NVIDIA's GeForce GTX Titan, Part 1: Titan For Gaming, Titan For Compute
by Ryan Smith on February 19, 2013 9:01 AM ESTTitan For Compute
Titan, as we briefly mentioned before, is not just a consumer graphics card. It is also a compute card and will essentially serve as NVIDIA’s entry-level compute product for both the consumer and pro-sumer markets.
The key enabler for this is that Titan, unlike any consumer GeForce card before it, will feature full FP64 performance, allowing GK110’s FP64 potency to shine through. Previous NVIDIA cards either had very few FP64 CUDA cores (GTX 680) or artificial FP64 performance restrictions (GTX 580), in order to maintain the market segmentation between cheap GeForce cards and more expensive Quadro and Tesla cards. NVIDIA will still be maintaining this segmentation, but in new ways.
| NVIDIA GPU Comparison | ||||||
| Fermi GF100 | Fermi GF104 | Kepler GK104 | Kepler GK110 | |||
| Compute Capability | 2.0 | 2.1 | 3.0 | 3.5 | ||
| Threads/Warp | 32 | 32 | 32 | 32 | ||
| Max Warps/SM(X) | 48 | 48 | 64 | 64 | ||
| Max Threads/SM(X) | 1536 | 1536 | 2048 | 2048 | ||
| Register File | 32,768 | 32,768 | 65,536 | 65,536 | ||
| Max Registers/Thread | 63 | 63 | 63 | 255 | ||
| Shared Mem Config |
16K 48K |
16K 48K |
16K 32K 48K |
16K 32K 48K |
||
| Hyper-Q | No | No | No | Yes | ||
| Dynamic Parallelism | No | No | No | Yes | ||
We’ve covered GK110’s compute features in-depth in our look at Tesla K20 so we won’t go into great detail here, but as a reminder, along with beefing up their functional unit counts relative to GF100, GK110 has several feature improvements to further improve compute efficiency and the resulting performance. Relative to the GK104 based GTX 680, Titan brings with it a much greater number of registers per thread (255), not to mention a number of new instructions such as the shuffle instructions to allow intra-warp data sharing. But most of all, Titan brings with it NVIDIA’s Kepler marquee compute features: HyperQ and Dynamic Parallelism, which allows for a greater number of hardware work queues and for kernels to dispatch other kernels respectively.
With that said, there is a catch. NVIDIA has stripped GK110 of some of its reliability and scalability features in order to maintain the Tesla/GeForce market segmentation, which means Titan for compute is left for small-scale workloads that don’t require Tesla’s greater reliability. ECC memory protection is of course gone, but also gone is HyperQ’s MPI functionality, and GPU Direct’s RDMA functionality (DMA between the GPU and 3rd party PCIe devices). Other than ECC these are much more market-specific features, and as such while Titan is effectively locked out of highly distributed scenarios, this should be fine for smaller workloads.
There is one other quirk to Titan’s FP64 implementation however, and that is that it needs to be enabled (or rather, uncapped). By default Titan is actually restricted to 1/24 performance, like the GTX 680 before it. Doing so allows NVIDIA to keep clockspeeds higher and power consumption lower, knowing the apparently power-hungry FP64 CUDA cores can’t run at full load on top of all of the other functional units that can be active at the same time. Consequently NVIDIA makes FP64 an enable/disable option in their control panel, controlling whether FP64 is operating at full speed (1/3 FP32), or reduced speed (1/24 FP32).
The penalty for enabling full speed FP64 mode is that NVIDIA has to reduce clockspeeds to keep everything within spec. For our sample card this manifests itself as GPU Boost being disabled, forcing our card to run at 837MHz (or lower) at all times. And while we haven't seen it first-hand, NVIDIA tells us that in particularly TDP constrained situations Titan can drop below the base clock to as low as 725MHz. This is why NVIDIA’s official compute performance figures are 4.5 TFLOPS for FP32, but only 1.3 TFLOPS for FP64. The former is calculated around the base clock speed, while the latter is calculated around the worst case clockspeed of 725MHz. The actual execution rate is still 1/3.
Unfortunately there’s not much else we can say about compute performance at this time, as to go much farther than this requires being able to reference specific performance figures. So we’ll follow this up on Thursday with those figures and a performance analysis.
Facebook
Twitter
RSS
157 Comments
View All Comments
CeriseCogburn - Thursday, February 21, 2013 - link
Congratulations to all you whining amd fanboy freaks who lost the POINT of the rebuttal....Some amd fruitcake OWS idiot squealed profits are going through the roof (jhonny dough boy)...
It was pointed out AMD is a failing, debt ridden, dying, sack of in the red losses loser company just about blwon to bankruptcy smithereens...
But then when do amd fanboys pay attention to their helping hand demise of their big fave AMD ?
Answer: NEVER
Reason: They constantly beg and bag for lower pricing and squeal amd does just that every time. Then they wait for nVidia to DESTROY amd pricing with awesome nVidia hardware... at that point they take their pauper pennies and JUMP on that AMD card-- raping the bottom line into more BILLIONS of losses for amd, while they attack nVidia and scream corporate profit...
LOL
Moral of the history and current idiocy of the flapping lipped fools who get it 100% incorrect while they bloviate for their verdetrol master amd.
The amd fanboys have just about destroyed amd as a going venture...
I congratulate the clueless FOOLS for slaying the idiot crash monkey lousy video card junk producer amd.
Scott586 - Thursday, February 21, 2013 - link
Wow TheJian, shut up already. We stopped reading 1/2-way thru the 2nd sentence.Spunjji - Thursday, February 21, 2013 - link
Thanks for taking what was a relevant comment correcting the previous poster and turning it into an inaccurate, ill-informed vitriolic rant about politics. That was... unprecedented.CeriseCogburn - Sunday, February 24, 2013 - link
I enjoyed it, a lot. It was not however unprecedented, mr liar.As I so clearly recall, very recently on another little review here, the left wing raging amd fan OWS protester went on a wild diatribe cursing out those who dared not use their powerful system to organize political action and other such ventures like curing cancer, thus "disusing" their personal systems...
So no, the wacko collectivist beat him to it.
BurtGravy - Thursday, February 21, 2013 - link
That escalated quickly.iEATu - Wednesday, May 29, 2013 - link
I stopped reading when you started talking about Reagan and cutting taxes creates businesses. Based off of history with Reagan, revenue did NOT come in because big businesses took advantage of the tax cuts and nothing changed except to make big business more powerful and the richest people richer.3DPA - Saturday, August 10, 2013 - link
Wow....this certainly got off topic. The only hope for America is for the US to became a tax haven for the rest of the world...similar to what Singapore did (and they are hugely prosperous). When you factor in under funded pensions, social security and medicare, the real US debt comes to around $86 trillion. You need to drop the US corporate tax rate from the highest in the world (I think around 30% to 35%) to the lowest. When you look at electronics manufacturing, the cost of building off-shore in China really only saves you 7% in labor because 90% of electronic assembly is automated. So you save 7%, but then you lose 4% shipping it back to the US thanks to the price of oil. So you really end up saving 3%. Given all the hassles of dealing with manufacturing half-way around the world, 3% savings is really not that attractive. So why do companies move electronics manufacturing over-seas? Because they save on tax...to the tune of 10% to 15%. You want to bring electronics manufacturing jobs back to the US? Just lower the corporate tax rate to something more globally competitive like 15% to 20%. This will create HUGE influx of job growth in the US. And what happens when more people are working? You broaden the tax base so tax revenue increases. Also, remember that close to $1 Trillion of corporate profits is sitting in foreign banks ($100 Billion of that belongs to Apple) because they don't want to pay the huge US corporate tax rates. Lower that tax rate and all that money flows back into the US where it can be used for investment in the US...which creates more opportunity...etc....etc...etc...So don't believe it when the Obama administration can't solve the US debt issue. It can be solved. But it just kills a socialist/communist president to cut corporate America a break even though not to do so causes us even more harm.
You know.....I have yet to meet anyone that actually thinks Obama is doing a good job or is good for the country. I hear more negative talk, more jokes, seen more negative books published about him and heard more dooms-day scenarios with his name at the middle of it than any other president in memory.
So how did he get re-elected?
Well...there is one good thing I can say about Obama: He made me appreciate Bill Clinton.
azixtgo - Sunday, November 10, 2013 - link
higher prices mean fewer sales. Lower prices mean lower margins. They have to find some balance and I doubt anyone here knows enough to know where that balance is. My thinking is that both companies make too many chips (desktop market). They price desirable chips too high for the average person. Maybe they should make fewer chips and make upgrades more desirable.Both consoles have sold millions already. I doubt it will be that game changing for them. Another thing is that AMD is letting their desktop CPU market suffer. They have not put out better chipsets for their FX processors in a good while so even people who would buy the fx processors may not do so when they see what they get with the motherboard (they do not even have mATX motherboards for these processors AFAIK.) I doubt the APUs will make up for the FX shortfalls. Intel on the other hand offers a less complicated setup. Companies do not just lose profits on lower prices and I am inclined to think the prices can stand to be lower based on what both nvidia and amd have done recently
eg. 280X is a 7970 being sold for $300. GTX 780 now sells for 499. Are they at the balance point? I don't know. i just don't think they are optimizing their market presence at all
Sabresiberian - Wednesday, February 20, 2013 - link
No, the GK110 isn't cheaper to produce, not by a long shot, and Intel's profit margin for their CPUs are quite a bit higher than Nvidia's or AMD's for the graphics solutions. It's pretty amazing considering the GPUs are far bigger.I was hoping for a price around $800, but expecting it to be $900-1200. Sure, I wish it was less, but Nvidia isn't out of line here.
CeriseCogburn - Sunday, February 24, 2013 - link
I agree. It's quiet, it sips power, it's the fastest, it has the features amd lacks, which are many and considerable. It has stable drivers, it is well built, it has 6G of ram, thus making it the most future proof to date, just recently of utmost importance to all the amd fanboys 3G quacking Skyrim mods baggers, now they can eat dirt, then mud from the delicious tears.Yeah pretty sick of the constant CEO price "absolutears" here on every dang release, squealing out price perf bang for buck whines and lies and pauper protestations in betwixt bragging about their knowledge and prowess and spewing fanboy fave.
If you're a fanboy get off your butt and toss a few papers so you can buy a ding dang video card without wailing like a homeless vagrant.