AMD Announces 2018 roadmap at CES

[Cyrops]

What a shitty time if you need a graphics card.

It's been that way in my country ever since Vega came out, I check my biggest online shop every week, Vega was in stock for a few days just as it hit, haven't seen a single one available since, all 1070's and up are pretty much sold out 24/7, for decent brands at least, just built PC for a friend last week with 8350k and 1060, but the 1060 was the 3GB, all the 6GB were sold out or costed 60e more (1/4th of a total).

[Evildeffy]

I think the only reference has been for 7nm Vega but that's a compute/AI card. Not a GFX card. My hope is that the APU's, provided they are fast enough, trigger a bit of a drop in pricing. If NVidia want to compete then they will need to have lower priced entry level cards for the discrete market. That will hopefully lead to a drop in the mid range cards because the gap between the entry level ones and the midrange cards will be massive. As it is, I was going to pick up new CPU and GFX cards for my kids but will seriously consider the APU's as an alternative. They make sense from both from a pricing perspective and an availability perspective.

Well ... It's only my guess, it's entirely possible that it is indeed sacrificed this year until Navi.

That said ... if the current rumours are true that the performance of an APU/Intel RX Vega M is at the performance level of a GTX 1060/RX 580 than I can only really say "Holy fuck!" really... because that'd be seriously good... like stupidly good.
So I'm going to wait till I see some proper benchmarks for that... but if they can pack that in then fuck yes we're going into a good direction.

- - - Updated - - -

You said this before and I asked then. Can you provide a reference for this?

You're never going to get an answer for this from Kagthul, you'd have more luck in getting struck by lightning from Zeus' arse in your face than him actually giving out a proper answer.
But in that light I'm going to give you the answer you very likely already know:
No, this is completely false information as all AMD uArchs all the way back to Soecket 754/939 Athlon64 (unsure of XP) CPUs are completely unaffected by Meltdown because even then uArchs were different, AMD may have started as an Intel clone but it diverged paths VERY soon after when Intel tried sabotaging it from even it's earliest days.
Thus his statement is completely false

I will state again for the misreading jockeys present on this forum:
Spectre affects all CPUs and is almost irrelevant for AMD as only 1 variant of the 2 affects them and that's patchable in the OS without performance deficit.
Meltdown is not even on AMD's radar as it's uArch doesn't give a fuck about it, in fact the patches to fix it for Intel have done more harm to AMD than Meltdown ever could do for AMD already with MS' botched update for a uArch that isn't affected.

Here is the thing. There are 2 very good reasons why maintaining a socket are a good thing.

1. Upgrading - I am not talking about going from a R5 1600 to a R5 2600 here. I am talking about going from a R3 1200 to a R5 2600. I am sitting with 2 PC's at home with I3-4160's and it's a nightmare to upgrade them. I bought them as a cheap gaming machine and I can't do squat with them now to upgrade them. I need to go and take a chance on E-Bay. There are a lot of people who buy small so they can upgrade later.
2. Maintenance - Motherboards go faulty, CPU's don't. That's my experience. I have an old I7-3770 and a I7-2600 and I have to replace the MB and CPU if the motherboards stuff out because you can't get them any more.

Maintaining a socket for an extended period of time is a very good thing for customers. Intel change sockets because it means more people upgrade and that's bad for consumers.

The fact he's already arguing that CPUs die more than motherboards is laughable enough.
Every techie worth his salt is aware of the fact that motherboards are considerably weaker (especially P4/PD/Athlon64 S754/S939 era motherboards) than CPUs, to even suggest that that's false is beyond stupid especially considering that some components simply do not live that long, Capacitors are a prime example.

Having said that ... I do agree that at some point upgrades are a good thing ... For those that actually have a disposable income.

I often help in charity situations where Dutch communities request for and build a lot of PCs for the less fortunate people among us where even an 8 year old CPU is fantastic for light gaming and basic use... but finding the motherboards for them is atrocious and that's when I'd honestly give money to be able to buy cheap motherboards capable of just being compatible for these types of events without the need of being high-end uber boards.

I can to this day still find perfectly functional Pentium 4 and Athlon64 S939 CPUs ... but no motherboards to properly support them.

[Zenny]

That said ... if the current rumours are true that the performance of an APU/Intel RX Vega M is at the performance level of a GTX 1060/RX 580 than I can only really say "Holy fuck!" really... because that'd be seriously good... like stupidly good.
So I'm going to wait till I see some proper benchmarks for that... but if they can pack that in then fuck yes we're going into a good direction.

Bit of clarification here, it's about as powerful as a 1060 Max-Q, which is significantly slower then a regular 1060 or 580. These stats also coming directly from Intel marketing so assume very bast case scenario as well. Even a 1060 3GB/ Radeon 570 will still have a commanding lead, especially when overclocked.

That being said, they are very good for APU's and can really shake the laptop market up. If they get priced cheap enough then they might even serve as potential console replacements in a tiny form factor, for example I can see them doing very well in a Steam Machine.

[Evildeffy]

Bit of clarification here, it's about as powerful as a 1060 Max-Q, which is significantly slower then a regular 1060 or 580. These stats also coming directly from Intel marketing so assume very bast case scenario as well. Even a 1060 3GB/ Radeon 570 will still have a commanding lead, especially when overclocked.

That being said, they are very good for APU's and can really shake the laptop market up. If they get priced cheap enough then they might even serve as potential console replacements in a tiny form factor, for example I can see them doing very well in a Steam Machine.

Yeah that's the Intel one but there's a Ryzen APU being built with 28CUs which is 50% of Vega 56 which brings it (roughly) on the level of GTX 1060/RX 580 in a single package, considerably cheaper than a separate card.

The uses I can think of for this is just drool worthy ... but then also something terrifying occurs to me ... 50% of a Vega 56 is GTX 1060/RX 570 level of mining performance ... dickbag miners might gobble that stuff up as well for extra "free" mining ><

[Zenny]

Yeah that's the Intel one but there's a Ryzen APU being built with 28CUs which is 50% of Vega 56 which brings it (roughly) on the level of GTX 1060/RX 580 in a single package, considerably cheaper than a separate card.

The uses I can think of for this is just drool worthy ... but then also something terrifying occurs to me ... 50% of a Vega 56 is GTX 1060/RX 570 level of mining performance ... dickbag miners might gobble that stuff up as well for extra "free" mining ><

Oh, I wasn't aware of that, AMD Fenghuang is apparently the name, that looks to be a killer chip indeed. As for miners, they might not really be interested, generally they want to bank as many GPU's as possible per CPU, so this APU would very likely not be cost effective for mining. Price and stock it right and you basically have the premier solution of mid->low tier gaming.

[Life-Binder]

Nvidia is going to be gods at this rate

absolutely untouchable, while Intel is hitting bumps


god, if only Nvidia supported VRR on HDMI 2.1, then it would be so perfect =/ .. if they dont then its going to be hell of deciding how to mix the components



in 2019 I will likely need to make a decision on getting either a Zen2 7nm or Ice Lake 10+nm

- - - Updated - - -

is 50% of Vega 56 which brings it (roughly) on the level of GTX 1060/RX 580

50% of a Vega 56 is not in any way equal to a 1060/580

[Evildeffy]

50% of a Vega 56 is not in any way equal to a 1060/580

50% of a Vega 56 is ROUGHLY (understanding the meaning of the word?) equal to a 1060/580 as you should have CLEARLY been shown in the past that having more or less processing units (stream processors or CUDA cores, doesn't matter) does not scale linearly in current day performance.

But in certain compute terms it is since a Vega 56 can do 44MH/s in cryptocurrency mining and since that scales relatively linearly in the same uArch with stream processors/CUDA cores you can make the deduction of it being 50% of as RX 580 is indeed slightly higher at 26 - 28 vs. GTX 1060 which is 20 - 22.

Half a Vega 56 on a CPU is entirely possible to be close-to or equal levels of performance to a 1060 6GB/580, and even if it was closer to GTX 1060 3GB/RX 570... who cares? That's still an insanely good package with saving a shitton of money.

[Zenny]

Half a Vega 56 should fit in just under a RX570 in terms of shaders and TMUs, but should have the same ROP count. So RX570 -10% performance I would guess?

As for saving a ton of money? Difficult to say, you would save on board cost and connectors a bit, but the two most expensive things on a GPU (the chip and memory) still remain.

[Evildeffy]

Half a Vega 56 should fit in just under a RX570 in terms of shaders and TMUs, but should have the same ROP count. So RX570 -10% performance I would guess?

Yes but as you can see with reduced counts of stream processors (and inherent other parts tied with it) that the reduction doesn't scale linearly.
So even it'd have some differences in amount of uArch parts .. it could entirely perform at that level.
Remember the Fury X2 card (forgot the name) gave up like 25 - 30% frequency vs. the Fury X card but didn't lose as much performance (only 1 GPU counting, not both GPUs) as it lost frequency, in fact it was only like ~10% or something?
Same scaling counts for the actually present ... everything.

Chances are because it's a scaled back uArch designed for higher throughput that it will perform close to a 1060/580 level of performance if the 28CU chip can flex it's muscles.

As for saving a ton of money? Difficult to say, you would save on board cost and connectors a bit, but the two most expensive things on a GPU (the chip and memory) still remain.

You'd be surprised how much cost the PCB and other components costs you because it includes marketing etc.
That is basically shifted to motherboard now which tends to have a slightly more robuust design of power circuitry and you no longer have to put engineering into the card as well as testing and validation, instead that is now mixed with a CPU's testing, saving enormous amounts of cost.

Granted the GPU core and HBM itself are still expensive but it's not as expensive as you may think component wise.
Overall everything around it is more expensive (considerably) than the core components.

That said right .... let's say it was 570 - 10% (which is already slower than the Intel Vega M according to leaks as the 1060 Max Q is about the same level of 570, so what will a 28CU unit do?) ... it's STILL in-friggin-sane for fitting on a CPU without an added card.

These Ryzen-Vega APUs are going in the right direction if these numbers hold up.

[Life-Binder]

theres nothing in Vegas arch that will let it magically perform with considerably less shaders, at the same level as 580

Vega IPC is essentially the same as Polaris


so Id advise agaisnt those claims unless/until they are verified by 3-rd party reviews

[Evildeffy]

theres nothing in Vegas arch that will let it magically perform with considerably less shaders, at the same level as 580

Vega IPC is essentially the same as Polaris

so Id advise agaisnt those claims unless/until they are verified by 3-rd party reviews

Vega and Polaris are not the same IPC, nor can you compare the stream processors (shaders) between the 2.
I am not claiming it IS equal to RX 580 / GTX 1060, I'm ESTIMATING it is ROUGHLY the same same, roughly being +/- 5 - 10%

uArchs are different between each other and Like Fury (X) before it when reaching a certain speed it plateaus but prior to that you can drop quite a bit and barely see any performance impact compared to what you're lowering it with.

It's always been like this because most games cannot fully utilize all the resources of a card (stuff like DOOM/Wolfenstein 2 and Sniper 4 (I think it was Sniper 4? Not sure) not included) and thus get limited, it is again not a linear scale.

Gamers Nexus' Titan V (5120 CUDA cores) review clearly shows this as the 1080Ti (which have 3584 CUDA cores (shaders) was under 4% behind in games where low level APIs were not fully native (titles mentioned above) and that's an increase of about 43%.

The comparison in this case is identical, the difference doesn't have to be that big, if you look back historically it's always been the same way.

AMD's Fury and Fury X, R9 290 and R9 290X (and the 300 series), HD 7970 and 7950 etc.
nVidia has similar comparisons between their Titan and Ti models.

If the Intel leaks are correct with it being on par or greater than 1060 Max-Q edition and the Ryzen APU has a 28CU (25% more CUs orso?) variant then the performance will be roughly close to the GTX 1060 / RX 580 cards.

Granted it's always the best waiting by reviewers ... but the potential alone is insane having the ability to just straight out dominate everything up to (and potentially including) the mid-range market on a single CPU die.

[Rennadrel]

Miners would still snipe them in .3 seconds in any case so us regular folk may as well come to terms that Vega cards are unicorns. I should have pulled the trigger on a 1070 Ti Duke the other day since it was $600CAD but now I see shortages in the GPU industry has come back with a vengeance in regards to availability and pricing for both sides. 580s are still over the moon and 1060s seem to be not far behind. 1080 and 1080 Ti? Forget about it.

What a shitty time if you need a graphics card.

$600? Christ, i paid less than that after taxes for my GTX 1070 when I bought it almost two years ago! Bought it on sale from NCIX for an Asus Strix model.

[Dald]

I'm glad I got my 1080ti sc2 when I did. Got it within minutes of it being released by EVGA for retail price.

[chronia]

I will state again for the misreading jockeys present on this forum:
Spectre affects all CPUs and is almost irrelevant for AMD as only 1 variant of the 2 affects them and that's patchable in the OS without performance deficit.

I wouldn't be to sure of this as AMD is now also going to issue Microcode updates for the Spectre 2nd Spectre variant that needs microcode updates. They would never do that if it was 100% sure that it would not affect them. Somehow it feels to me AMD has jumped the gun a little bit in some of their statements.

Nvidia is going to be gods at this rate

On the high end, probably yes, they won't have any serious competition till Navi, and Navi needs to very very good to be competition.

On the lower end and laptop market i can see some serious issues for Nvidia though as the AMD APU's and Intel / Vega package seem so competive that i wouldnt be surprised if they manage to push Nvidia basicly out of this market except for the very very high end laptops that come with GTX 1080's and the likes.

[Evildeffy]

I wouldn't be to sure of this as AMD is now also going to issue Microcode updates for the Spectre 2nd Spectre variant that needs microcode updates. They would never do that if it was 100% sure that it would not affect them. Somehow it feels to me AMD has jumped the gun a little bit in some of their statements.

They actually don't NEED those updates.
Papermaster has followed up on the speculative execution blog post in which he stated there will be OPTIONAL microcode updates to alleviate fear surrounding this bug.
It has already been tested on AMD CPUs and it doesn't work on their CPUs.

Here's the follow-up on their point:
https://www.hardocp.com/news/2018/01...down_statments

I mean this could be an elaborate dig hiding "Oh shit we found issues! Fix secretly and put it under the guise of optional!" ... but I think AMD would know that if it were found out in doing so it'd be in deep investor shit.
Considering that Google's Project Zero pretty much hasn't found the weakness on AMD CPUs either I'm inclined to believe that statement really.

[chronia]

I don't see where they state that they don't NEED those microcode updates looking at their own statement:

Google Project Zero (GPZ) Variant 1 (Bounds Check Bypass or Spectre) is applicable to AMD processors.
We believe this threat can be contained with an operating system (OS) patch and we have been working with OS providers to address this issue.
Microsoft is distributing patches for the majority of AMD systems now. We are working closely with them to correct an issue that paused the distribution of patches for some older AMD processors (AMD Opteron, Athlon and AMD Turion X2 Ultra families) earlier this week. We expect this issue to be corrected shortly and Microsoft should resume updates for these older processors by next week. For the latest details, please see Microsoft’s website.
Linux vendors are also rolling out patches across AMD products now.
GPZ Variant 2 (Branch Target Injection or Spectre) is applicable to AMD processors.
While we believe that AMD’s processor architectures make it difficult to exploit Variant 2, we continue to work closely with the industry on this threat. We have defined additional steps through a combination of processor microcode updates and OS patches that we will make available to AMD customers and partners to further mitigate the threat.
AMD will make optional microcode updates available to our customers and partners for Ryzen and EPYC processors starting this week. We expect to make updates available for our previous generation products over the coming weeks. These software updates will be provided by system providers and OS vendors; please check with your supplier for the latest information on the available option for your configuration and requirements.
Linux vendors have begun to roll out OS patches for AMD systems, and we are working closely with Microsoft on the timing for distributing their patches. We are also engaging closely with the Linux community on development of “return trampoline” (Retpoline) software mitigations.

Sure they call it optional, but they do state that the bugs are applicable to their processors, and they only "believe" that its hard to exploit. And i certainly don't believe thats it a lie that it will be hard to exploit. But if there was 0 risk they would have never released the microcode updates, and if there less than 0 zero risk, depending on the type of businesses the hardware will be used its hard to call something like this optional.

[Evildeffy]

I don't see where they state that they don't NEED those microcode updates looking at their own statement:

Sure they call it optional, but they do state that the bugs are applicable to their processors, and they only "believe" that its hard to exploit. And i certainly don't believe thats it a lie that it will be hard to exploit. But if there was 0 risk they would have never released the microcode updates, and if there less than 0 zero risk, depending on the type of businesses the hardware will be used its hard to call something like this optional.

We have seen some initial stories with a couple of inaccuracies so want to make sure we are being perfectly clear.

* There is no change to AMD’s position on our susceptibility to GPZ Variant 1 or GPZ Variant 2 (collectively called Spectre in many news reports).

* The update in relation to Variant 2 is that even though Variant 2 has not been demonstrated to work on AMD products due to differences in our micro architecture, out of an abundance of caution we are making optional micro code updates available to further contain the threat.

Again, to make it perfectly clear we have not changed our statement related to our susceptibility to Variant 2. Let me know if you have questions or need additional details.

Let me rephrase it:

Google has been unable to find any way to utilize Spectre v2 to exploit AMD CPUs.
AMD themselves have been unable to find a way to utilize Spectre v2 to exploit their CPUs with current OS designs due to uArch differences.

They do not dismiss the idea that it is somehow possible to exploit it like Intel's CPUs but that as of so far they've not found any way to do so yet.

So these optional updates are simply built to, if necessary, stop it from happening entirely if it is even remotely possible.

They have stated from the initial release that the expected risk is "Near zero" (their words) for Spectre v2 due to differences in uArch.

Whilst AMD has been known for stupid ass marketing (it really is atrocious) they've never been known to dick about in these matters.

Considering that, barring Google's own Project Zero team, security experts all across the globe have (so far) been unable to exploit Spectre v2 on AMD but "easily" so on Intel I'm inclined to believe the statement, it can always change in the future of course but I honestly don't think it will since the people trying today are incredibly smart on this front, far more so than any of us on this forum.

Like I said prior though... the hilarity of this situation is that the Windows/Linux patches originally rolled out for fixing these gaps (Meltdown) have been a bigger damage point to AMD than these issues themselves when they aren't affected.

You want to see something damaging though... check out what the Spectre/Meltdown fix does to Apple's performance... (not pretty)

[Triggered Fridgekin]

$600? Christ, i paid less than that after taxes for my GTX 1070 when I bought it almost two years ago! Bought it on sale from NCIX for an Asus Strix model.

I'm pretty much over it. Instead I just went and upgrade my board and processor to an i7-8700k with a nice beefy board since they were both under MSRP at Mikes and will re-evaluate in, like, June and hope for the best with a bigger budget.

[Zenny]

Let me rephrase it:

Google has been unable to find any way to utilize Spectre v2 to exploit AMD CPUs.
AMD themselves have been unable to find a way to utilize Spectre v2 to exploit their CPUs with current OS designs due to uArch differences.

Apparently that is not quite true anymore:

https://bugs.chromium.org/p/project-...detail?id=1272
https://bugs.chromium.org/p/project-...aid=287305#585

Even AMD has finally admitted it: https://www.engadget.com/2018/01/12/amd-spectre-patch/

True, it appears to be harder to exploit then on Intel chips, but the threat is still there.

[Evildeffy]

Apparently that is not quite true anymore:

https://bugs.chromium.org/p/project-...detail?id=1272
https://bugs.chromium.org/p/project-...aid=287305#585

Even AMD has finally admitted it: https://www.engadget.com/2018/01/12/amd-spectre-patch/

True, it appears to be harder to exploit then on Intel chips, but the threat is still there.

What you're looking at on the Google docs is from June 2007 and is Spectre v1, not v2. (from the first 2 links, link 2 being a write-up of link 1)

And I quoted an article up above from Papermaster that is in direct answer to link 3.
Again it is conceivable that it is exploitable but they have been unable to find 1 for v2, so they're releasing an optional micro-code update for it.

Here's that link for you: https://www.hardocp.com/news/2018/01...down_statments

From all angles I've looked about regarding finding any instance of Spectre v2 being shown exploited on AMD chips and have found nothing, Google's Project Zero hasn't found the v2 vulnerability either so again I'm inclined to believe Papermaster's "Just in Case" only.

That said it's entirely possible Cyber Security firms have already found the weakness but are under embargo/NDA for a while, but even so I'd expect them to post success if they would've cracked the vulnerability already since AMD is already bringing out an optional patch.

Though in the advent of exploits ... everything is possible.