Jump to content

AMD R9 390X Fiji XT Leaked For the Third Time on Shipping Manifesto - Might Make a Surprise Appearance at CES

MetroP

So Fiji XT AMD's upcoming flagship Rx 300 series GPU core has been spotted on the reliable Zauba shipping database going from AMD's Radeon headquarters in Canada to their testing facility in India.

Unlike the previous two times when only one card was shipped this time four GPUs were shipped at once and they were all in "Full Operational Capacity"

Zauba-2.jpg

The New R9 390X card will have 4096 GCN Cores, 4GB of stacked High Bandwidth Memory, a 4096bit wide i/o memory bus and 20nm process technology.

Full a article.

Link to comment
Share on other sites

Link to post
Share on other sites

4 GB vram is kinda dissapointing from AMD. Looks like it will be anounced on CES anyways...

who cares...

Link to comment
Share on other sites

Link to post
Share on other sites

They are releasing 4GB of RAM, half a year later everones upgrade plans are gone and then the are releasing 8GB Ram models, same with NVIDIA

CPU: Xeon 1230v3 - GPU: GTX 770  - SSD: 120GB 840 Evo - HDD: WD Blue 1TB - RAM: Ballistix 8GB - Case: CM N400 - PSU: CX 600M - Cooling: Cooler Master 212 Evo

Update Plans: Mini ITX this bitch

Link to comment
Share on other sites

Link to post
Share on other sites

4 GB vram is kinda dissapointing from AMD. Looks like it will be anounced on CES anyways...

They are releasing 4GB of RAM, half a year later everones upgrade plans are gone and then the are releasing 8GB Ram models, same with NVIDIA

They have no choice, you can't use more than 4GB of HBM until HBM 2.0 comes out in 2016. Until then 4GB is the maximum that can be used,

 

HBM-Memory-Roadmap.jpeg

Link to comment
Share on other sites

Link to post
Share on other sites

They have no choice, you can't use more than 4GB of HBM until HBM 2.0 comes out in 2016. Until then 4GB is the maximum that can be used,

 

HBM-Memory-Roadmap.jpeg

 

What do all the GTX 780s with 6 GB vram do then? Or R9 290s with 8 GB?

who cares...

Link to comment
Share on other sites

Link to post
Share on other sites

4096 seems a little high for a relatively similar architecture that the previous cards have, I'm intrigued about the stock cooler.

.

Link to comment
Share on other sites

Link to post
Share on other sites

That's GDDR5 not HBM -_- . HBM is the brand new 3D stacked memory that has never been used in any graphics card before.

Ke8mR92.jpg

 

Hmmm. I am behind on my tech news. :D

who cares...

Link to comment
Share on other sites

Link to post
Share on other sites

Hmmm. I am behind on my tech news. :D

Each HBM module has an effective 256-wide bus (4 parallel 64-wide channels per module), and AMD could only stack 2 pairs of 1GB modules together before you'd run into inconsistent performance due to bank switching and increasing distance from the GPU. Or AMD would have to further increase costs to widen the memory bus. The final memory bus width is still 512-bit, but it will have a much higher effective clock due to HBM's architecture.

 

It's an early release and 4K is still too expensive for the vast majority of consumers. We'll see 8GB cards before the end of 2015 from AMD with HBM aboard, likely in a 1024-wide configuration.

Software Engineer for Suncorp (Australia), Computer Tech Enthusiast, Miami University Graduate, Nerd

Link to comment
Share on other sites

Link to post
Share on other sites

Price per unit = approx. 76,000 INR = 1,200 USD

 

This can't be right

It could be the dual gpu card..... releasing it with the rest of the series, makes more sense....

Hello This is my "signature". DO YOU LIKE BORIS????? http://strawpoll.me/4669614

Link to comment
Share on other sites

Link to post
Share on other sites

Each HBM module has an effective 256-wide bus (4 parallel 64-wide channels per module), and AMD could only stack 2 pairs of 1GB modules together before you'd run into inconsistent performance due to bank switching and increasing distance from the GPU. Or AMD would have to further increase costs to widen the memory bus. The final memory bus width is still 512-bit, but it will have a much higher effective clock due to HBM's architecture.

 

It's an early release and 4K is still too expensive for the vast majority of consumers. We'll see 8GB cards before the end of 2015 from AMD with HBM aboard, likely in a 1024-wide configuration.

4k is comming down in price AND FAST. I bet it will be < 299 pounds next year.

Hello This is my "signature". DO YOU LIKE BORIS????? http://strawpoll.me/4669614

Link to comment
Share on other sites

Link to post
Share on other sites

How much will it cost?

Pricing is unknown, tho it shouldn't cost that much more than the R9 295x2 once pricing stabilizes.

 

It could be the dual gpu card..... releasing it with the rest of the series, makes more sense....

The R9 390X is a single large GPU card. It's AMD's own TITAN.

Link to comment
Share on other sites

Link to post
Share on other sites

Pricing is unknown, tho it shouldn't cost that much more than the R9 295x2 once pricing stabilizes.

 

The R9 390X is a single large GPU card. It's AMD's own TITAN.

I would hope not since its a 1150$ card.... 

 

Hopefully its around 650$. 700 max. 

Link to comment
Share on other sites

Link to post
Share on other sites

I would hope not since its a 1150$ card.... 

 

Hopefully its around 650$. 700 max. 

Yes but with HBM being an expensive premium technology fresh out of Hynix's production line, you can expect any GPUs using it for the time being will be forced to increase in price. It's on the order of $30/GB between manufacturing and attaching it to the PCB.

Software Engineer for Suncorp (Australia), Computer Tech Enthusiast, Miami University Graduate, Nerd

Link to comment
Share on other sites

Link to post
Share on other sites

Yes but with HBM being an expensive premium technology fresh out of Hynix's production line, you can expect any GPUs using it for the time being will be forced to increase in price. It's on the order of $30/GB between manufacturing and attaching it to the PCB.

Oh boy this gonna be expensive especially when im from Canada  prob 50-100$ more on top of that not being from States. 

 

Everything here is more a cpu cooler is usually 20-30$ more compared newegg.com to newegg.ca 

 

290x is around 50$ more here. Launch of cards is even worse i think cards were 100$ more here with 290x. 

Link to comment
Share on other sites

Link to post
Share on other sites

Oh boy this gonna be expensive especially when im from Canada  prob 50-100$ more on top of that not being from States. 

 

Everything here is more a cpu cooler is usually 20-30$ more compared newegg.com to newegg.ca 

 

290x is around 50$ more here. Launch of cards is even worse i think cards were 100$ more here with 290x. 

This is exactly why AMD didn't put HBM on Carrizo. The issue wasn't space. It was price.

Software Engineer for Suncorp (Australia), Computer Tech Enthusiast, Miami University Graduate, Nerd

Link to comment
Share on other sites

Link to post
Share on other sites

Yes but with HBM being an expensive premium technology fresh out of Hynix's production line, you can expect any GPUs using it for the time being will be forced to increase in price. It's on the order of $30/GB between manufacturing and attaching it to the PCB.

Could you please remind me what the advantages of HBM are? If it costs AMD that much to implement it, the price increase should better have some advatages.

GPU: Gigabyte GTX 970 G1 Gaming CPU: i5-4570 RAM: 2x4gb Crucial Ballistix Sport 1600Mhz Motherboard: ASRock Z87 Extreme3 PSU: EVGA GS 650 CPU cooler: Be quiet! Shadow Rock 2 Case: Define R5 Storage: Crucial MX100 512GB
Link to comment
Share on other sites

Link to post
Share on other sites

Could you please remind me what the advantages of HBM are? If it costs AMD that much to implement it, the price increase should better have some advatages.

Okay, think of it in terms of the power and heat budget of a GPU. a 512-bit bus requires 512/32 = 16 GDDR5 modules. That's a lot of PCB real estate and a lot of power and heat (about 30-35W) 4HBM modules in this configuration delivers slightly better performance for only 8W. Now, don't forget the bus may eventually be widened to cover individual modules in perfect parallel instead of pairing them up, but that does require a great deal of engineering.

Software Engineer for Suncorp (Australia), Computer Tech Enthusiast, Miami University Graduate, Nerd

Link to comment
Share on other sites

Link to post
Share on other sites

Could you please remind me what the advantages of HBM are? If it costs AMD that much to implement it, the price increase should better have some advatages.

With HBM you can cut out a lot of power consumption used by several higher power consuming GDDR5 modules. On top of that you can add massive amounts of bandwidth without consuming a large portion of PCB space. GPU performance is outgrowing GDDR5 technology. Take a look at the TITAN-Z PCB and you will start to understand what I mean.

GF_GTX_Titan_Z_PCB.jpg

Link to comment
Share on other sites

Link to post
Share on other sites

With HBM you can cut out a lot of power consumption used by several higher power consuming GDDR5 modules. On top of that you can add massive amounts of bandwidth without consuming a large portion of PCB space. GPU performance is outgrowing GDDR5 technology. Take a look at the TITAN-Z PCB and you will start to understand what I mean.

-snip-

@ICantThinkOfAnyGoodName And if you look at that PCB, that's 2 384-bit busses (2 sets of 12 GDDR5 modules), less complex than the R9 295x2 which has a double 512-bit bus on its PCB.

Software Engineer for Suncorp (Australia), Computer Tech Enthusiast, Miami University Graduate, Nerd

Link to comment
Share on other sites

Link to post
Share on other sites

Fuji XT is not the dual card, if amd follow the naming of the past the dual card will have a different name. I just hope that this pricing (1200 USD per card) don't have anything to do with the actual price...

Link to comment
Share on other sites

Link to post
Share on other sites

Each HBM module has an effective 256-wide bus (4 parallel 64-wide channels per module), and AMD could only stack 2 pairs of 1GB modules together before you'd run into inconsistent performance due to bank switching and increasing distance from the GPU. Or AMD would have to further increase costs to widen the memory bus. The final memory bus width is still 512-bit, but it will have a much higher effective clock due to HBM's architecture.

 

It's an early release and 4K is still too expensive for the vast majority of consumers. We'll see 8GB cards before the end of 2015 from AMD with HBM aboard, likely in a 1024-wide configuration.

HBM rides on a 128-bit physical interface with two 128-bit internal TSV to the main compute die. Eight channels is where you get the specified 1024 wide throughput. Tho the physical interface for the R9 390X per example should still be 512-bit (4 stacks of 4-Hi HBM). With the massive amount of throughput allows HBM to run at a much lower clock than GDDR5. Not sure if this will change with 8-Hi stacks in the future tho by then AMD cards will be shipping with over 1TB of memory bandwidth. They probably could of stacked 8 stacks of 4-Hi on the R9 390X for a 1024-bit interface and 8 GB of VRAM density. Tho the pricing likely killed the cards value while only being two R9 280X in the same die.

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×