While true, we do need more competition even if only to keep dGPUs from being priced out of reach of most gamers at least.
Yotsugi
Golden Member
- Oct 16, 2017
- 1,029
- 487
- 106
We don't, AMD/Intel wasting their money to please the kids with cheaper nVidia GPUs is hands down the last thing I'd ever want.
As long as they can make a fair profit then they are not wasting thier money, now are they?
Yotsugi
Golden Member
- Oct 16, 2017
- 1,029
- 487
- 106
Making a profit in dGPU with increasing tapeout costs is getting hella difficult unless you're the PC gaming's darling, being Nv.As long as they can make a fair profit then they are not wasting thier money, now are they?
And yet AMD is still making money of their dGPUs, even with low margins. BTW, dGPUs being priced out most consumer hands will end up killing off the PC gaming market aside from casual games.
Yotsugi
Golden Member
- Oct 16, 2017
- 1,029
- 487
- 106
Something meager outside of mining booms, and that's with active design reuse.
So why even bother, Intel is free money piniata.
Which is why they won't ever be, since it's not like nV has a CPU biz worth mentioning haha.
* We don't really know what the marketshare is, based off of financial numbers AMD is doing okay compared to their history, albeit it does probably have about 25% marketshare.
* AMD are already working hard on their Ryzen 3 CPU, which will be coming in about 3 months, they are also working on next gen Threadripper, and also on their 7nm+ Epyc processors. So they are in an extremely good position CPI wise. They have the process node advantage, the performance per power advantage, power per core advantage, price to performance, price to core advantage and better performance in multithreaded applications. Where its solely lacking in single threaded applications.
From the various reports Ryzen 3 should be about 18% faster than Ryzen 2 equivalents, with a combination of process node advantage, clock speeds advantages and IPC advantages. If they can squeeze out 3-4% more speed, as they are taking their time with these processors, they should be golden.
* Vega 20 is an updated architecture over Vega 10. Vega 20 was always meant to be purely for computing and the professional market and that is where its been sold. What they did is take the defective dies that couldn't be MI60 and excess MI50 inventory and converted it to Radeon 7.
Rather than have all this excess MI50 that they aren't selling that well, they created the Radeon 7 with it. So now they can get some money off of it, sell it as competition to the 2080 and clear excess inventory.
* AMD have been working on Navi for the past 3 years, we know its coming either late Q2, though most likely early Q3. They will have the process advantage over Nvidia with it and the ability to see what Nvidia has done and one up it now. I think Navi will be very competitive, I mean if you look at the RX 460/470/480/560/570/580 they were extremely competitive with Nvidia's offerings. Vega not so much, but they tried to do everything with Vega, computing, workstation and gaming and they failed at the gaming part in a big way, but in terms of pricing it was/is competitive with the 1070, 1070ti and 1080.
Considering the expense of nodes, AMD will have to make due with less designs. Considering Nvidia has a no compromise compute chip in the Gx100 series and a gaming focused gp102 series, I think it is impossible for AMD to make a compromise chip that competes with either without some sort of temporary crutch like getting to 7nm first. This is not a long terms solution.
What I believe AMD needs to do is focus. That is do a pure compute card at the high end which doesn't does not get extended to the gaming market.
And something along the lines of polaris as far as pricing and target the entire market below this? So what about the high end? Ignore it and spend that time on getting this chip as good as possible as soon as possible. E.g RX 590 or 580 on day 1 and have sufficient drivers which show almost all the potential day 1. That doesn't mean AMD has to abandon the high end.
I believe the solution lies in something AMD has abandoned mostly but it will be something they need to focus on in the future and that is crossfire and mGPU.
The thing is, AMD for the chiplet design to work, needs to get their drivers to the point where crossfire works in 90% of games. That is they need it to work enough that it isn't used as a strike against them. Because of the rising cost of designs on higher nodes, there is no way AMD will be able to keep up tit for tat with Nvidia. So what should they do, focus on less designs, save that money and spend it on software development which has been a clear weakness for them(bad initial performance). Similarly crossfire support is even worse than SLI as of late which is pretty bad to begin with. Since Nvidia has removed SLI support entirely from the GTX 1660/1660 ti and RTX 2060, its a potential vulnerability which AMD can exploit if they get their crossfire performance fixed. In addition, it would help alleviate, some of these market saturation issues. If crossfire was good and well supported across lots of games, with so many owners of Polaris 480/580, AMD would have sustainable long terms customer base as soon as they drop the price of these chips over time on top of laggard buyers. The high end lately for AMD have been expensive gambles that typically don't work out for AMD.
From what we have seen with cards like Vega and the GTX 480, making a compromise chip that does everything to target the high end market leads to products very late to market having performance more along the lines of the competitors smaller and cheaper chips.These chips are extremely intensive for resources in terms of engineers which is why they come out so late.
Trying to make a compromise chips that does it all like a GTX 480/580 is mostly a fruitless effort at this point and Nvidia already learned from those mistakes. This is because making a jack of all trades cards leads to efficiency issues and extra silicon that doesn't help gaming performance. AMD high ends cards compete with Nvidia GP104 chips but have cost to build more along the lines of GP102 chips. This is not good because it puts AMD in a position where Nvidia can price them out of the market with just a few price drops. If it wasn't for the mining boom, Vega would have been dead in the water. What we are seeing repeatedly now is a pattern where AMD comes up short, far too late and the story was largely the same with Fiji and the again with Radeon VII.
Trying to build a fermi that does it all when Nvidia has products that separately target the compute market and gaming market will be all but impossible to do in the long run without a huge R and D budget. AMD will just fall further and further behind as cost go up and dilute AMD already limited resources. Under funding a project that needs more resources simply leads to a slow death. We are witnessing this with the GPU division. This is why it was right for AMD to can and sell off imageon, and cut development on ARM server processors. All the major companies making cellphone chips vertically integrate and develop their own GPU(or buy a company out right) or use ARM's reference design. Developing custom ARM architectures are super resources heavy and AMD doesn't have the budget to compete in this market, while designing CPU and GPUs. Time to market is everything and spreading your resources across too many projects will make your products late and substandard. Cutting off these expensive projects and removing resources from the GPU division to ensure the success of Ryzen(both time to market and performance) was the correct move for the long term survival of the company. But since AMD CPU division is saved, they need to start correcting their mistakes with their GPU and be realistic with the resources because what they are doing now will simply kill their GPU division.
AMD might be proud of Radeon VII right now, but it is far too late for the card to make a splash in the professional market and data center market. This is because GP100 and GV100 have been on the market far too long and combined with CUDA, it makes it very difficult for AMD to penetrate this market. Building a super computer for examples using a Radeon VII is simply undesirable because everything about the super computing market is about time to market and all the big contract have already gone to Nvidia and long since built. AMD needed something like Radeon VII to happen 2 years ago or more. GV100 was being tested in 2016 and has specs a bit better than radeon VII on top of the Cuda advantage. At this point, the next super computing cards will be looking more for next generation beyond Volta performance and have it in prototype form by the middle of 2020. Releasing hardware that is competitive with your competitors 2 years later is not good when the super computing market is all about epeen and having the fastest super computer in the world for as long as possible. Since AMD is using 7nm as a crutch to catch up to GV100, they will always be late to market since Nvidia can deliver on 16/12nm what AMD needs 7nm to accomplish. So AMD needs a 100% compute focused chip to have a chance in the long term data center market where cost don't matter as much and to have a chance to gain a performance advantage to make up for the lack of support for Cuda. They are diluting their staff focusing on too many products with their limited budget and it is making all their products late or flawed(AMD needs to be able to deliver RX 590 performance day one on the launch of Polaris).
Like Ryzen, AMD RTG needs to make less products but use that focus to build better designs and creating solutions which increase their versatility.
What I believe AMD needs to do is focus. That is do a pure compute card at the high end which doesn't does not get extended to the gaming market.
And something along the lines of polaris as far as pricing and target the entire market below this? So what about the high end? Ignore it and spend that time on getting this chip as good as possible as soon as possible. E.g RX 590 or 580 on day 1 and have sufficient drivers which show almost all the potential day 1. That doesn't mean AMD has to abandon the high end.
I believe the solution lies in something AMD has abandoned mostly but it will be something they need to focus on in the future and that is crossfire and mGPU.
The thing is, AMD for the chiplet design to work, needs to get their drivers to the point where crossfire works in 90% of games. That is they need it to work enough that it isn't used as a strike against them. Because of the rising cost of designs on higher nodes, there is no way AMD will be able to keep up tit for tat with Nvidia. So what should they do, focus on less designs, save that money and spend it on software development which has been a clear weakness for them(bad initial performance). Similarly crossfire support is even worse than SLI as of late which is pretty bad to begin with. Since Nvidia has removed SLI support entirely from the GTX 1660/1660 ti and RTX 2060, its a potential vulnerability which AMD can exploit if they get their crossfire performance fixed. In addition, it would help alleviate, some of these market saturation issues. If crossfire was good and well supported across lots of games, with so many owners of Polaris 480/580, AMD would have sustainable long terms customer base as soon as they drop the price of these chips over time on top of laggard buyers. The high end lately for AMD have been expensive gambles that typically don't work out for AMD.
From what we have seen with cards like Vega and the GTX 480, making a compromise chip that does everything to target the high end market leads to products very late to market having performance more along the lines of the competitors smaller and cheaper chips.These chips are extremely intensive for resources in terms of engineers which is why they come out so late.
Trying to make a compromise chips that does it all like a GTX 480/580 is mostly a fruitless effort at this point and Nvidia already learned from those mistakes. This is because making a jack of all trades cards leads to efficiency issues and extra silicon that doesn't help gaming performance. AMD high ends cards compete with Nvidia GP104 chips but have cost to build more along the lines of GP102 chips. This is not good because it puts AMD in a position where Nvidia can price them out of the market with just a few price drops. If it wasn't for the mining boom, Vega would have been dead in the water. What we are seeing repeatedly now is a pattern where AMD comes up short, far too late and the story was largely the same with Fiji and the again with Radeon VII.
Trying to build a fermi that does it all when Nvidia has products that separately target the compute market and gaming market will be all but impossible to do in the long run without a huge R and D budget. AMD will just fall further and further behind as cost go up and dilute AMD already limited resources. Under funding a project that needs more resources simply leads to a slow death. We are witnessing this with the GPU division. This is why it was right for AMD to can and sell off imageon, and cut development on ARM server processors. All the major companies making cellphone chips vertically integrate and develop their own GPU(or buy a company out right) or use ARM's reference design. Developing custom ARM architectures are super resources heavy and AMD doesn't have the budget to compete in this market, while designing CPU and GPUs. Time to market is everything and spreading your resources across too many projects will make your products late and substandard. Cutting off these expensive projects and removing resources from the GPU division to ensure the success of Ryzen(both time to market and performance) was the correct move for the long term survival of the company. But since AMD CPU division is saved, they need to start correcting their mistakes with their GPU and be realistic with the resources because what they are doing now will simply kill their GPU division.
AMD might be proud of Radeon VII right now, but it is far too late for the card to make a splash in the professional market and data center market. This is because GP100 and GV100 have been on the market far too long and combined with CUDA, it makes it very difficult for AMD to penetrate this market. Building a super computer for examples using a Radeon VII is simply undesirable because everything about the super computing market is about time to market and all the big contract have already gone to Nvidia and long since built. AMD needed something like Radeon VII to happen 2 years ago or more. GV100 was being tested in 2016 and has specs a bit better than radeon VII on top of the Cuda advantage. At this point, the next super computing cards will be looking more for next generation beyond Volta performance and have it in prototype form by the middle of 2020. Releasing hardware that is competitive with your competitors 2 years later is not good when the super computing market is all about epeen and having the fastest super computer in the world for as long as possible. Since AMD is using 7nm as a crutch to catch up to GV100, they will always be late to market since Nvidia can deliver on 16/12nm what AMD needs 7nm to accomplish. So AMD needs a 100% compute focused chip to have a chance in the long term data center market where cost don't matter as much and to have a chance to gain a performance advantage to make up for the lack of support for Cuda. They are diluting their staff focusing on too many products with their limited budget and it is making all their products late or flawed(AMD needs to be able to deliver RX 590 performance day one on the launch of Polaris).
Like Ryzen, AMD RTG needs to make less products but use that focus to build better designs and creating solutions which increase their versatility.
Yotsugi
Golden Member
- Oct 16, 2017
- 1,029
- 487
- 106
Load of nonsense.
It's an accelerator, it's inherently bad at versatility.
Also Navi is gaming and gaming only.
:^)
NTMBK
Lifer
- Nov 14, 2011
- 10,448
- 5,831
- 136
The reason that SLI/Crossfire is dying is because it doesn't work well with modern graphics techniques. If you need data from the previous frame for your temporal AA solution, then it's pointless trying to do interleaved frames on two GPUs. AMD can't solve that with magic drivers. (Plus the fundamental problem that SLI does nothing to improve frame latency, only throughout.)
AtenRa
Lifer
- Feb 2, 2009
- 14,003
- 3,362
- 136
Dont thing so, they need to address the entry, low and meddle-end Professional market, much like Polaris
Yotsugi
Golden Member
- Oct 16, 2017
- 1,029
- 487
- 106
"Professional" cards are the same thing different driver.
You're paying for the driver.
AtenRa
Lifer
- Feb 2, 2009
- 14,003
- 3,362
- 136
Same chip, different product, different price.
Example,
Chip = Polaris 10
Consumer card = RX 480
Professional card = Pro WX 7100
Yotsugi
Golden Member
- Oct 16, 2017
- 1,029
- 487
- 106
It's the very same product, IIRC you can very much load the gaming drivers on WX7100.
They have already addressed the professional market, unless you've been living under a rock. It's called MI60 and MI50. These things are selling like hot cakes and AMD beat Nvidia in the professional market last quarter.
Navi is gaming only, it seems to be a scalable chip, from high end to low end.
Eh? AMD had about 12% of the enterprise market last quarter. Workstation market I haven't seen any numbers.
There just don’t seem to be enough facts out available to get any concrete sense of where AMD is going. AMD's path appears to be more challenging than Nvidia's. Doesn’t seem like there is much more that can be said, unless there are more *facts* that can be put on the table.
Yotsugi
Golden Member
- Oct 16, 2017
- 1,029
- 487
- 106
Between the obvious Vega drawbacks and their patent history, sure there is.
I dunno, x86 is a money printing machine, GPU IP just makes it even more money-printing.
To remember, graphics IP is used for much more than consumer gaming cards. Consoles and APUs come to mind. Are some saying that AMD will abandon these markets in addition to the desktop and compute, by abandoning staying competitive? Well they might as well start planning now to close shop.
When the expenditure for 7nm started, they were in a much more precarious position financially than they are now, but still the money was invested.
When the expenditure for 7nm started, they were in a much more precarious position financially than they are now, but still the money was invested.
DrMrLordX
Lifer
- Apr 27, 2000
- 22,931
- 13,014
- 136
AMD won't abandon enterprise compute markets. They will stick with consoles as well (Navi seems aimed at consoles first). Everything else is up in the air.
Why?
If consoles are on the table then desktop as well. They would have to give up any thought of competing for the high-end, but in the end, once a profit can be made, I expect them to continue producing products. In spite of the fact that Nvidia outsold them easily, the Polaris models still contributed to both revenue and profits.
A previous slide showing the costs for the various nodes gives the impression to some that this cost is repeated for each design. In reality, quite a bit of those can be shared, as just being able to "handle" that node and its rules.
Yotsugi
Golden Member
- Oct 16, 2017
- 1,029
- 487
- 106
Navi is gaming first.
Noticed the keyword?
Not consoles, or DT. Gaming.
Lisa will throw money there either way.
Navi+1 is compute focus again.
I said last quarter, you can't take the overall market that has been buying stuff over the past 10 years. Marketshare over the last quarter and last few years has been improving SIGNIFICANTLY for AMD.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 22K
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 384
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
Facebook
Twitter