NVIDIA GTX TITAN reviews

[aussie-revhead]

I still think BF3 performance is poor, has anyone else found this?

[nossy23]

I still think BF3 performance is poor, has anyone else found this?

I had high framerates, but poor performance (ftp looked like they were going really low sometimes, even though this was not the case) in BF3 with my Geforce Titan. However after a lot of testing I found that at least on my system, the cause of the mircostutter etc was the Asus AI 2 Suite. I have removed this SW now, and everything is really smooth.

[aussie-revhead]

Really? How did you diagnose this? Maybe I should try a clean install with little extra and see how it goes. I am sure its no better than my GTX670 and my 680 Lightning was better (for me anyway, regardless of what reviews say).

[nossy23]

Well, when I was testing the new GPU in my system I saw that my framerates were high in BF3, but it didn't look like they were. So I ran OCCT. This showed some weird things. Core 0 ran in a fraction of a second too hot (from like 40 C to about 80, other cores stayed at 40-45 C), and the benchmark stopped. I was thinking either the CPU or motherboard was faulty or there is something interfering with the readings.
The only thing I could think of on my system that could do this was the Asus AI 2 Suite. Even though I thought I was not really running anything with it (I had deselected all the flags in that program), I saw it was still running in the background. So I threw it off my system, and heay presto!!! No more weird readings in OTTC and BF3 is running great!

[Andreas]

I do have the cards for 2 weeks now. I can't share a review, but some experiences.

The Coolermaster HAF case is a good fit


10.752 shader cores


The hot zone


First, I don't do gaming, so please take one of the many sites for further information about the cards in a gaming setup.

Memory
I did use the cards in 2 different systems - a single LGA-2011 socket mobo and a dual LGA-2011 Xeon motherboard. The main difference is the available bandwidth over the PCI Express interconnect between cards and main memory. Due to the limitation of 40 PCI lanes per LGA-2011 socket, a 4 card setup has to utilize PCI-e x8 connections per card, limiting the available bandwidth to main memory to approx 5.6 GB/sec when PCI-e 3.0 is enabled (it is off by default).

The aggregated transferrate of the cards is above 20 GB/sec to main memory. Given that 1600 MHz main memory in an LGA-2011 system achieves 40 GB/sec, it would be easy to assume, that 50% of the available memory bandwidth is taken by the I/O operations of the cards and another 50% is still available to the CPU. There is one caveat though - cache consistency. By the very design of current motherboards, a PCI transaction flushes the respective line in the cache to avoid any inconsistency between main memory and cache. This requires the CPU to load the dataset not from cache, but from main memory - which is significantly slower.

To cut a long story short. If 4 cards are not only crunching with data within the cards, but need to regularily transfer data to the main system, most systems will exhibit performance limitations on the CPU/memory side to feed the cards appropriately fast.

The situation is much more "stable" on the dual Xeon side. Out of the 80 PCIe 3.0 lanes the 2 sockets provide, some motherboards make up to 72 lanes available in the 6 or 7 PCI slots. In a balanced system, 2 times 2 PCI slots with x16 connects each are available to unleash more bandwidth between cards and main memory. in x16 mode PCI 3.0, data transfer rates per card are in the range of 11.2 GB/sec read or write.

Energy
Running 4 cards at maximum TDP stretched my Silverstone 1200W PSU beyond its capabilities. I had to replace it with a 1500W which provides enough stability and headroom for quick power surges (caused by sudden load changes). Highest power consumption observed was (single socket system i7-3930K):1180Watt, dual E5-2687W Xeon: 1460 Watt.
Idle power draw on the single socket system with 4 cards: 120 Watt

Used as pure compute cards, power consumption goes down by 20-40 Watt (Texture units and ROP units aren't used), adding headroom to the TDP and temp budgets.

Temperature
For noise insensitive people, basically a non-issue. By default the fans of the cards max out at 60% of its max rpm. This speed keep the whole system silent enough for office noise levels to not disturb anybody. But cards hit under load temperatures above 80 degrees with stock settings. Cranking up the fanspeed to 85% (4400 rpm, and the max setting in EVGA precision 4.0), temperature goes down by 20-25 degrees. A good thing. The bad thing: Noise levels are very high. Nothing somebody would like to experience for hours. In my case, the solution was easy - move the rig in the basement. It doesn't make any difference where numbers are calculated.

The first few days I was careful with regards how far a sustainable load for hours and days would be a good from a power,temp,lifecycle balance side.
I kept the cards at stock frequency, cranked up the fans to 70% to maintain the temps under 70-75 degrees. The cards are very effective to avoid heating up the case. Behind the case it is very hot, but RAM and system board temperature barely increased.

Over a period of 5 days I kept the cards running with higher settings:
Increased shader frequency by 260 MHz to 1097 MHz
Memory + 160 MHz to 6320 MHz
Fan to 85%
TDP to 106%
Voltage: default 1,162V
max temp set to 80 degrees, cards were between 65 and 75 degrees, depending on load.

rock solid experience. Not a single glitch.

You might ask: Why Titan's?
For me, 2 reasons why I waited and chose GK110 cards.
It is the only Geforce card with CUDA compute capability 3.5 (GTX 680, 690 are on 3.0). This is one step further to simplify programming and increase performance in simpler ways by reducing the need for execessive roundtrips between GPU and the CPU.
1,5 TFlop/s double precision performance ( > 6 TFlop/s per system with embarrasingly parallel workloads)

What's missing:
Currently, NVidia is only supporting OpenCL 1.1, but not OpenCL 1.2
A lock to fix the cards to the case. They tend to disappear eventually and end up temporarily in my son's gaming system ....

Cheers,
Andy

[nossy23]

I have not yet seen the review of the Titan card comparing it's Bitcoin performance against the AMD cards. Is it still weak like the other Nvidia GPU's?

[Andreas]

I have not yet seen the review of the Titan card comparing it's Bitcoin performance against the AMD cards. Is it still weak like the other Nvidia GPU's?

https://en.bitcoin.it/wiki/Why_a_GPU...ter_than_a_CPU
Please scroll down for an explanation

Forgot to mention:
NVidia added the missing logic and shift instruction in Version 3.1 of the instruction set (PTX-ISA 3.1) available in SM 3.5. It will take some time, when new CUDA kernels from the respective applications will be available which take benefit of this extension.

Andy

[iddqd]

I have not yet seen the review of the Titan card comparing it's Bitcoin performance against the AMD cards. Is it still weak like the other Nvidia GPU's?

Don't they have dedicated hardware for hasing bitcoins now? That's way faster than any GPU because it's designed to do it in hardware?

[mbreslin]

I don't know the cost of the dedicated hardware but I can tell you when I recently sold my 5970s on that big auction site a ton of bitcoin miners were bidding.

[mmorpgfocus]

The Bitcoin ship has sailed.

Using NVIDIA or AMD cards to mine coins is not going to be very profitable.

You are now competing with THIS: https://products.butterflylabs.com/h...oin-miner.html

The TITAN is useful for hardcore extreme gamers, and people who use GPGPU, especially that second group. TITAN would be great for science work like brain mapping, weather simulation, nuclear implosion modeling etc. etc.

If you are buying TITAN to mine bitcoins you are making poor choices.

However that is not the point, all these reviews are wrong/incorrect. The new drivers NVIDIA released on the 25th of March drastically improve TITAN performance, and the old reviews do not reflect the gaming power of the chip.

Two TITANS in SLI will probably last you through the entire PS4/XBOX Next gaming generation. I can play anything maxed at almost 4K reolution in DX11 Ultimate modes (5760x1200).

Plain 1080P would be a joke!

[eXa]

Eh... $29,899.00

Amd is better at bitcoin mining than nVidia anyways btw...

[mmorpgfocus]

Eh... $29,899.00

Amd is better at bitcoin mining than nVidia anyways btw...

I linked to a $30k monster to show what you are up against running a GPU trying to software mine bitcoins.

If you understand how they work at all, you know that with machines like that (and there are also $500 ASICS) - well just with ASIC hardware bitcoin miners proliferating - you won't be able to ever make back even a fraction of the cost of electricity used with a GPU to mine coins now no matter what AMD GPU is used. ASICS rule, and the difficulty is going to skyrocket.

That being said, TITAN was way under-rated in ALL reviews. Just trying to get the word out there that the new drivers are adding 10%-50% FPS increase in most games. If you want to know TITANS true value you need to add 10FPS at least to just about every score. It puts it into a class that is truly untouchable by AMD or 690 GTX.

SLI TITAN will rule handsomely well into 2013, I will not be surprised if it stands up to next gen 780 series, giving TITAN owners no reason to upgrade! Also I don't understand why these were not stock clocked at 1GHZ.

[STEvil]

I keep seeing these references to how good the titan is with GPGPU but nobody ever really posts numbers......

[Andreas]

I keep seeing these references to how good the titan is with GPGPU but nobody ever really posts numbers......

Here we go with 2 familiar numbers - Folding at Home and Furmark:




Andy

[STEvil]

Furmark isnt really much of a GPGPU benchmark.

edit

Looks like 66-70k for F@H, 3x a GTX580/680 (~20-22k). Good increase there at least.

[Olivon]

Added nVidia GK110 models: GeForce GTX Titan LE

http://www.hwinfo.com/forum/Thread-H...-Beta-released

[GoldenTiger]

http://www.hwinfo.com/forum/Thread-H...-Beta-released

Titan LE (GK110), and some 7xx notebook parts (GK208) mentioned.... hrrrrrm.

[aussie-revhead]

Titan LE (GK110), .... hrrrrrm.

Lets hope this doesnt kills our Titan prices lol ... or give 80% of the performance at 50% of the price .....

[paulbagz]

"There have been several rumors for a while now that NVIDIA would be building an LE version of the GeForce GTX Titan. We can now expand that even further, as it seems there will be two new models release. An even faster Ultra edition and thus the cheaper LE edition. German website 3dcenter claims that the GeForce GTX TITAN LE has 2496 Shader cores, 208 TMUs, 40 ROPs, and a 320-bit wide memory interface (5120 MB memory.) The tertiary GK110 product would then be the GeForce GTX TITAN Ultra.

The Ultra would be a small upgrade for the 'regular' Titan, opening up one disabled SMX cluster bringing 2880 Shader cores with 240 TMUs towards the Ultra.

There remains a lot of speculation of course, but this could mean the Titan LE at 599 USD. The Original Titan at 799 USD, and the GeForce Titan Ultra at 999 USD."


http://www.guru3d.com/news_story/nvi..._titan_le.html

Very keen!

-PB

[GoldenTiger]

"There have been several rumors for a while now that NVIDIA would be building an LE version of the GeForce GTX Titan. We can now expand that even further, as it seems there will be two new models release. An even faster Ultra edition and thus the cheaper LE edition. German website 3dcenter claims that the GeForce GTX TITAN LE has 2496 Shader cores, 208 TMUs, 40 ROPs, and a 320-bit wide memory interface (5120 MB memory.) The tertiary GK110 product would then be the GeForce GTX TITAN Ultra.

The Ultra would be a small upgrade for the 'regular' Titan, opening up one disabled SMX cluster bringing 2880 Shader cores with 240 TMUs towards the Ultra.

There remains a lot of speculation of course, but this could mean the Titan LE at 599 USD. The Original Titan at 799 USD, and the GeForce Titan Ultra at 999 USD."


http://www.guru3d.com/news_story/nvi..._titan_le.html

Very keen!

-PB

I'm skeptical at the speculated Titan LE pricing, but if it does come out at that, I'm hopping right on it!

[paulbagz]

I'm skeptical of the standard Titan pricing rofl!

Seems like this is the stop gap until 7XX series later this year.

If they're going to make 3-4 cards out of the name, why not just call it 7XX to begin with lol >.<

-PB

[behrouz]

lol , I laugh at Titan owners if Nv drop value!, Nv if you want to Release new version and drop Titan's Price to lower , wouldn't you bother for someone tell you :

I Just bought Two TITAN cards , Now there is an ultra version, coming soon , Fu!!!!!!! you NVIDIA!

but Hope Nv wont touch their Price in other to respect consumers who have Titan.

[Kallenator]

I have my doubts that people who buy the absolutely hottest hardware available think it wont depreciate like a rock falling in perfect vacuum? ^^

[Andreas]

I keep seeing these references to how good the titan is with GPGPU but nobody ever really posts numbers......

On a second note:

You might be interested in this post:
http://www.xtremesystems.org/forums/...t=#post5183120

Meanwhile max PPD rates have been going up.

[LordEC911]

Meanwhile max PPD rates have been going up.

What does it look like at the end?