NVIDIA: PS4 Not Worth The Cost

There is so much wrong info posted in this thread:

- Trinity uses Piledriver cores, a slight revision of Bulldozer. It's a big and hot and isn't a good fit for consoles (it would also be expensive given the low TDP required meaning crap yields)

- Jaguar is a brand new core microarchitecture from AMD. It's a redesign of the older Bobcat cores and promises up to 50% better performance.

- 8 Jaguar cores are being used for the PS4 as well as the new Xbox.

- The Kabini/Kaveri APU (it's the same die) uses the same Jaguar cores and 2 GCN CUs. It was designed to provide very good performance for a small TDP and great perf-per-watt (5-25W TDP for dual and quadcore versions)

- The PS4's die is actually an APU with a slightly-more-powerful 7850 GPU coupled with 8 Jaguar cores and who knows what else. The GPU and the CPU feed off of the same memory and sit right next to each other, chatting it up while mixing up the workload.

- The 8GB of GDDR5 means that both the CPU and GPU will be fed from the same, incredibly fast memory. This gives the CPU and GPU a lot more wiggle room and prevents and potential bottlenecks (which you see in even high end desktop CPUs). Because of the 8GB GDDR5, it wouldn't surprise me to see some console games having better physics performance on the console than on a PC with much better hardware.

- The ISAs aren't brand-spanking new, but it does feature goodies such as AVX2 and F16C. It lacks FMA3/4, but I'm guessing it wasn't worth the die space.

Now the nVidia portion:

Sony and Microsoft both wanted/needed x86. Obviously, nVidia can't provide that, but they could have provided them with a decent ARM SoC and a fast GPU. nVidia is right in that it would have been far too expensive for them to provide the consoles with the proper CPU they need, but that's mainly to due with nVidia still failing to provide a core microarchitecture of their own. Their current ARM SoCs use bog standard ARM core which just wouldn't be enough to power a next-gen console (you could argue that the A15s have that power, but there's still a tremendous benefit with opting to go with x86). In short, nVidia just doesn't have the engineers nor the x86 license to provide a CPU for the next-gen consoles. AMD has the CPU - though AMD stick at the high end, their low end processors are great - and they can couple that with their GCN architecture.

nVidia never had a shot at winning the next gen consoles given AMD marketing their APUs. Where nVidia could have provided a GPU, AMD is able to provide both sides of the die and do it cheaper than nVidia+CPU can do it. I don't think AMD will make a killing from these consoles off the hardware sales, but the potential advantage that AMD gains across their entire lineup of products (all of the discrete GPUs will use the same architecture as the consoles) means nVidia will have to work much much harder to equal AMD's natural performance advantage due to developers targeting their cards specifically. IMO, that's the biggest issue facing nVidia going forward. Are they going to dedicate millions and millions of dollars with a newer version of TWIMTBP just to get on an even keel with AMD? I'm not sure about that.

nVidia is going to have to win the architectural race for years to come just to break even with respect to performance with AMD, and I just don't see that happening considering how the last few years have played out. They've been pretty much neck and neck, but now the developers are heavily in AMD's favor.
 
- The Kabini/Kaveri APU (it's the same die) uses the same Jaguar cores and 2 GCN CUs. It was designed to provide very good performance for a small TDP and great perf-per-watt (5-25W TDP for dual and quadcore versions)

Shouldn't that be Kabini and Temash?

Both APUs are using Jaguar cores. However, Kaveri APU is supposed to be the successor to Trinity and Richland given that it's using Steamroller cores, and su[posedly the first implementation of AMD's HSA design.

http://www.anandtech.com/show/6201/amd-details-its-3rd-gen-steamroller-architecture

They've been pretty much neck and neck, but now the developers are heavily in AMD's favor.

Actually, that's a good sign for us gamers on the PC. We may actually see games have equal or better footing against Nvidia, who many of us criticize for sleeping with the devs and having games favor their video cards for TWIMTBP BS.

Maybe for once we'll see developers move off of PhysX and onto something like Bullet or Havok physics engines so that both video cards will have a somewhat equal footing against each other when it comes to performance wars.
 
There is so much wrong info posted in this thread:

- 8 Jaguar cores are being used for the PS4 as well as the new Xbox.

- The PS4's die is actually an APU with a slightly-more-powerful 7850 GPU coupled with 8 Jaguar cores and who knows what else. The GPU and the CPU feed off of the same memory and sit right next to each other, chatting it up while mixing up the workload.

- The 8GB of GDDR5 means that both the CPU and GPU will be fed from the same, incredibly fast memory. This gives the CPU and GPU a lot more wiggle room and prevents and potential bottlenecks (which you see in even high end desktop CPUs). Because of the 8GB GDDR5, it wouldn't surprise me to see some console games having better physics performance on the console than on a PC with much better hardware.

What is not clear is the actual speed of the DDR5, and the speed of the DDR5 to the each or all of the 8 CPUs, plus the bus width to each CPU.
 
Sorry, mistyped :D Kabini/Temash. Kabini is the notebook part while Temash is ~5W TDP at tablets. Kaveri is indeed a Steamroller core and due out around November of 2013 (which will probably be released at the Fusion/HSA conference)
 
What is not clear is the actual speed of the DDR5, and the speed of the DDR5 to the each or all of the 8 CPUs, plus the bus width to each CPU.

DDR5 is expensive, but it's also not worth opting to go with the cheaper RAM either.



Update: I've confirmed that there are 8 Jaguar based AMD CPU cores inside the PS4's APU. The CPU + GPU are on a single die. Jaguar will still likely have better performance than the PS3/Xbox 360's PowerPC cores, and it should be faster than anything ARM based out today, but there's not huge headroom going forward. While I'm happier with Sony's (and MS') CPU selection this time around, I always hoped someone would take CPU performance in a console a bit more seriously. Given the choice between spending transistors on the CPU vs. GPU, I understand that the GPU wins every time in a console—I'm just always an advocate for wanting more of both. I realize I never wrote up a piece on AMD's Jaguar architecture, so I'll likely be doing that in the not too distant future.

Serial code could potentially be a bottleneck, but most serial CPU code in games has to due with poor development in the first place (Blizzard, anyone?). Given the thread-laden architecture of the prior Xenon, I doubt having a comparatively lower IPC CPU (when compared to desktop CPUs) is going to be a hindrance here.
 
What is not clear is the actual speed of the DDR5, and the speed of the DDR5 to the each or all of the 8 CPUs, plus the bus width to each CPU.

And, we don't know what capacity each GDDR5 chip is at. When I did the math a few months ago, we're looking at most SIXTEEN GDDR5 chips at 512MB each. To my knowledge, the largest chip manufactured is 256MB GDDR5 which is common in many video cards today.

Bus width will have to be 512-bit minimum-- 64-bit x8 memory controllers.

My best bet is the one from Samsung:
4Gb per chip at 128Mx32,16K/32ms 1.5V ± 0.045V

Maximum bandwidth per chip is 7000Mbps at 0.28ns. I'm going to guess frequency is somewhere north of 5GHz, and I'm low-balling it.

Sorry, mistyped :D Kabini/Temash. Kabini is the notebook part while Temash is ~5W TDP at tablets. Kaveri is indeed a Steamroller core and due out around November of 2013 (which will probably be released at the Fusion/HSA conference)

I'm holding off on buying a laptop until Kabini is out, and hopefully soon Richland-based laptops. Kaveri is still a long ways off for me.
 
DDR5 is expensive, but it's also not worth opting to go with the cheaper RAM either.

4800MHz GDDR5 memory doesn’t seem to be too expensive, to me, when I look at the price of ATI 7850 cards, and how much GDDR5 memory comes with them.

For example, there is a “Sapphire HD 7850 1024MB” for £114.99 ex VAT (retail) and another “MSI HD 7850 OC 2048MB” for £137.49 ex VAT (retail), which is a £22.5 difference; equating to about £180 (retail) for 8GB of 4800MHz GDDR5 memory. A lot cheaper than some of the DDR3 modules.
 

It depends on what costs they take into account when they calculate the profit/loss. It could be proven that everything makes a loss at the beginning. That’s business.

The way that I look at it though; is if a single ”AMD Bulldozer FX-4 Quad Core 4100 Black Edition 3.60Ghz” can be bought for £66.66 ex VAT (retail), then discounting any R&D costs, and if there is a high silicon yield, then < $50 per PS4 chip can still give good profit. At high enough quantity & yield, it my be possible to sell a half that unit price, and still come into profit towards the end.
 
I didn't miss it. Looked fairly cooked to me. I am calling it now - end product will look and play worse than what was shown.
 
I didn't miss it. Looked fairly cooked to me. I am calling it now - end product will look and play worse than what was shown.

A more intelligent guess would be it would do better.

Beta shiit normally sucks. It's often compiled with debug flags.
 
And, we don't know what capacity each GDDR5 chip is at. When I did the math a few months ago, we're looking at most SIXTEEN GDDR5 chips at 512MB each. To my knowledge, the largest chip manufactured is 256MB GDDR5 which is common in many video cards today.

Bus width will have to be 512-bit minimum-- 64-bit x8 memory controllers.

My best bet is the one from Samsung:
4Gb per chip at 128Mx32,16K/32ms 1.5V ± 0.045V

Maximum bandwidth per chip is 7000Mbps at 0.28ns. I'm going to guess frequency is somewhere north of 5GHz, and I'm low-balling it.

You suck at math!

176GB/s confirmed bandwidth.

That's 1408 Gbits/sec!

Assume 256-bit bus (just like the HD 7850), and you get GDDR5 clocked at 5.5 GHz!

That's perfectly doable with today's chips (GTX 680, 670 AND 660 Ti ALL ship with 6 GHz effective GDDR5), and even AMD has been shipping cards with memory that fast for over two years (HD 6970), so they have plenty of experience driving those speeds.

Plus, that means you could do it with only 8 chips (each GDDR3 / GDDR5 chip is 32-bits wide), assuming you could get 8Gbit GDDR5, which makes it a stretch goal for Sony to wait for these chips to get higher density.

Unfortunately for them, there are only 2Gbit GDDR5 chips available for the foreseeable future, and since GDDR5 is low-volume (compared to desktop ram), it will not see as fast a push for greater densities. I have not seen any public announcements for 4Gbit GDDR5, and there may never be any because GDDR6 is supposed to replace it in 2014!
 
Unfortunately for them, there are only 2Gbit GDDR5 chips available for the foreseeable future, and since GDDR5 is low-volume (compared to desktop ram), it will not see as fast a push for greater densities. I have not seen any public announcements for 4Gbit GDDR5, and there may never be any because GDDR6 is supposed to replace it in 2014!

http://www.samsung.com/global/busin...t/graphic-dram/detail?productId=7824&iaId=759

Samsung has 4Gbit GDDR5 chips by the way. Maximum bandwidth is 7000Mbit/s per chip at 0.28ns access time.

2013-03-17_12-31-05.png


I was surprised when I found it when I wrote that post. And, it seems they're the only ones manufacturing it. Since they're the only manufacture I found manufacturing them, these could be the GDDR5 chips the PS4 is using. Neither is Qimonda or Hynix has them since they only manufacture at the highest 2048Mbit GDDR5 chips (256MB GDDR5 chip).

Samsung began mass production of the 4096Mbit chips in September 2012. We will probably not see them in consumer and professional video cards until late 2013 or so.

And, my math is correct as you can see below.

4096Mbit GDDR5 per chip = 512MB per chip

With 8GB of GDDR5 RAM in the PS4 (8192MB total), you need 16 4Gbit GDDR5 chips to achieve that total.

16 x 512 = 8192MB

The HD 7850 has a 256-bit bus. That's four 64-bit dual channel memory controllers.
http://images.anandtech.com/doci/5625/PitcairnArch.png

4 x 64-bit = 256-bit

That gives it 2 GDDR5 chips per memory controller, or eight chips total. With the current maximum 256MB GDDR5 chips multiplied by 8, that gives 2048MB or 2GB of GDDR5 RAM for the Radeon HD 7850.

To achieve this in the PS4, you will need to double the amount of memory controllers. Therefore, that's eight 64-bit dual channel memory controllers. At two GDDR5 memory chips per controller, that's 16 chips total. That also gives a 512-bit bus maximum.

8 x 64-bit = 512-bit

8 memory controllers x 2 GDDR5 chips = 16 GDDR5 chips x 512MB GDDR5 per chip = 8192MB of GDDR5 RAM

To do this on a 256-bit bus, you will need an 8Gbit density GDDR5 memory chip which does not exist. But, to humor you, here is the math:

4 x 64-bit = 256-bit

4 controllers x 2 GDDR5 chips = 8 GDDR5 chips x 1024MB GDDR5 per chip = 8192MB of GDDR5 RAM

There is also an alternative method I've thought about: Dual 256-bit memory controllers.

In other words, the PS4 could use two banks of 4GB GDDR5 memory each separated by two 256-bit memory controllers-- one 256-bit on the GPU and another 256-bit on the CPU within the APU SoC. I don't know if it's cost-effective or cheaper to do it that way than have a single row of 64-bit memory controllers totaling 512-bits.
 
I just don't trust them one bit :)

I don't trust myself. I am afraid that I might be led by the hype that comes with an expensive card like the nVidia Titan. The memory bandwidth of the Titan is barely double that of this new PS4 AMD gaming chipset, and it probably has features that I trust games developers will avoid using.
 
I don't trust myself. I am afraid that I might be led by the hype that comes with an expensive card like the nVidia Titan.

I wouldn't trust you either, considering you couldn't even back up any of your arguments from the previous thread; and yes, you are being lead by the hype that the PS4 is bringing. :rolleyes:

The memory bandwidth of the Titan is barely double that of this new PS4 AMD gaming chipset, and it probably has features that I trust games developers will avoid using.
Like what features, and please, give cited examples, not your opinions.
 
To achieve this in the PS4, you will need to double the amount of memory controllers. Therefore, that's eight 64-bit dual channel memory controllers. At two GDDR5 memory chips per controller, that's 16 chips total. That also gives a 512-bit bus maximum.

8 x 64-bit = 512-bit

8 memory controllers x 2 GDDR5 chips = 16 GDDR5 chips x 512MB GDDR5 per chip = 8192MB of GDDR5 RAM

To do this on a 256-bit bus, you will need an 8Gbit density GDDR5 memory chip which does not exist

See, this is why I said you suck at math - it's because you don't understand memory interfacing, and ther way you described what was on your mind was VERY CONFUSING.

Adding a second set of memory chips for each controller channel means you just have to use the Chip Select, and add another line to your ADDRESS BUS to handle double the memory space.

At no time do you need to increase the size of your DATA BUS from 256-bit to 512-bit. You can multiplex more devices without widening the bus, although these are limits to how much you can add before it's bandwidth-starved. Anyway, you don't want to double the size of the memory bus "just because" you have the chips to feed it - it comes at a huge cost to die space.

Also, thanks for the information on the 4Gbit GDDR5 devices. I did not hear a thing about them being introduced, and it cuts the devices required by Sony from 32 down to 16.
 
See, this is why I said you suck at math - it's because you don't understand memory interfacing, and ther way you described what was on your mind was VERY CONFUSING.

Adding a second set of memory chips for each controller channel means you just have to use the Chip Select, and add another line to your ADDRESS BUS to handle double the memory space.

At no time do you need to increase the size of your DATA BUS from 256-bit to 512-bit. You can multiplex more devices without widening the bus, although these are limits to how much you can add before it's bandwidth-starved. Anyway, you don't want to double the size of the memory bus "just because" you have the chips to feed it - it comes at a huge cost to die space.

Also, thanks for the information on the 4Gbit GDDR5 devices. I did not hear a thing about them being introduced, and it cuts the devices required by Sony from 32 down to 16.

I did not know about multi-plexing or chip-select, sorry. Last I read similar terms was on how DDR4 memory controllers would work. There, different banks of memory would be used on a single channel and could be switched between them. (There is a lengthy and detailed explanation somewhere, think it was PC-Watch's website.)

I didn't think GDDR5 memory controllers would be capable of that seeing that it is dual channel and have only seen 2 chips per controller, nothing more. If they can keep it to 256-bit, that'll be great since it'll reduce the complexity and cost of the console itself. 512-bit would be costly.
 
I did not know about multi-plexing or chip-select, sorry. Last I read similar terms was on how DDR4 memory controllers would work. There, different banks of memory would be used on a single channel and could be switched between them. (There is a lengthy and detailed explanation somewhere, think it was PC-Watch's website.)

I didn't think GDDR5 memory controllers would be capable of that seeing that it is dual channel and have only seen 2 chips per controller, nothing more. If they can keep it to 256-bit, that'll be great since it'll reduce the complexity and cost of the console itself. 512-bit would be costly.

AFAIK, DDR4 will be point-to-point, but I don't see why you couldn't again multiplex as many chips as you wanted on the end of a DDR4 bus drop (e.g. as long as the memory controller sees 8GB memory space, does it care if it's one 64Gbit chip, or 8x8Gbit?) I'd have to read-up on the standard to tell you for sure.
 
AFAIK, DDR4 will be point-to-point, but I don't see why you couldn't again multiplex as many chips as you wanted on the end of a DDR4 bus drop (e.g. as long as the memory controller sees 8GB memory space, does it care if it's one 64Gbit chip, or 8x8Gbit?) I'd have to read-up on the standard to tell you for sure.

Very good article to read though: http://pc.watch.impress.co.jp/docs/column/kaigai/20100816_387444.html

You'll need to translate to English though. English version (I think this would be the same):
http://www.bit-tech.net/hardware/memory/2010/08/26/ddr4-what-we-can-expect/1

If GDDR memory controller could do something like have 4 chips per 64-bit memory controller (2 chips per channel for each memory controller), they wouldn't have to resort to a large, complex 512-bit bus.
 
AFAIK, DDR4 will be point-to-point, but I don't see why you couldn't again multiplex as many chips as you wanted on the end of a DDR4 bus drop (e.g. as long as the memory controller sees 8GB memory space, does it care if it's one 64Gbit chip, or 8x8Gbit?) I'd have to read-up on the standard to tell you for sure.

I guess what I think of when I think about point-to-point is like how you can create wider bus interfaces on PCI Express and use a smart bus arbitrator (like Nvidia's NF200) to make the most of the shared bandwidth. Just like chip select on parallel data bus, there's limits of how many lines you can add before you're bandwidth-starved, but if everything downstream from the bus arbitrator thinks the signals are up to spec, the upstream PCI Express 16x controller doesn't care.

If DDR4 uses a similar interface, I see no problem with adding more devices to the bus.
 
AMD has time to languish for many years unless they really screw things up. Gamble on their future and buy stock??

Of course, NVidia is gambling their own future. I applaud pushing technology, but I fear and suspect an era of stagnation ahead
 
AMD has time to languish for many years unless they really screw things up. Gamble on their future and buy stock??

Of course, NVidia is gambling their own future. I applaud pushing technology, but I fear and suspect an era of stagnation ahead

I think Nvidia is gambling in the wrong place. At the right time (Their window now is very small as I see ARM devices leveling off really quick in the next 2-3 years). Their Tegra chips so far has been proven to be worse than their competitors and worse very late to the market. Nvidia has graphics expertise but have really nothing in the baseband, cpu etc that makes the SoC. They would have to license a lot of the stuff or come up with it on their own and they so far have proven they cant. I would much rather have Snapdragon and Exynos than Tegra. Also for all the graphics expertise they had Tegra 3 wasnt at all faster than the PowerVRs. (At least as far as I remember.)
 
Since the last thread was locked:


I think that it is AMD&#8217;s roadmap to produce better APU, which combine ATI GPU tech with AMD CPU tech. There is plenty of proof about it on the internet. Such a device is pretty much useless without high bandwidth memory. (PhysX is useless, unless it writes into the game-play memory).

I know, physics on a GPU or non-CPU device is useless, that's why it's been in use since 2006. :rolleyes:


There is plenty of proof about it on the internet.
Then post some sources backing your claims.
Just saying "there is proof on the internet" does not constitute 3rd party, cited proof.

We keep going around in circles because you refuse to do just this.
I am completely willing to believe what you say, but without another source that backs your claims, I, and no one else, is going to believe you.

So post some sources that will back all of your "bottlenecking" claims up to this point, and I will gladly admit that I am wrong, and will STFU and GTFO just for you. :)
If you won't post sources, however, then please, don't post any more and quit wasting our time with your opinions and speculations.


This isn't "distorting the facts" or "trolling", as you keep stating.
This is me saying, "put your cards on the table and let's see who actually won the game."


Deal?
 
They would have to license a lot of the stuff or come up with it on their own and they so far have proven they cant. I would much rather have Snapdragon and Exynos than Tegra. Also for all the graphics expertise they had Tegra 3 wasnt at all faster than the PowerVRs. (At least as far as I remember.)

Or the Adrenos. (Which a lot of people forget... is Radeon tech. Adreno = Radeon with the letters rearranged.) This was introduced in 2002, way before nVidia thought they could "me, too!" this market. Kinda heartbroken to say it was for the better that AMD sold it off because Qualcomm has done great things with the Snapdragon SOCs.... though what we see now with those is largely what ATI started.
 
At the end of the day execution is the only thing that matters. Yes AMD wanted to pair higher end GPUs with CPUs, and they did it but they dropped the ball on the CPU part by that time. And in the mean time Intel really brought their GPUs up in speed, still not as good but the whole deal with APUs is, saving cost and power and at a lower process node and poorer process AMD was not delivering on what should have been the strength of their APUs.
Second notice I said higher power, not highest power. Highest power is always going to be reserved discreet graphics since that's the way you can get rid of the heat and apply a lot of amps.

Also as far as adreno, ya just about every company had some mobile play back in the day and for whatever reason almost all of them sold off their interest or separated it ironically before the mobile explosion took off. AMDs lack of foresight really cost them since in the end NVidia is further ahead on mobile SOCs. Once again its all about execution. Theres a company called archos which pretty much made every all of apples hit products before apple but they didn't make the billions of dollars.
 
At the end of the day execution is the only thing that matters. Yes AMD wanted to pair higher end GPUs with CPUs, and they did it but they dropped the ball on the CPU part by that time. And in the mean time Intel really brought their GPUs up in speed, still not as good but the whole deal with APUs is, saving cost and power and at a lower process node and poorer process AMD was not delivering on what should have been the strength of their APUs.
Second notice I said higher power, not highest power. Highest power is always going to be reserved discreet graphics since that's the way you can get rid of the heat and apply a lot of amps.

Also as far as adreno, ya just about every company had some mobile play back in the day and for whatever reason almost all of them sold off their interest or separated it ironically before the mobile explosion took off. AMDs lack of foresight really cost them since in the end NVidia is further ahead on mobile SOCs. Once again its all about execution. Theres a company called archos which pretty much made every all of apples hit products before apple but they didn't make the billions of dollars.

I can see that Nvidia is further along in SOC , i can see them Kill AMD on Kabini and Temash , they just have to do some funky time travel stuff and GO NVIDIA GO!

The process might go somehting along this way:
Sony : what can you offer us and how much of an improvement can you give us for our next gen project shipping in 2013.

If Nvidia and Intel can't do it in several number of years then that says enough doesn't it?
The other thing you say I'm not so sure about is AMD dropping the ball on the CPU part, what has been proven from Bulldozer->Vishera and Trinity->Richland is that the blame does go elsewhere.
 
The Tegra 4 hasn't exactly been a hit with the OEMs, from what I'm seeing/hearing. Very few design wins thus far and, again, late to the market. This time nVidia has again opted to go with standard A15 cores but have coupled them with a non-unified GPU architecture. I actually thought the Tegra 3 was pretty good despite its relatively weak performance. What made it a great little tablet SoC was its rock bottom pricing and decent overall performance. Unfortunately, the Tegra4 isn't going to be dirt cheap and nor is it coming to market at the right time. nVidia has been touting their own LTE but that isn't going to come to market until late 2013, and that's assuming they get picked over Qualcomm or Broadcom, which is incredibly unlikely.

Personally, I'd have liked to see nVidia get a console win this time, but given their history with MS and the original Xbox, the PS4 was the only real option. I really don't think things are going to go all that well in the mobile market for nVidia in the next few years. They're late to market again and don't have anything special to offer. The influx of Chinese companies providing ARM SoCs with great performance are going to lock them out of the bottom end as well.
 
I dont think the Tegra 3 was cheap because nvidia wanted it to be or designed it that way. It was cheap because no one bought it, it was late and slower than everything else. Tegra 4 seems to be a repeat. I wouldnt trust nvidia over broadcomm or qualcomm for LTE chipset. I also agree that nvidia will probably get locked out of the bottom end.
 
The Tegra 4 hasn't exactly been a hit with the OEMs, from what I'm seeing/hearing. Very few design wins thus far and, again, late to the market. This time nVidia has again opted to go with standard A15 cores but have coupled them with a non-unified GPU architecture. I actually thought the Tegra 3 was pretty good despite its relatively weak performance. What made it a great little tablet SoC was its rock bottom pricing and decent overall performance. Unfortunately, the Tegra4 isn't going to be dirt cheap and nor is it coming to market at the right time. nVidia has been touting their own LTE but that isn't going to come to market until late 2013, and that's assuming they get picked over Qualcomm or Broadcom, which is incredibly unlikely.

Personally, I'd have liked to see nVidia get a console win this time, but given their history with MS and the original Xbox, the PS4 was the only real option. I really don't think things are going to go all that well in the mobile market for nVidia in the next few years. They're late to market again and don't have anything special to offer. The influx of Chinese companies providing ARM SoCs with great performance are going to lock them out of the bottom end as well.

I dont think the Tegra 3 was cheap because nvidia wanted it to be or designed it that way. It was cheap because no one bought it, it was late and slower than everything else. Tegra 4 seems to be a repeat. I wouldnt trust nvidia over broadcomm or qualcomm for LTE chipset. I also agree that nvidia will probably get locked out of the bottom end.

I asked in the Mobile Phones forum this question: Why aren't there a lot of Tegra and Intel Atom-based smartphones?
http://hardforum.com/showthread.php?t=1751361

Curiously, I have not seen any devices announced that use Tegra 4 outside of Nvidia. Most of the talk nowadays are Qualcomm's Snapdragon 600 and 800 SoCs and Samsung's Exynos 5. Apple, well, they're Apple and only their devices use the Ax line.

Tegra 4? Not any device I've heard or read recently that I can recall. You would think that an OEM would use it given it has a seemingly powerful GPU and some interesting features. But, it doesn't seem to be the case.

Majority of the Tegra 3 devices I've seen are in tablets and of course the Ouya. Smartphones? Not a whole lot. Tegra is like the Windows Phone 8 of the mobile SoC world-- not a lot of OEMs have it and not a large marketshare.

What is Nvidia doing wrong with Tegra where Qualcomm, Apple, and Samsung have succeeded in? Besides being late to market, slow, lack of LTE on chip, what else are they missing?

They can't make x86 processors because if I recall, they were denied licensing or was it they failed to purchase a company that makes x86 processors like Via? I don't remember the whole Nvidia-x86 situation now.

If they can't make it in the mobile market, then that leaves them with 3 options...
  • Manufacture their own Android tablet, smartphone, and gaming device (i.e.- Shield)
  • Desktop and workstation graphics cards
  • GPGPU compute server solutions (i.e.- Tesla
 
I dont think the Tegra 3 was cheap because nvidia wanted it to be or designed it that way. It was cheap because no one bought it, it was late and slower than everything else. Tegra 4 seems to be a repeat. I wouldnt trust nvidia over broadcomm or qualcomm for LTE chipset. I also agree that nvidia will probably get locked out of the bottom end.

I'm just curious where you guys get this stuff. Of all my tablets my nexus 7 is hands down the fastest. Faster than my phones to. Sgs2 and 3. IPad 3, galaxy tab 2. I have been very pleased with it, it gets the most use lasts all day on Wi-Fi browsing and ask other tasks I use it for.

I really don't get it. Do you guys just repeat what people say or do you own and use the things you trash talk? Just curious.
 
I personally think nvidia gets greedy with their prices, rumor has it the next nexus 7 won't use a tegra setup. This I think is also why Sony went a different direction with PS4. They must like to burn bridges I know they did with MS.
 
good news for AMD imo they could use the boost..competition is always better for us
 
Back
Top