|
If AMD sells even a thousand of these things they've made a damned mint that would easily show up in Q1 2017 earnings. It'd be so drat weird for the Radeon Pro SSG to turn their GPU division profitable again, but AMD has been kind of borderline already so tipping over into a few million worth of black may not be too much of a task. And of course, once they've bought into the Radeon SSG, AMD would be ready to sell them processors...boards...hell, the whole drat complex.
|
#
?
Jul 27, 2016 16:42
|
|
|
|
| # ? May 22, 2024 03:42 |
|
|
Fallorn posted:Oil and gas is will be on top of this also. Image analysis with really large datasets will also be neat to do. Question is if it'll be better or worse than just using a server bank of Xeons and a server bank of RAM.
|
|
#
?
Jul 27, 2016 16:57
|
|
|
Yes, it will. MIT published a paper last year that demonstrated that with sufficiently large datasets, flash memory is as good as DRAM, if not better. Data is manipulated directly on the disks, as opposed to being copied out to a much smaller pool of system memory, manipulating the data, and then writing it all back to the disk. Proceeding to put the flash storage right up against GPU-accelerated compute, then, should provide great dividends, as the recent proof-of-concept seems to demonstrate. http://news.mit.edu/2015/cutting-cost-power-big-data-0710 quote:The researchers also presented experimental evidence showing that, if the servers executing a distributed computation have to go to disk for data even 5 percent of the time, their performance falls to a level that’s comparable with flash, anyway. SwissArmyDruid fucked around with this message at 17:15 on Jul 27, 2016 |
|
#
?
Jul 27, 2016 17:08
|
|
|
That's really cool; hope they don't screw things up.
|
|
#
?
Jul 27, 2016 17:21
|
|
|
I am curious as to what this does to the SSD lifespan with a load that would make use of such storage. I also wonder what the performance could scale to with Intels new memory coming down the pipe. The next few years are going to be exciting for memory alone. This SSG though could really become a big niche for AMD not only in this server crowd, but say in workstation laptops and such. I sometimes like to work on video not on my main desktop and my laptop currently chokes even on the 1440P stuff my GoPro 3 spits out. To be able to work on 4K-8K video on a laptop SSG would be pretty freaking incredible. Could their APU's possible benefit from this as well?
|
|
#
?
Jul 27, 2016 17:24
|
|
|
That might be the first smart thing AMD's done in a while. But how much faster is that supposed to be over using pci-e SSDs?
|
|
#
?
Jul 27, 2016 18:27
|
|
|
It might be slower, but enable Prediction Modelling on datasets larger than 12gb (or whatever the largest amount of vram is on the card with the largest amount). Instead of banging data through the PCIe bus the card can see all of it at all times. Also the very impressive 8k video editing preview which is impossible through a PCIe bus at any acceptable level of lag.
|
|
#
?
Jul 27, 2016 18:37
|
|
|
EdEddnEddy posted:I am curious as to what this does to the SSD lifespan with a load that would make use of such storage. I also wonder what the performance could scale to with Intels new memory coming down the pipe. XPoint is supposed to be able to scale up to near DRAM bandwidth and latency but the launch stuff, particularly for 'affordable' SSD's, won't come close to that if some of the benches they gave a while back are any indication. At least not at launch. The DIMM mounted stuff may be a different story. EdEddnEddy posted:Could their APU's possible benefit from this as well? edit: \/\/\/\/\/Why would you think they couldn't feed the chip enough power based on the die size? 200mm+ die sizes really aren't all that small. And a A10-7860K Godavari will pull a bit north of 220W without issue through the socket if you overclock them. Something like the old Durons (100mm or so ) would be small. I don't remember exactly how much power those things would pull when you overclocked them but I don't think they were power limited at all either. PC LOAD LETTER fucked around with this message at 20:52 on Jul 27, 2016 |
|
#
?
Jul 27, 2016 18:58
|
|
|
PC LOAD LETTER posted:I think this is the sort of thing that is aimed squarely at big data number crunching and their APU's are just going to be too power and die limited to have the processing oomph to make much use of it too. Maybe if we start seeing practical multi die APU's that have a fairly powerful GPU, CPU, and SSD/XPoint cache all mounted together on one interposer. But who knows when we'll see something like that. Years away at least it seems to me right now. The problem here as I gather is power delivery, as you can only feed a tiny chip so much power through the motherboard.
|
|
#
?
Jul 27, 2016 19:29
|
|
|
Sounds like a good thing if they are having trouble at GloFo... http://www.overclock3d.net/news/gpu_displays/amd_will_be_creating_future_products_with_samsung_as_well_as_global_foundries/1
|
|
#
?
Jul 28, 2016 17:56
|
|
|
I'm know almost sure that the differentiation between Greenland and Vega (Greenland is confirmed to be not Vega 10 or 11 exactly) is that Greenland is this GPU+SSD+HBM2 concept integrated into a coherent fabric with Naples CPUs. If Fiji is running the gently caress away with 8K video editing merely by giving it very close access to decent M.2 SSDs, a dual Vega using better interconnect speeds would be mindblowing. That and it allows AMD to sell you an entire package deal you'd keep coming back to for compatibility reasons (they do CPUs, memory, SSDs, and GPUs)
|
|
#
?
Jul 28, 2016 18:39
|
|
|
mobby_6kl posted:That might be the first smart thing AMD's done in a while. But how much faster is that supposed to be over using pci-e SSDs?
|
|
#
?
Jul 29, 2016 03:51
|
|
EdEddnEddy posted:Sounds like a good thing if they are having trouble at GloFo... http://www.overclock3d.net/news/gpu_displays/amd_will_be_creating_future_products_with_samsung_as_well_as_global_foundries/1 Well, afaik amd is still solely using Glofo. No samsung stuff coming in the next 6-9 months, and tsmc would take a huge commitment since they are already using '14nm'
|
|
|
#
?
Jul 30, 2016 16:26
|
|
|
Rumor mill is saying AMD will be launching a $100 X8K 8 cat core +2CU processor for AM4 as an alternative to Zen. Speculation is that it beats the FX 8350 for half the power consumption. Supposed specs quote:AM4 Socket Die shot  Dresdenboy, the guy who has done a lot of analysis on Zen, thinks it either an incredibly sophisticated fake, or it's very real.
|
|
#
?
Aug 3, 2016 00:03
|
|
|
How close would this be to the PS4 refresh (the neo) that's rolling out this christmas?
|
|
#
?
Aug 3, 2016 00:27
|
|
|
I think unlikely. Only 2 CUs isn't going to cut it. This looks to be the AMD equivalent of an Intel chip with integrated graphics, but not an APU. edit: I concede that this *could* go into a PS Neo, but only if it's paired with a discrete GPU. Socket AM4, though, that probably means not. If they don't BGA whatever powers the PS Neo, I'll eat my hat. SwissArmyDruid fucked around with this message at 09:14 on Aug 3, 2016 |
|
#
?
Aug 3, 2016 01:07
|
|
|
That TDP is really high for what looks like a Kabini replacement.
|
|
#
?
Aug 3, 2016 01:31
|
|
|
Tanreall posted:That TDP is really high for what looks like a Kabini replacement. It has double the cores and is clocked 50% higher just to start, 65W TDP doesn't seem that far fetched. It's also running GCN3 so it's likely this is a derivative of Excavator, which puts at least a generation apart from Kabini. With theoretical performance exceeding an FX 8350, I'm not sure this is meant as a Kabini replacement, rather it's meant as the last hurrah of performance Construction Cores with the bare minimum CU attached to it so it can display the desktop. Of course grain of salt and everything so eh.
|
|
#
?
Aug 3, 2016 03:19
|
|
|
incoherent posted:How close would this be to the PS4 refresh (the neo) that's rolling out this christmas? GPU wise it wouldn't be close to the original PS4 let alone the Neo, CPU wise it would easily dust the 8 Jaguar cores at 2.1 ghz. So not really comparable.
|
|
#
?
Aug 3, 2016 03:35
|
|
|
It should be an evolution of the Puma+ cores if it's actually part of the Cat family of processors. Which is why I think it's odd that it's TDP is so high.
|
|
#
?
Aug 3, 2016 05:10
|
|
|
AMD to use Rambus DDR4 PHY? I guess this means 3200Mhz DDR4 standard for Zen? I've seen people read into this way than it needs to be though.Good news for the APUs however!
|
|
#
?
Aug 7, 2016 17:13
|
|
|
Well THAT'S not a good sign.
|
|
#
?
Aug 7, 2016 17:51
|
|
|
SwissArmyDruid posted:Well THAT'S not a good sign. How exactly? This just seems to be Rambus's DDR4 PHY design, not the usage of actual Rambus DRAM. This just seems to enable better stock frequencies support for Zen.
|
|
#
?
Aug 7, 2016 18:41
|
|
|
it's the physical interface that's licensed -- Intel uses Synopsys' DDR4 PHY in the Haswell Xeon and 100-series chipsets.
|
|
#
?
Aug 7, 2016 18:54
|
|
|
New Zen roadmap leaks. I have come to discover that I am actually surprisingly okay with the idea of AMD re-using the old ATI All-in-Wonder branding for APUs. Usual WCCFT caveats apply. http://wccftech.com/amd-roadmap-2016-2017-leaked-zen/FaustianQ posted:How exactly? This just seems to be Rambus's DDR4 PHY design, not the usage of actual Rambus DRAM. This just seems to enable better stock frequencies support for Zen. They're only JUST getting the DDR4 PHY sorted now? They've been touting DDR4 for over a year now. 
|
|
#
?
Aug 9, 2016 20:36
|
|
|
Tapeouts of a completely new architecture are kind of a hilarious clusterfuck
|
|
#
?
Aug 10, 2016 01:59
|
|
|
This is not just a hilarious clusterfuck, this is "Jesus H. Christ on a pogo stick with Mary and Judas, AMD." Also, acknowledge my idea of using "All-in-Wonder" again, but this time to flog APUs, damnit. 
|
|
#
?
Aug 10, 2016 02:12
|
|
|
SwissArmyDruid posted:acknowledge my idea of using "All-in-Wonder" again, but this time to flog APUs, damnit. 
|
|
#
?
Aug 10, 2016 02:16
|
|
|
SwissArmyDruid posted:New Zen roadmap leaks. I have come to discover that I am actually surprisingly okay with the idea of AMD re-using the old ATI All-in-Wonder branding for APUs. Usual WCCFT caveats apply. http://wccftech.com/amd-roadmap-2016-2017-leaked-zen/ I highly doubt it's an issue of sorting out the PHY now, rather all the work with Rambus was done before, they're announcing it now, and the whole license renewal is merely formality. Like everything's been handled for awhile now, rather this is just paperwork to formalize it now surfacing.
|
|
#
?
Aug 10, 2016 03:08
|
|
|
SwissArmyDruid posted:
I'm surprised they haven't tarnished the Opteron and FirePro names to market a business version of APUs
|
|
#
?
Aug 10, 2016 13:12
|
|
|
SwissArmyDruid posted:Also, acknowledge my idea of using "All-in-Wonder" again, but this time to flog APUs, damnit. AIW cards with the hardware encoders for game streamers. MAX FPS REVOLUTION
|
|
#
?
Aug 10, 2016 15:44
|
|
|
Welp, Mac's with Polaris are on the way... among other tech that PC has had for years now.
|
|
#
?
Aug 10, 2016 16:37
|
|
|
EdEddnEddy posted:Welp, Mac's with Polaris are on the way... among other tech that PC has had for years now. The Mac Pro has AMD GPUs in it too. Apple is REALLY far behind on Mac hardware now. The ~~new~~ Mac Pro is literally untouched since the launch just shy of 3 years ago making it probably the worst purchase of a workstation you can make now.
|
|
#
?
Aug 10, 2016 17:18
|
|
|
At least the power efficiency of Polaris will actually fit one of the stronger use cases for an otherwise-overpriced brand.
|
|
#
?
Aug 10, 2016 17:18
|
|
|
FuturePastNow posted:I'm surprised they haven't tarnished the Opteron and FirePro names to market a business version of APUs http://www.amd.com/en-us/products/graphics/workstation/firepro-3d/apu
|
|
#
?
Aug 10, 2016 18:27
|
|
|
Potato Salad posted:At least the power efficiency of Polaris will actually fit one of the stronger use cases for an otherwise-overpriced brand. Over priced is very market dependent. Scandinavia laptop market is crazy overpriced, except for Apple laptops which are very reasonable (as reasonable as high-end laptops are).
|
|
#
?
Aug 10, 2016 18:28
|
|
|
Engineering samples of Zen @2.8 to 3.05Ghz vs i7 4790K @ 3.6Ghz Same profile of the guy who had leaked (?) scores of the RX480, so highly legit. 12.75% difference in score, for ~18% difference in clockspeed, so okay I guess?
|
|
#
?
Aug 10, 2016 19:00
|
|
|
FaustianQ posted:Engineering samples of Zen @2.8 to 3.05Ghz vs i7 4790K @ 3.6Ghz Man, you got me all excited because I thought it was a CPU-bottlenecked task like encoding video, SQL, or photoshop filters, but I don't think AotS is likely to be representative of general purpose computing.
|
|
#
?
Aug 10, 2016 19:09
|
|
|
Twerk from Home posted:Man, you got me all excited because I thought it was a CPU-bottlenecked task like encoding video, SQL, or photoshop filters, but I don't think AotS is likely to be representative of general purpose computing. I'd hope for more too, but I'll grasp the straws the AMD engineers throw out there. 8C/16T seems acceptable for ~400$ if it's going to match or barely beat an i7-4790K.
|
|
#
?
Aug 10, 2016 19:20
|
|
|
|
| # ? May 22, 2024 03:42 |
|
|
Well, just being vaguely competitive in gaming is pretty impressive for AMD, sadly. They're 23% behind the i7-4790 with a 11% clock speed advantage right now. Measured in the very unscientific environment of AotS at least, the Zen ES' performance per clock is 1.76 points/MHz compared to the average Vishera performance per clock of 1.08 points/MHz. That's 63% more performance per clock, which is... kind of loving incredible. edit: Apparently, if it has to do with computers, Zen is always spelled with an X, even when it isn't. Sinestro fucked around with this message at 20:26 on Aug 10, 2016 |
|
#
?
Aug 10, 2016 19:28
|
|
