi

[pezezin]

Take a look at this: https://www.cpubenchmark.net/c... It's faster than even the latest i5-6600, and cheaper. Only the i7's are faster. And don't tell me about single core performance, it's 2017, any program I care about is multithreaded. Now if you talk about energy consumption, you may have a point...

Re:trash posting

[Anonymous Coward]

Most of the time the article is two weeks old when it hits Slashdot so it shouldn't be surprising if there are plenty of people that read the article or even another article based on the same original source.
Unfortunately this means that a comment can seem uninformed because it rambles on about information that is misrepresented or weren't included in the linked article but that existed in the original source.

First step towards integrated CPU

[ET3D]

Given the relative sizes of CPU's and GPU's, it makes sense that an 'APU' will be a GPU with a bundled CPU, rather than the other way round. Having a large address space is one requirement for doing virtual memory on a card.

512 TB should be enough for any *CLUSTER*...

[DrYak]

Fixed the title for you.

And I think that's genuinely the point of this:
it would be possible to run a whole cluster of compute nodes with VEGA GPUs,
and have all the data within a single unified address space across the whole cluster.

Just throw in a few IOMMU to handle access rights, and a fabric like Infiniband, or some PCI-E based one.

For bonus point have the storage it self being memory mapped non-volatile RAM (X-Point, etc.)
(But then you DO run out of address space - clusters tend to have data in the peta-byte range).

AMD ruined ATI with shit drivers

[BeCre8iv]

Then they effectively discontinue a whole range of DX11 4gb GPUs before even writing working drivers. Waste of time, money and silicon.

512TB of ADDRESSABLE memory!!! Seriously, RTFA.

[Calabacin]

It's funny how not RTFA in Slashdot is even a meme but yet I can see that 95% of the people here didn't read it. The 512TB number is the amount of ADDRESSABLE memory, which means that you can reserve for example 300Gb of that memory to read a texture file that big. Then, as you start reading it, a secondary controller will transfer data there from main memory, directly from disk or from wherever. To you it will be as if you were reading a 300GB block from Video memory and thanks to that external controller (IOMMU, DMA, etc.) that transfer will be super fast. While you are doing that other application will be doing the same with a 1TB block. Both will be easily accesible at the same time thanks to having 512TB of ADDRESSABLE memory. Each of those applications will only use a part of the total REAL memory for this access. No more manual overlay/pagination will be necessary.

This technique is commonly used in modern systems, but it was revolutionary when it first appeared. AMD has now brought it to the GPU now so that it can be used with many more applications. Apparently they are bringing the CPU to the GPU instead of the usual other way around.

This new GPU also has "over 200 new features", for those who say it's nothing new.

Not a fanboy or anything, but all these rants that could all be avoided reading the article or paying attention simply got to me. Sorry about that.