Monday, April 10th 2017

NVIDIA Beats AMD to Market On HBM2 - Announces Tesla P100

NVIDIA has announced availability of their latest data center accelerator, the Tesla P100, which is the world's first HBM2-powered add-in-card. this means that NVIDIA effectively beat AMD in time to market with HBM2 technology, which AMD pioneered (in its HBM form) with the Fury line of graphics cards.

NVIDIA naturally touts this as the world's most advanced data center accelerator, for workloads such as "Artificial intelligence for self-driving cars. Predicting our climate's future. A new drug to treat cancer." NVIDIA's green graphics show an almost 50x increase in computing power from 8x Tesla P100 accelerators when compared to a dual CPU server based on Intel's Xeon E5-2698 V3 (which isn't really all that surprising.) NVIDIA further brings in the PR talk with examples on how a single GPU-accelerated node powered by four Tesla P100s - interconnected with PCIe - can replace up to 32 commodity CPU nodes for a variety of applications - saving up to 70% in overall data center costs.

Source: NVIDIA
Add your own comment

105 Comments on NVIDIA Beats AMD to Market On HBM2 - Announces Tesla P100

#1
Trompochi
But... Can it play Crysis 3? (sorry I had to XD)
Posted on Reply
#2
RejZoR
Business segment maybe, but not for consumers. Where AMD still holds the first for HBM and also for HBM2. Because RX Vega cards already exist, you still can't actually buy them yes, but they exist far more than anything NVIDIA might have. We've acually seen real, operating Vega cards.
Posted on Reply
#3
PowerPC
RejZoR said:
Business segment maybe, but not for consumers. Where AMD still holds the first for HBM and also for HBM2. Because RX Vega cards already exist, you still can't actually buy them yes, but they exist far more than anything NVIDIA might have. We've acually seen real, operating Vega cards.
Which doesn't mean Nvidia doesn't have them. AMD just always chooses to show off what they have before they can actually deliver it in any reasonable form or quantity.
Posted on Reply
#4
RejZoR
But they've shown a real functioning card.
Posted on Reply
#5
robal
Where is HBM2 equipped Titan then...
Posted on Reply
#6
Caring1
But can it be used for Crunching or Folding?
Posted on Reply
#7
Naito
RejZoR said:
Business segment maybe, but not for consumers. Where AMD still holds the first for HBM and also for HBM2. Because RX Vega cards already exist, you still can't actually buy them yes, but they exist far more than anything NVIDIA might have. We've acually seen real, operating Vega cards.
The bias is strong in this one.
Posted on Reply
#8
Nokiron
robal said:
Where is HBM2 equipped Titan then...
This is a cut-down GP100. It would perform on par with the first Pascal Titan, the key difference would be significantly increased FP64-performance.
Posted on Reply
#9
RejZoR
Naito said:
The bias is strong in this one.
Is it? Are you going to accuse me of being an AMD fanboy now? I suggest you clickidy click to my specs...
Posted on Reply
#10
Naito
RejZoR said:
Is it? Are you going to accuse me of being an AMD fanboy now? I suggest you clickidy click to my specs...
Not accusing you and owning hardware from a different camp isn't exactly cover. Just an observation that you often have a strong biased opinion regarding most topics you post on (posts regarding Nvidia, for example). It shows. Does it not grow tiresome?
Posted on Reply
#11
Prima.Vera
Caring1 said:
But can it be used for Crunching or Folding?
Is all about the drivers and if the Bios is optimised for those kind of tasks. In theory can be a monster.
Posted on Reply
#12
jabbadap
Uhm what is this, year old news?
Posted on Reply
#13
Relayer
Naito said:
Not accusing you and owning hardware from a different camp isn't exactly cover. Just an observation that you often have a strong biased opinion regarding most topics you post on (posts regarding Nvidia, for example). It shows. Does it not grow tiresome?
Just checked Newegg. No joy though. /s

jabbadap said:
Uhm what is this, year old news?
Looking closely for wood screws. :D
Posted on Reply
#14
oxidized
Naito said:
Not accusing you and owning hardware from a different camp isn't exactly cover. Just an observation that you often have a strong biased opinion regarding most topics you post on (posts regarding Nvidia, for example). It shows. Does it not grow tiresome?
Don't bother, being AMD biased has become the new cool thing in the last years. Eventually people will realize how stupid it is to support something or someone, just because they're the underdogs
Posted on Reply
#15
Naito
oxidized said:
Don't bother, being AMD biased has become the new cool thing in the last years. Eventually people will realize how stupid it is to support something or someone, just because they're the underdogs
People can support whatever they may like - each to their own and all that, but is it expecting too much to scroll past the article and read some level-headed discussion and debate about tech? Not jump straight into brand bias spew from regular members...

Sorry, for being off topic. Over and out.
Posted on Reply
#16
RejZoR
Naito said:
Not accusing you and owning hardware from a different camp isn't exactly cover. Just an observation that you often have a strong biased opinion regarding most topics you post on (posts regarding Nvidia, for example). It shows. Does it not grow tiresome?
Bias? For stating facts? NVIDIA doesn't have even a working prototype shown to the public. AMD has. The end. If we'd play that game, we can also make up things like uh oh, but AMD has HBM3 already. They haven't told anyone about it or shown it, but they already have one.
Posted on Reply
#17
Nokiron
jabbadap said:
Uhm what is this, year old news?
Yeah, seems weird. Last I checked we had a couple of those running in the rack.
Posted on Reply
#18
Naito
RejZoR said:
Bias? For stating facts? NVIDIA doesn't have even a working prototype shown to the public. AMD has. The end. If we'd play that game, we can also make up things like uh oh, but AMD has HBM3 already. They haven't told anyone about it or shown it, but they already have one.
No need to take it so personally.
Posted on Reply
#19
oxidized
Naito said:
People can support whatever they may like - each to their own and all that, but is it expecting too much to scroll past the article and read some level-headed discussion and debate about tech? Not jump straight into brand bias spew from regular members...

Sorry, for being off topic. Over and out.
I'll suggest you go for a small tour on overclock.net, lol.

/OT
Posted on Reply
#20
NeDix!
Wasnt VEGA(HBM2) server/pro workstation released/announced already ?
Posted on Reply
#21
silentbogo
RejZoR said:
Business segment maybe, but not for consumers. Where AMD still holds the first for HBM and also for HBM2. Because RX Vega cards already exist, you still can't actually buy them yes, but they exist far more than anything NVIDIA might have. We've acually seen real, operating Vega cards.
Same could be said about NV showing the prototype last year, or argued backwards that it was a dud. What matters that P100 is in retail and MI25 is not.

Relayer said:
Just checked Newegg. No joy though. /s
Newegg became a horrible place to look for enterprise stuff.
Here's one I found just a few days ago:
http://www.thinkmate.com/hardware/co-processors?a=YToxOntpOjE0Mzc7YToxOntpOjA7czoxMDoiVGVzbGEgUDEwMCI7fX0%3D

Supermicro also started to offer P40 and P100-based server solutions.
Posted on Reply
#22
Nokiron
silentbogo said:
Same could be said about NV showing the prototype last year, or argued backwards that it was a dud. What matters that P100 is in retail and MI25 is not.


Newegg became a horrible place to look for enterprise stuff.
Here's one I found just a few days ago:
http://www.thinkmate.com/hardware/co-processors?a=YToxOntpOjE0Mzc7YToxOntpOjA7czoxMDoiVGVzbGEgUDEwMCI7fX0=

Supermicro also started to offer P40 and P100-based server solutions.
I wonder what the news is though? I'm pretty sure my work has P100s and have had for quite some time.
Posted on Reply
#23
jabbadap
Nokiron said:
This is a cut-down GP100. It would perform on par with the first Pascal Titan, the key difference would be significantly increased FP64-performance.
And fp16 performance. If price does not matter, you actually can buy quadro GP100 now.
Posted on Reply
#24
Fluffmeister
So much drama! These things have been up and running and available for months now, but only as part of their DGX-1.

This guy for example was playing with a rack of em last year: https://linustechtips.com/main/topic/688793-benchmark-10x-tesla-p100-gpu-nvtp100-16/

Point is if you have deep pockets you can buy em indivually now.

https://www.scan.co.uk/products/16gb-pny-nvidia-tesla-p100-module-pcie-30-(x16)-hbm2-gpu-tba-3584-cores-93-tflops-sp-47-tflops-dp-pa
Posted on Reply
#25
RejZoR
silentbogo said:
Same could be said about NV showing the prototype last year, or argued backwards that it was a dud. What matters that P100 is in retail and MI25 is not.


Newegg became a horrible place to look for enterprise stuff.
Here's one I found just a few days ago:
http://www.thinkmate.com/hardware/co-processors?a=YToxOntpOjE0Mzc7YToxOntpOjA7czoxMDoiVGVzbGEgUDEwMCI7fX0=

Supermicro also started to offer P40 and P100-based server solutions.
From where did you drag MI25 in? I was talking about CONSUMER card, the RX Vega.
Posted on Reply
Add your own comment