bugI must have missed it, what's revised in the new Ryzens?

There's the cache & mem latency numbers which look great, also higher (speed) mem support.

R0H1TThere's the cache & mem latency numbers which look great, also higher (speed) mem support.

The (slightly) lower latencies could be an effect of everything getting smaller, thus needing less electrons to fill up, it's not a confirmed tweak. And I can't find anything about faster memory support at a glance.

bugThe (slightly) lower latencies could be an effect of everything getting smaller, thus needing less electrons to fill up, it's not a confirmed tweak. And I can't find anything about faster memory support at a glance.

Simply put: first gen. ryzens are specced to 2667MHz and second gen. are specced as Jedec 2933MHz.

Fabiome neither, because i don't want to help those miners to repay their cards and because they are abused

that is pure supposition.. mining cards are not abused.. in fact most run under powered to keep the power consumption down.. i have 10 x 1070 cards that have been running 24/7 at 75% power for the last six months.. all of them still look and work like new..

i am curious as regards how long they will last running 24/7 but i strongly suspect quite a long time.. if gpu mining really does die i aint gonna get to find out but i would like to.. :)

trog

AMD is adopting 7nm first just as they did with with 40nm. 8K processor count should be doable. Mining cards may be well cooled and underpowered but the memory chips are burning hot, use infrared to see for yourself. I see 90C om mem, whereas the core is only 50C, hint memory chips should operate under 85 according to spec, and 65 according to common sense... Those cards will be artifacting very badly the day after the warranty expires.

trog100that is pure supposition.. mining cards are not abused.. in fact most run under powered to keep the power consumption down.. i have 10 x 1070 cards that have been running 24/7 at 75% power for the last six months.. all of them still look and work like new..

i am curious as regards how long they will last running 24/7 but i strongly suspect quite a long time.. if gpu mining really does die i aint gonna get to find out but i would like to.. :)

trog

still remain the fact i will never ever help a f* miner to return his investment after destroying pc market. Prefer to buy new card, thank you.

OMG they released the patches for Vega 10 at the same time last year, 3 months ahead of the Frontier Editions announcement.
All aboard the hype train!!!

iOOMG they released the patches for Vega 10 at the same time last year, 3 months ahead of the Frontier Editions announcement.
All aboard the hype train!!!

even when mining didn't exist, gpu manufactures underestimated demand every single launch. nothing will change, you won't get your hands on one for 6-10 months from now. even with a 3 month from now launch. also, I am hesitant AMD will ever be able to match a 1080 ti OC'd. i wish they could, but I just don't see it in the stars. same as AMD will never be able to match Intel in min fps game rates across the board.

if ryzen 2 and vega 2 in 2019 can change the game - I'm all in AMD... but i just have a feeling Intel/Nvidia will decimate them with their equal releases around the same time... sigh. sadly I need the best performance, because I need my 165hz 1440p, or life has no meaning anymore. :D

edit: not kidding btw ^ I tried playing a PS4 recently and it was so sluggish looking I couldn't stand it.

I hope they can just build them this time. Vega 56 was actually competitive, and I wanted a vega 64 watercooled, even if it wasnt a great deal compared to the 1080/ti.

But AMD just couldnt build the things. By the time they were in stock, prices were going through the roof.

TheinsanegamerNBut AMD just couldnt build the things. By the time they were in stock, prices were going through the roof.

Perspective. AMD manufactured as many chips as one could reasonably expect , remember that Vega silicon is comparable in terms of costs to GP102 inside the Titan Xp , not GP104. As a matter of fact they probably produced way more chips than Nvidia did for their consumer Titan products.

So 12nm refresh, 10nm is 1/2 of a shrink. 7nm is 1/3 the size of VEGA64 and this could be positioned as VEGA264, RX680 or RX660, great.

Vya DomusPerspective. AMD manufactured as many chips as one could reasonably expect , remember that Vega silicon is comparable in terms of costs to GP102 inside the Titan Xp , not GP104. As a matter of fact they probably produced way more chips than Nvidia did for their consumer Titan products.

Additional perspective: If Vega 64 was their 1080 competitor, then AMD needed to produce them as fast as nvidia could produce 1080s. If VEGA was too big to mass produce, that is a flaw in the design, and AMD needs to fix that flaw if they introduce new vega cards.

Vya DomusPerspective. AMD manufactured as many chips as one could reasonably expect , remember that Vega silicon is comparable in terms of costs to GP102 inside the Titan Xp , not GP104. As a matter of fact they probably produced way more chips than Nvidia did for their consumer Titan products.

Well, Nvidia doesn't really care how many Titans they sell, but Vega was supposed to cover AMD's high end. If they made it so hard to build, that's an engineering failure.
But I still don't think manufacturing the chips held them back as much as the lack of HBM2.

bugWell, Nvidia doesn't really care how many Titans they sell, but Vega was supposed to cover AMD's high end. If they made it so hard to build, that's an engineering failure.
But I still don't think manufacturing the chips held them back as much as the lack of HBM2.

That's pretty much it, from TPU2 to Intel Nervana & now Volta, everything has HBM2 in it. Now compare the price of Vega to them & few other ASIC, there's no way Vega can be cost effective for AMD.
Probably the same reason why they went with 2 stacks of HBM2 instead of 4 (16GB) as was rumored.

TheinsanegamerNthat is a flaw in the design, and AMD needs to fix that flaw if they introduce new vega cards.

There is no flaw , that's just the result of the card they designed with the features they wanted. You don't make a chip huge by mistake and filled with nothing , there is nothing to be fixed other than the unfitting manufacturing process.

At least we're likely to get rx500x cards with slightly higher stock performance to tide us over and the 2700x is looking pretty cool!

ppnAMD is adopting 7nm first just as they did with with 40nm. 8K processor count should be doable. Mining cards may be well cooled and underpowered but the memory chips are burning hot, use infrared to see for yourself. I see 90C om mem, whereas the core is only 50C, hint memory chips should operate under 85 according to spec, and 65 according to common sense... Those cards will be artifacting very badly the day after the warranty expires.

These chips do not operate at 90 degrees by themself, but are being heated up by nearby VRM. That is the problem for any card packing a chip that requires a high current.



You can see the much cooler area at the other side of the card. The only way to solve this is to increase VRM cooling or simply turn up the fan.

Performance wont be any better when cooling, but the longevity of your GDDR chips will be extended and perhaps higher clocks on the long run, since it's not throwing so much heat through the card.

Vya DomusThere is no flaw , that's just the result of the card they designed with the features they wanted. You don't make a chip huge by mistake and filled with nothing , there is nothing to be fixed other than the unfitting manufacturing process.

Actually, you don't make a chip huge at all. Unless you have a really, really good justification for it. Because huge chips have always been problematic to build.

Vya DomusThere is no flaw , that's just the result of the card they designed with the features they wanted. You don't make a chip huge by mistake and filled with nothing , there is nothing to be fixed other than the unfitting manufacturing process.

If you have to make a halo sized chip to compete against a high end chip, and use HBM to keep total TDP in check, you have a pretty big flaw somewhere.

The problem is the chip doesnt fit in line with what its competitor can do in the same space. The manufacturing process works fine for nvidia, the problem is AMD's design, not the manufacturing process. Unless you are telling me AMD engineers wanted a titan XP sized chip with expensive HBM to only compete with a much smaller 1080 while pulling tons of power.

Hopefully AMD took some of those lessons to heart when modifying vega for a new process.

TheinsanegamerNThe manufacturing process works fine for nvidia, the problem is AMD's design, not the manufacturing process.

How do you know ? They use a different process. TSMC is miles ahead of everyone.

TheinsanegamerNUnless you are telling me AMD engineers wanted a titan XP sized chip with expensive HBM to only compete with a much smaller 1080 while pulling tons of power.

They wanted a compute chip first , Vega was not made for gaming as it's primary use. If they wanted an equivalent for the 1080 they would've just scaled up Polaris but they didn't , because that wasn't their goal.

TheinsanegamerNIf you have to make a halo sized chip to compete against a high end chip, and use HBM to keep total TDP in check, you have a pretty big flaw somewhere.

The problem is the chip doesnt fit in line with what its competitor can do in the same space. The manufacturing process works fine for nvidia, the problem is AMD's design, not the manufacturing process. Unless you are telling me AMD engineers wanted a titan XP sized chip with expensive HBM to only compete with a much smaller 1080 while pulling tons of power.

Hopefully AMD took some of those lessons to heart when modifying vega for a new process.

More RAM perhaps and maybe a few more cores, but probably still a smaller, hopefully less powerhungry, gpu is likely. Gddr6 adoption would help keep costs down as well, so the next completely new rx580(x) replacemnt should still be priced decently and perform much better or at least much more efficiently (much closer to 1060 efficiency).

bugThe (slightly) lower latencies could be an effect of everything getting smaller, thus needing less electrons to fill up, it's not a confirmed tweak. And I can't find anything about faster memory support at a glance.

No, everything would be exactly the same for a pure straight shrink. Cache latency changes indicate either improvements were made, or timings were tweaked with the process change. It's hard to know, perhaps both.

Vya DomusHow do you know ? They use a different process. TSMC is miles ahead of everyone.



They wanted a compute chip first , Vega was not made for gaming as it's primary use. If they wanted an equivalent for the 1080 they would've just scaled up Polaris but they didn't , because that wasn't their goal.

So, they wanted a compute chip, but then sold it as a gaming chip? Clearly they wanted a gaming chip, because they released vega 56/64.

If their goal was to have a compute chip, they failed massively, because they sold it as a gaming chip, not a compute chip. As you said, they could have scaled up polaris, the fact they didnt indicated they wanted vega to be their high end chip, and as a high end chip it failed. That is not GloFo's fault, that is AMDs.

Also, the 1050/ti are built on GloFo's 14nm, and dont have the same issues vega did (high power consumption, production issues). nvidia managed to make their arch work on two different processes without issue, AMD didnt get vega working properly on 1. If the process was an issue, nvidia wouldnt have used it. And yes, I know the 1050ti si a lot smaller, but the fact that they got it to work, with power consumption comparable to their TSMC parts, indicated that GloFo wasnt at fault for AMD's failures.

TheinsanegamerNAlso, the 1050/ti are built on GloFo's 14nm, and dont have the same issues vega did.

They are a fraction of the size and a totally different architecture. There is no way to know for sure how a similarly sized but higher clocked GP102 would have fared on that node vs Vega. But what is known for sure and confirmed by industry experts is that TSMC's 16nm was superior.

www.semiwiki.com/forum/content/6713-14nm-16nm-10nm-7nm-what-we-know-now.html

TheinsanegamerNClearly they wanted a gaming chip

Because you say so ?

Major overhaul of GCN cores in order to achieve 2x FP16 and 4x 8-bit Int operations per clock compared to the FP32 performance , all that for gaming ? It's a compute card above all else.

TheinsanegamerNIf their goal was to have a compute chip, they failed massively, because they sold it as a gaming chip, not a compute chip.

GP102 must have been a failure as well because they also sold it as a gaming card ? No it wasn't , Nvidia and AMD did the same thing , GP102(GP100) and Vega were clearly designed for compute first and then sold as gaming cards to each company's needs. This isn't a novelty unseen before , it's been done previously as well. The only reason you find it so hard to believe it's because AMD pretty much left it as it is and sold is a gaming card whereas Nvidia disabled or severely cut down some compute specific capabilities and then they sold it as the Titan Xp.

Vya DomusHow do you know ? They use a different process. TSMC is miles ahead of everyone.

It's not the process. Larger dies have been problematic to build for ages. If you understood a bit how chips are made, we wouldn't be having this conversation.