I'm new be nice
Member
Yeah I think the 780ti will be my last top tier purchase. The performance increase over the midrange cards just isn't worth the money compared to having a bit more cash on hand for an entire architecture change.
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
AMD introduces "Radeon Pro Duo" 16TFLOPS graphics card
- Thread starter Caayn
- Start date
Yeah I think the 780ti will be my last top tier purchase. The performance increase over the midrange cards just isn't worth the money compared to having a bit more cash on hand for an entire architecture change.
//DEVIL//
Member
is there any single self respected motherboard that doesn't have 2 slots ?
I mean, if someone is buying a PC aimed for gaming and then he buys a motherboard that has 1 slot then lolol
If they pulled this out at the end of last year it would have been great timing but what is the point when their Polaris and Nv's Pascal architectures are right around the corner?
I'd rather one gpu that gives me even 70% of the Pro Duo's power then one pcb that is xfire and needs driver profiles to run properly.
I'd rather one gpu that gives me even 70% of the Pro Duo's power then one pcb that is xfire and needs driver profiles to run properly.
In the case of multi GPU 2x4=4.
Despite it being pointed out everywhere this card isn't really being targeted at consumers. It is aimed at VR developers mostly. The Anandtech article has an interesting take on their motivations behind the card.
Vulcano's assistant
Banned
People who put looks firsts and only want mini ITX cases
Everything uses a lot of juice as you increase bandwidth.
Look for benchmarks of 2 x Fury X in Crossfire. This "Radeon Pro Duo" will perform almost identically. Although we don't know the Duo's GPU clock yet, which could be lower than a Fury X for thermal reasons.
Either way, Pascal is a code name for a GPU family not yet released, so who knows? Realistically Pascal will probably bring ~40% improvement over Maxwell. The only products in the Pascal range that will have any shot at performing similarly to the Radeon Pro Duo (in games where Crossfire works properly) will be the enthusiast level cards like Titan or anything else based on GP100, which probably wont be arriving until 2017 anyway.
It strikes me that the Radeon Pro Duo is really aimed at workstation customers, a bit like Nvidia's Titan Z.
Everything uses a lot of juice as you increase bandwidth.
Well OK. This can be a serious problem with DRAM that's stacked on the GPU though, not so much because of power draw, but the heat that's created.
OmegaDL50
Member
Micro ATX.
Considering the wonkiness of some SLI / Crossfire configurations having issues and not performing as well as single card setups. Generally a single more powerful card is preferable anyways.
dr_rus
Member
AMD naming scheme
2x 7970 -> 7990
2x R9 290x -> R9 295 X2
2x Fury X -> Pro Duo
Keep 'em guessing.
roadrunn3r
Member
Yes and yes, did you read what I was replying to?
Rinãã¨ã
Banned
edit: this comment was a mistake
lol
lol
Caayn
Member
Micro-ATX can have multiple pcie slots, my MSI Z97m board has room for two two slots pcie cards.
Thanks for sharing.Rinさとり;198374080 said:
I'm in the same boat. The 980 Ti for ~$500 is a good deal I suppose (if you can find it for that) but like you, I also have a 780 and I paid something in the high 600s. It was really great for a while but in the long run, I don't feel like it was worth getting over a 770 and then a 970 upgrade.
I think the better move for value is just buy midrange and sell every cylce, or maybe every other cycle.
It's not always about looks. I live in a small apartment in NYC with no space for a PC desk so I put together a mini ITX build with a 980, 3770k, and 16 GB of RAM and connected it to my Pansonic plasma. Sometimes u just have to work with what u have.
Rinãã¨ã
Banned
Your welcome :>
Par for the course then. XD
lyrick
Member
Just making sure you know that your comparison was about as nonsensical as comparing an AMD 7750 to a GTX 280.
Maybe even more so as the 280 didn't start out as a lower mid-tier card.
If you compare Both companies within competitive performance brackets at competitive time frames (say 980ti vs Fury X), both products have very similar die sizes ~600mm2, similar power consumption 250-275W, and similar performance depending on title.
You could make the case that Nvidia uses less slightly larger cores to do the job 3K v 4K, but since they both take up the same amount of silicon it really doesn't matter.
Horsemama1956
Member
These cards are great if you have a little patience and can wait a little bit for Crossfire/SLI support(if it's not already there) and you like spending a bunch of money. AMD has made great strides in these single dual gpu cards.
Personally they are not for me as I don't bother with anything over midrange any more, but they have a pretty die hard following.
Personally they are not for me as I don't bother with anything over midrange any more, but they have a pretty die hard following.
Nikodemos
Member
Regular Crossfire is still shitty, though it no longer has the tremendous issues it had during the 6990 era. This card is mostly for LiquidVR purposes.
Good luck convincing devs to go the extra mile in that regard, multi GPU must be extremely difficult to implement and is very niche. I'm all for devs to cater to enthusiasts within reason but I don't foresee wide multi GPU support in DX12 games.
LordOfChaos
Member
True enough for gaming performance, but to explain the difference, FLOPS are attributed to compute, and Nvidia really cut out a lot of the compute performance of their consumer level cards in the last few architectures.
You can argue it two ways, they do have more efficient gaming performance for it, but less compute potential.
i.e, why Apple chose an old inefficient GCN 1.0 part for the rMBP 15 --
Even Intels Iris Pro beats Nvidias 750M-860M at that!
That also makes the "pro" part of the dual chip AMD card make sense, there's a lot of compute performance on one card, for upper end consumer pricing rather than upper end professional pricing which would run even more.
roadrunn3r
Member
There was no comparison at all, again I replied to someone doubting that TFLOPS mean different things for Nvidia and AMD and it's not directly comparable. Do I need to make a drawing next time?
dr_rus
Member
True enough for gaming performance, but to explain the difference, FLOPS are attributed to compute, and Nvidia really cut out a lot of the compute performance of their consumer level cards in the last few architectures.
You can argue it two ways, they do have more efficient gaming performance for it, but less compute potential.
i.e, why Apple chose an old inefficient GCN 1.0 part for the rMBP 15 --
Even Intels Iris Pro beats Nvidias 750M-860M at that!
That also makes the "pro" part of the dual chip AMD card make sense, there's a lot of compute performance on one card, for upper end consumer pricing rather than upper end professional pricing which would run even more.
It's a common knowledge that NV's OpenCL performance is bad because they are pushing CUDA instead. When you look at the bigger picture the compute on NV's h/w isn't nearly as bad as you make it sound:
From Anandtech, blue is R9 380, red is GTX960.
From Anandtech, blue is R9 390, red is GTX970.
Also in you comparison a lot is dependent on what memory type both 650M and 750M are running as this can be both DDR3 and GDDR5 with three times the bandwidth difference between them. M370X on the other hand is GDDR5 only.
Schnozberry
Member
On the plus side, it's as fast as two Fury X's crossfired.
On the down side, it works like two Fury X's crossfired.
On the down side, it works like two Fury X's crossfired.
Coulomb_Barrier
Member
Yeah this Radeon Pro Duo is pretty much sure to be the fastest single card this year, unless NV bring out a dual-GPU card. I don't see that happening though.
Looks like NV will only debut lower power next-gen cards this year like AMD.
Possumowner
Member
Well that's a helluva price.Let's hope for the price it runs like Usain Bolt
It's right there in the article linked in the OP.
Thanks. Can't read the article in detail right now. But it appears to be for VR content creation.
cyen
Member
Unfortunably it seems thats the case, no hbm2 at least on AMD and i presume NV as well. Smaller cores this year and the big guns in 2017.
dr_rus
Member
On the even downer side it costs $200 more than two Fury X crossfired.
n0razi
Member
Thats what i mean... drop $1500 on a GPU, $300 on a motherboard, and get a cheapo $150 SSD? I would expect an M2 950 Pro at the minimum. Every $80 budget SSD from a year ago does 500MB/s. Its not that impressive anymore
We have had 1000, 1200+ watt PSUs for years now, its not that crazy.
It's a common knowledge that NV's OpenCL performance is bad because they are pushing CUDA instead. When you look at the bigger picture the compute on NV's h/w isn't nearly as bad as you make it sound:
From Anandtech, blue is R9 380, red is GTX960.
From Anandtech, blue is R9 390, red is GTX970.
Also in you comparison a lot is dependent on what memory type both 650M and 750M are running as this can be both DDR3 and GDDR5 with three times the bandwidth difference between them. M370X on the other hand is GDDR5 only.
I really wonder about the CUDA performance of AMD GPU's.
FYI, I'm talking about this.
Dee Dah Dave
Member
I was hoping this would be positioned as a genuine "980 Ti Killer". A competitively priced high end gaming GPU.
That doesn't look to be the case and another wasted opportunity from AMD.
That doesn't look to be the case and another wasted opportunity from AMD.
Netherscourge
Banned
I mostly agree.
However, Titan Z NOW sells for around $1800 to $2300, depending on where you shop, so the savings from AMD's product isn't really $1500. It's more like $300.
It's good, but not great.
Anything above $1000 is still a tough sell, no matter how powerful it is. Staying below $1000 should be the universal goal for these Top-tier GPUs.
Doesn't work that way. GPUs in that performance category were released less than 1 year ago. These companies will usually leave nearly 2 years before replacing a product at a particular price/performance level.
The thing looks nuts, $1500 though... Why not just crossfire some Fiji cards and call it a day? Same tech, just takes two slots instead of one..
I remember when I bought the 7950GX2 (dual 7900 GT card) it cost me $600 and it was nvidias top of the line dual GPU card (like this card) man prices have changed since 2006.
I remember when I bought the 7950GX2 (dual 7900 GT card) it cost me $600 and it was nvidias top of the line dual GPU card (like this card) man prices have changed since 2006.
E-Cat
Member
I believe this slide from yesterday's presentation is key to understanding AMD's ambitions going forward (2017-2019 timeframe):
On the surface it's a bit confusing, with the Pro Duo seemingly running against this philosophy of increasing price-performance. However, in the pcper interview Koduri mentioned the extension of the dual-GPU strategy to smaller dies, which is where the sweet spot lies--thus keeping in line with Moore's Law like leaps in price-perf.
Really excited for Vega/Navi!
On the surface it's a bit confusing, with the Pro Duo seemingly running against this philosophy of increasing price-performance. However, in the pcper interview Koduri mentioned the extension of the dual-GPU strategy to smaller dies, which is where the sweet spot lies--thus keeping in line with Moore's Law like leaps in price-perf.
Really excited for Vega/Navi!
LordOfChaos
Member
For what it's worth, to their credit, Apple ships the GDDR5 models and has often favored faster GPU memory.
The end memory bandwidth ends up being 80.3 GB/s vs 72 GB/s, which doesn't by itself explain that large of a difference. Nvidias lack of care for OpenCl partly does, though it's also true they cut DP performance out of their consumer line.
The Iris Pro by the way does better than the GDDR5 750M with its 50GB/s eDRAM cache plus 20odd GB/s system memory bandwidth assuming it has full access. So I don't think bandwidth explains much in that chart.