Guru3D.com
  • HOME
  • NEWS
    • Channels
    • Archive
  • DOWNLOADS
    • New Downloads
    • Categories
    • Archive
  • GAME REVIEWS
  • ARTICLES
    • Rig of the Month
    • Join ROTM
    • PC Buyers Guide
    • Guru3D VGA Charts
    • Editorials
    • Dated content
  • HARDWARE REVIEWS
    • Videocards
    • Processors
    • Audio
    • Motherboards
    • Memory and Flash
    • SSD Storage
    • Chassis
    • Media Players
    • Power Supply
    • Laptop and Mobile
    • Smartphone
    • Networking
    • Keyboard Mouse
    • Cooling
    • Search articles
    • Knowledgebase
    • More Categories
  • FORUMS
  • NEWSLETTER
  • CONTACT

New Reviews
Guru3D Rig of the Month - February 2021
ASUS GeForce RTX 3060 STRIX Gaming OC review
EVGA GeForce RTX 3060 XC Gaming review
MSI GeForce RTX 3060 Gaming X TRIO review
PALIT GeForce RTX 3060 DUAL OC review
ZOTAC GeForce RTX 3060 AMP WHITE review
Fractal Design Meshify 2 Compact chassis review
Sabrent Rocket 4 PLUS 2TB NVMe SSD review
MSI Radeon RX 6900 XT GAMING X TRIO review
Guru3D Q1 Winter 20/21 PC Buyer Guide

New Downloads
Media Player Classic - Home Cinema v1.9.10 Download
GeForce 461.72 WHQL driver download
AIDA64 Download Version 6.32.5640 beta
CrystalDiskInfo 8.11.2 Download
AMD Radeon Adrenalin Edition 21.2.3 driver download
GPU-Z Download v2.37.0
Intel HD graphics Driver Download Version: DCH27.20.100.9313
HWiNFO Download v6.43 - 4380 Beta
AMD Radeon Adrenalin Edition 21.2.2 driver download
3DMark Download v2.17.7137 + Time Spy


New Forum Topics
Judge orders Valve to hand over Steam data AMD Radeon Software Adrenalin 2020 Edition 21.2.3 Meltdown and Specter patch performance hit tested on Optane and VROC. GeForce 461.72 WHQL drivers: download & discussion Any way to "Half-Refresh V-Sync" with AMD GPUs? Thanks! RTSS 6.7.0 beta 1 New Upcoming ATI/AMD GPU's Thread: Leaks, Hopes & Aftermarket GPU's Tesla shuts down Model 3 production line for two weeks due to chip shortage MOSFET prices to increase on worsening component shortages The AMD Ryzen All In One Tread /Overclocking/Memory Speeds & Timings/Tweaking/Cooling Part 2




Guru3D.com » News » GeForce GTX TITAN P might see August Announcement

GeForce GTX TITAN P might see August Announcement

by Hilbert Hagedoorn on: 07/05/2016 04:32 PM | source: | 22 comment(s)
GeForce GTX TITAN P might see August Announcement

I've been doubting to post this news-item as the information is unsubstantiated and based on a very loose rumor from one website. But there currently is a rumor that GeForce GTX TITAN P would be announced at Gamescom. Personally I doubt it as it is way too soon, but hey I could be wrong here.

The news spreads based on a loose post on the source: VR World. Gamescom will be held in Cologne, Germany starting at the 17th of August. The rumor is that a card called GeForce GTX TITAN P (with the P for Pascal) will be announced, again I think it is too soon. That GP100 silicon would see light of day in two flavors, a 12 and 16 GB one. The 16 GB model would feature four HBM2 stacks over a 4096-bit memory bus; the 12 GB variant would feature three active HBM2 stacks and this would be based on a 3072-bit bus. Both similar to the Tesla P100-based PCIe super computer accelerators 

The Pascal based GPU driving the unit holds 15 Billion transistors which is roughly double that of the current biggest Maxwell chip. Gp100 is huge at 600mm^2. The prognosis performance (according to Nvidia) is 5.3TFLOPS using 64-bit floating-point numbers and is rated at 10.6TFLOPS using 32-bit and 21.2TFLOPS using 16-bit. The P100 has 4MB of L2 cache and 14MB of shared memory for just the register file. The following table provides a high-level comparison of Tesla P100 specifications compared to previous-generation Tesla GPU accelerators, I added the GP100 as a fully enabled product in the following diagram to get an idea of what such a GPU would entail the GP100 might end up as GP102 (consumer chip) btw :
   

Products Tesla K40 Tesla P100 GP100 GTX 1080
GPU GK110 (Kepler) GP100 (Pascal) GP100 (Pascal) GP104 (Pascal)
SMs 15 56 60 40
TPCs 15 28 30 20
FP32 CUDA Cores / SM 192 64 64 64
FP32 CUDA Cores / GPU 2880 3584 3840 2560
Base Clock 745 MHz 1328 MHz ~1328 MHz 1607 MHz
GPU Boost Clock 810/875 MHz 1480 MHz ~1480 MHz 1733 MHz
Texture Units 240 224 240 160
Memory Interface 384-bit GDDR5 4096-bit HBM2 4096-bit HBM2 256-bit
Memory Size Up to 12 GB 16 GB 16 GB 8 GB
L2 Cache Size 1536 KB 4096 KB 4096 KB 2048KB
Register File Size / SM 256 KB 256 KB 256 KB 256 KB
Register File Size / GPU 3840 KB 14336 KB 14336 KB 10240 KB
TDP 235 Watts 300 Watts ~300 Watts 180 Watts
Transistors 7.1 billion 15.3 billion 15.3 billion 7.2 Billion
Manufacturing Process 28-nm 16-nm 16-nm 16-nm

As the block diagram now shows, the GP100 features six graphics processing clusters (GPCs). Just look at the diagram and count along with me - each GPC holds 10 streaming multiprocessors (SMs) and then each SM has 64 CUDA cores and four texture units. Do the math and you'll reach 640 shader processors per GPC and 3840 shader cores with 240 texture units in total.

  • 6 (GPC) x (10x64) = 3840 Shader processor units in total.

Meaning the GP100 used on the Tesla P100 is not fully enabled. Nvidia is known to out GPU that have disabled segments, it helps them selling different SKUs, the Tesla P100 holds a shader count of 3584 and thus has 56 SMs enabled (from the 60).

GP100’s SM incorporates 64 single-precision (FP32) CUDA Cores. In contrast, the Maxwell and Kepler SMs had 128 and 192 FP32 CUDA Cores, respectively. The GP100 SM is partitioned into two processing blocks, each having 32 single-precision CUDA Cores, an instruction buffer, a warp scheduler, and two dispatch units. While a GP100 SM has half the total number of CUDA Cores of a Maxwell SM, it maintains the same register file size and supports similar occupancy of warps and thread blocks.GP100’s SM has the same number of registers as Maxwell GM200 and Kepler GK110 SMs, but the entire GP100 GPU has far more SMs, and thus many more registers overall. This means threads across the GPU have access to more registers, and GP100 supports more threads, warps, and thread blocks in flight compared to prior GPU generations.

Since the graphics memory is on-die HBM2, the VRAM amount is fixed. That means that ALL GP100 products will get 16GB of memory or less. HBM2 will run a wide 4096-bit HBM2 (1024 bit per IC stack) memory interface running an effective bandwidth anywhere up-to a full 1 TB/s.

This is a big chip, very big at 600mm^2 hence it is interesting to see that 16nm can offer a lot in terms of clock frequency, The Tesla P100 is an enterprise part that ends up in servers, however this part already is clocked at 1328 MHz with Boost capabilities towards a frequency of 1480 MHz. Combined the TDP still remains to be under 300W. 







« MSI Takes GTX 1080 4-way SLI 3DMark crown · GeForce GTX TITAN P might see August Announcement · X2 RINDJA Full-tower Chassis »

Related Stories

No SLI for GeForce GTX 1060 ? - 07/04/2016 12:48 PM
News on the web spreading like wildfire is that the upcoming GeForce GTX 1060 will not be SLI compatible. Judging from leaked photos it seems clear that the card does not have SLI connectors....

MSI GeForce GTX 1080 and GTX 1070 DUKE Edition Graphics Cards - 07/04/2016 09:10 AM
It seems that MSI is releasing the MSI GeForce GTX 1080 and GTX 1070 DUKE Edition Graphics Cards, these are intended for the the more Asia segmented markets....

GeForce GTX 1060 Launch Imminent + Leaked Specs - 07/01/2016 04:18 PM
More and more rumors on the GeForce GTX 1060 being released are hitting the web, including specifications.  Earlier on a GeForce GTX 1060 already had been photographed on a Hong Kong market. ...

Review: MSI GeForce GTX 1080 SEA HAWK X - 07/01/2016 04:18 PM
Oh yes we're going Hybrid today as we review a GeForce GTX 1080 being tested with Corsair hybrid cooling!, yes join us as we test the MSI GeForce GTX 1080 SEA HAWK X . The gear that everybody is wait...

Gigabyte to launch Mini-ITX version of GeForce GTX 1070 - 07/01/2016 01:45 PM
Gigabyte is releasing a mini-ITX version of the 1070 GTX . The card with a length of 16.9 centimeters will even receive a factory overclock. The Gigabyte GeForce GTX 1070 Mini ITX OC features a single...


5 pages 1 2 3 4 5


PrMinisterGR
Senior Member



Posts: 7519
Joined: 2014-09-27

#5301094 Posted on: 07/05/2016 04:56 PM
Titan P
enis
?

If they managed to drop that Tesla price from $10,000 to something like two grand (don't expect this to be under $1,500 if it launches now), and produce more than two at a time, it is possible I guess.

It will be good for PR, even if there are none to be found.

PrMinisterGR
Senior Member



Posts: 7519
Joined: 2014-09-27

#5301096 Posted on: 07/05/2016 05:00 PM
If a 1080 with literally less than half the hardware reaches $800 for a meaningless FE, then this at $1,500 would be a comparative bargain. Especially if they leave the compute part intact. Then it's a steal.

Fox2232
Senior Member



Posts: 11475
Joined: 2012-07-20

#5301103 Posted on: 07/05/2016 05:23 PM
Why would they be wasting wafers on Consumer grade Titan P?
As they stated, they have enough orders for Tesla P100, that they'll not be able to fulfill them till end of the year.

Loophole35
Senior Member



Posts: 9786
Joined: 2011-09-21

#5301106 Posted on: 07/05/2016 05:26 PM
Why would they be wasting wafers on Consumer grade Titan P?
As they stated, they have enough orders for Tesla P100, that they'll not be able to fulfill them till end of the year.

Because this will likely not be a full GP100/102. This will be a way to make money off of otherwise junk waffers.

I expect this to be like the original Titan and be a cut GPU.

PrMinisterGR
Senior Member



Posts: 7519
Joined: 2014-09-27

#5301107 Posted on: 07/05/2016 05:28 PM
Because this will likely not be a full GP100/102. This will be a way to make money off of otherwise junk waffers.

I expect this to be like the original Titan and be a cut GPU.

That would make quite a lot of sense, until we remember that they could do the same with compute GPUs too, and at higher margins. Unless they have some kind of marketing research that tells them which is the most profitable curve of CUs/Cost for compute GPUs (ie, the max amount of profit per CU they can make per CU/waffer with GPGPU), and that reached the point where it made sense to try the consumer market.

5 pages 1 2 3 4 5


Post New Comment
Click here to post a comment for this news story on the message forum.


Guru3D.com © 2021