[BitsAndChips]390X ready for launch - AMD ironing out drivers - Computex launch

[RussianSensation]

DX12 looks to improve multi GPU scaling quite well. So its not exactly silly to bet on something like Tonga x2 with HBM for the long run.

This is easily the least logical rumour of all time regarding R9 390X. And it basically fails in all areas:

1) No credible website or in fact no website in the world in the last 1.5 years even once leaked the possibility of R9 390X being comprised of dual Tonga chips. The only time it was ever mentioned that Fiji XT was a dual-chip was by Fudzilla who confused the requirement for Virtual reality running smoother on 2 cards and automatically assumed the test rig had a single Fiji card with dual GPUs instead of dual Fiji cards. Even if we went with the idea that Fiji XT is a dual-GPU card, AMD would simply use dual Hawaii XTs with HBM1. Considering AMD already released R9 295X2 with 500W of power easily cooled by a single 120mm AIO CLC, there is no need at all to compromise on performance and go with mid-range Tonga chips. In the $700 market segment, performance rules, not power usage. No one will buy a $700 dual Tonga XT card that uses 200W of power but loses to an R9 295X2 90% of the time. The highest end is all about performance.

2) What does DX12 have to do with CF? In all cases where CF works, R9 295X2 would obliterate a 390X consisting of dual 2048 SP, 32 ROPs, 128 TMUs Tonga XTs. AMD has never released a next gen flagship card that was slower than last gen's flagship, in any gaming benchmark.

R9 290X > HD7970Ghz > HD6970 > HD5870 > HD4890/70 > HD3870 > HD2900XT.

This point alone destroys your entire theory.

3) When CF doesn't work, one of those dual Tonga XT chips would end up with 512GB/sec+ memory bandwidth. This is easily the most inefficient use of expensive HBM1. This also guarantees that when CF doesn't work, R9 380/380X would outperform R9 390/390X. No, just no. Even since HD6990, 7990 and R9 295X2, the single chip inside AMD's dual-chip card is a flagship one. AMD does not make newest flagship GPUs out of mid-range $199 chips.

All of these and other points were already discussed. Either you don't pay attention or you are purposely spreading FUD.

Remember Kyle Bennett said "400 series will be the real new architecture"?

When Eric Demers introduced GCN, he said it will be around for at least 5 years. Anyone who expects R9 300 series to not be GCN has not been paying attention to GCN architecture and how it was designed from day 1.

In case you are forgetting, AMD's HD5870 matched HD4870X2 in performance but HD5870 and 4870 are the same VLIW-5 architecture. AMD has been known to use the same architecture for 5+ years and just revise it by improving it over time and adding new features.

HD5870 vs. 4870:
"What you're seeing truly is a doubling of resources versus the RV770. Cypress has twice as many SIMD arrays in its shader core, twice as many texture units aligned with those SIMD arrays, double the number of render back-ends, and even two rasterizers. The big-impact number may be 1600, as in the number of shader processors or whatever AMD is calling them this week. 1600 ALUs, at any rate, bring a prodigious amount of compute power to this puppy." ~ Source

Why would AMD need a new architecture if they can still double HD7970Ghz and add HBM1 to it, incorporate Tonga's architecture, new UVD, HDMI 2.0, etc. The end result is 2X the performance of HD7970Ghz.

Surprise! That would be 104% on this chart.

So again, where are people on this forum getting this crazy idea that R9 390 series was post-GCN / brand new architecture? All signs have pointed that AMD will continue evolving GCN 1.0, 1.1, 1.2 --> 1.3/2.0, 3.0, etc.

Anyway, looks like you have even better info than all of us, but I think it's fair to say you have been underestimating R9 300 series, right?

With the extra time, AMD might surpase the gm200. But even if they dont, there shouldnt be a huge gap. AMD should be able to beat, catch, and/or almost catch maxwell on every metric. History has shown this.

Exactly. Despite NV having a history of building massive monolithic die, it never managed to beat AMD's best by more than 15-20% other than the failed 2900XT/3870 series. Worst case scenario is Titan X is 15-20% faster than R9 390X, however given how 980 is less than 10% faster than R9 290X at 4K, I think this time AMD has a legitimate shot to not only tie Titan X at 4K, but beat it. Granted, NV has a bag of tricks in its sleeve and it wouldn't be difficult for NV to launch a 1250mhz GM200 6GB card.

Sales on Hawaii are getting better and better, fire sale to clear inventory for 300 series?

$200 R9 290

Pretty much. I posted a couple times how here in Canada in major stores (NCIX and Canada Computers), the inventory of R9 290/290x is drying up and neither of these stores is re-ordering new units. Once a particular SKU sells out, it's done.

There are practically single digits of inventory left of some SKUs.

Just 2 Sapphire Tri-X 290s are currently left at Canada Computers across 27 stores. It does not look like the store has any plans to replenish this inventory.

 

[Silverforce11]

But it is really hard for me to accept that AMD, all of a sudden, cant design. History has shown that AMD has been going toe to toe for the many many years now. Fiji should be no different. It should be pretty freaking great.

With the extra time, AMD might surpase the gm200. But even if they dont, there shouldnt be a huge gap. AMD should be able to beat, catch, and/or almost catch maxwell on every metric. History has shown this.

Anyone who can recall the last few generations, AMD has steadily creeped up on NV in terms of their smaller die versus NV's big die on performance. Hawaii > GK110. That was the first time thier smaller die actually match & beat NV's big die. The trend is up for AMD in fact.

NV is early to the next-gen with Maxwell, but to not expect AMD to be competitive with their own next-gen (when we all know its a huge GPU with a major advantage: HBM) is plain silly and unfounded given historic proof.

I expect a minimum 50% above R290X performance for the same power at typical 1080/1440p. At 4K, that could blow up to be massive, especially min-fps, given the large latency & bandwidth advantage from HBM.

 

[raghu78]

Has there actually been any proof of new dies other than Fiji existing? Plenty of proof about Fiji, starting from that 550mm^2 leak last year but I've yet to see anything else that would imply a new die. Anything remotely reliable that is.

Also the stuff about Fiji being a lower end GPU with HBM (4GB) and bermuda being a higher end HBM one with (8GB), the theory that raghu at least bases his speculations on, seems to be most likely false.

http://cdn.videocardz.com/1/2015/04/Hot-Chips-Symposium-Fiji-2.5-HBM.png

The names could be different. But I will take a bet with you that we will see two dies with HBM. One massive 520 - 550 sq mm and the other around 380 - 400 sq mm

R9 390X - 4096 SP, 8 GB HBM, 4 shader,raster and tesselation engines (4 x 1024 = 4096 sp) , 256 TMUs, 8 ACE, 64 or 128 ROPs, 1/8 fp64 rate as Radeon and 1/2 fp64 rate as Firepro, 520 - 550 sq mm

R9 380X - 3072 sp, 4 GB HBM, 4 shader, raster and tesselation engines (4 x 768 = 4096 sp) , 192 TMUs, 4 or 8 ACE, 64 ROPs, 1/16 fp64 rate. 380 - 400 sq mm

R9 370X - 1536 sp, 2 /4 GB GDDR5, 2 shader, raster and tesselation engines ( 2 x 768 = 1536 sp), 2 or 4 ACE, 32 ROPs, 1/16 fp64 rate. 250 sq mm

Tonga was a testbed for a few of the architectural improvements (tesselation , fill rate, color compression) which will make its way to GCN 1.3 or GCN 2.0 found in R9 3xx series. Tonga has served its purpose both as R9 285 and the M9 295x in the iMac with Retina Display. But its a highly imbalanced product and inefficient in terms of both power and die size / cost. Do not make the mistake of judging the R9 3xx based on Tonga. :thumbsdown:

If AMD had a higher end GPU than fiji and fiji was going to be used for the 380X instead of the 390X they wouldn't be holding a presentation about fiji in august. They'd be doing a bermuda presentation.

Nobody except AMD knows the exact codenames. So Fiji with 8 GB HBM could be R9 390X and R9 380X with 4 GB HBM could be Bermuda or Grenada. We will know their codenames and the actual product names by June.

And the rest of the lower end leaks for mobile and for desktop have showed reasonable proof for rebrands. Even the leaked XFX double dissipation card had a pcb that was identical to hawaii PCBs from everything we could actually see on the card.

What proof ? Right now everyone is speculating based on leaks. nobody knows how reliable is anyone's information. Let me show a wccftech article on XFX R9 370 with TDP around 110-130w and 167mm PCB.

http://wccftech.com/xfx-radeon-r9-3...-feature-2-gb-gddr5-vram-launches-april-2015/

So there we have proof of Trinidad being a highly efficient product. Manhattan offsreen results render to a fixed resolution size and is the accurate performance indicator. Onscreen performance is affected by actual device screen resolution to which its rendering. R9 370 is 36% faster in Manhattan offscreen compared to R9 270. That puts this maybe ahead of even R9 285 . Anyway I looked up Manhattan offscreen results which were comparable to these scores. R9 270 score on the below link compared to the wccftech article could be due to drivers.

https://gfxbench.com/result.jsp?ben...=true&arch-ARM=true&arch-x86=true&base=device

http://www.anandtech.com/show/8666/the-apple-ipad-air-2-review/4

You can see a ipad Air 2 crushes the iphone 5s and iPhone 6 in Manhattan offscreen while falling behind in onscreen as it has to render to 2048x1536 resolution compared to much lower screen res on iphone 5s and iphone 6.

So you have Trinidad which is highly efficient and performance much higher than Pitcairn and maybe even R9 285. I expect Trinidad R9 370X to be faster than GTX 960 and R9 370 on par. btw getting the exact same Tonga chip to fit into a 110-130w desktop product is not gong to happen.

Personally I think that AMD's architecture improvements over tonga are limited to nonexistent and as such it didn't make any sense to design any more GPUs which would be only slight improvements over the old stuff, but expensive to design.

We'll likely see Fiji with almost identical GCN to tonga that gets its extra performance from the power savings from HBM, added die size and added space on the die caused by the change in memory controllers.

This is a stopgap generation anyway. The real deal will be when 16nm rolls out, both manufacturers are on HBM and we'll see which architecture actually is the superior one.

Again given your extreme Nvidia bias its no wonder you always see AMD failing and doing nothing to improve its competitive situation. I am saying (again from leaks) that GCN 1.3 or GCN 2.0 is a improved microarchitecture which improves IPC and efficiency.

https://translate.google.com/transl...ande-chip-monolitico-anche-per-amd&edit-text=

"The third new feature is the micro-architecture. With Fiji design GCN should fully embrace the "tiled architecture" going to review the organization for ALU thread within Compute Units in order to improve workload management."

 

[raghu78]

This is easily the least logical rumour of all time regarding R9 390X. And it basically fails in all areas:

RS you have better things to do than to respond to that. Anybody saying Fiji is a dual GPU card does not have a clue as to how AMD designs its GPUs products.

1) No credible website or in fact no website in the world in the last 1.5 years even once leaked the possibility of R9 390X being comprised of dual Tonga chips. The only time it was ever mentioned that Fiji XT was a dual-chip was by Fudzilla who confused the requirement for Virtual reality running smoother on 2 cards and automatically assumed the test rig had a single Fiji card with dual GPUs instead of dual Fiji cards.

and what does fud stand for ? see fudzilla is really bottom of the barrel , click bait trash site. I consider videocardz a better site. There are few sites who have any actual info on the next gen AMD products.

Even if we went with the idea that Fiji XT is a dual-GPU card, AMD would simply use dual Hawaii XTs with HBM1. Considering AMD already released R9 295X2 with 500W of power easily cooled by a single 120mm AIO CLC, there is no need at all to compromise on performance and go with mid-range Tonga chips. In the $700 market segment, performance rules, not power usage. No one will buy a $700 dual Tonga XT card that uses 200W of power but loses to an R9 295X2 90% of the time. The highest end is all about performance.

The R9 390X GPU is going to be a monster that its going to cause a lot of heartache for a few of the strongest Nvidia supporters and the forums will be lively for the next 12 months.

In fact I expect that the entire R9 3xx series is going to be a fitting reply to Maxwell that will surprise everyone.

All of these and other points were already discussed. Either you don't pay attention or you are purposely spreading FUD.

come on buddy you know it. So lets not feed the FUD.

Anyway, looks like you have even better info than all of us, but I think it's fair to say you have been underestimating R9 300 series, right?

this . over and over and over. :thumbsup:

Exactly. Despite NV having a history of building massive monolithic die, it never managed to beat AMD's best by more than 15-20% other than the failed 2900XT/3870 series. Worst case scenario is Titan X is 15-20% faster than R9 390X, however given how 980 is less than 10% faster than R9 290X at 4K, I think this time AMD has a legitimate shot to not only tie Titan X at 4K, but beat it. Granted, NV has a bag of tricks in its sleeve and it wouldn't be difficult for NV to launch a 1250mhz GM200 6GB card.

I am optimistic of AMD's chances of having the GPU crown even with OC. GF 28SHP is a less leaky process than TSMC 28HP and I would not be surprised to see excellent OC headroom. So judging R9 390X OC headroom based on R9 290X is not right. We should wait for the actual reviews and I expect partner products like Sapphire R9 390X Toxic to compete favourably with the best GM200 SKUs.

 

[gamervivek]

Didn't remember that there were benchmark scores for the 370 leak. And did you miswrote the digits there, it looks to be 64% faster here,

http://lensfire.blogspot.in/2015/01/amd-radeon-r9-370-benchmarks-and.html

 

[JDG1980]

The names could be different. But I will take a bet with you that we will see two dies with HBM. One massive 520 - 550 sq mm and the other around 380 - 400 sq mm

R9 390X - 4096 SP, 8 GB HBM, 4 shader,raster and tesselation engines (4 x 1024 = 4096 sp) , 256 TMUs, 8 ACE, 64 or 128 ROPs, 1/8 fp64 rate as Radeon and 1/2 fp64 rate as Firepro, 520 - 550 sq mm

R9 380X - 3072 sp, 4 GB HBM, 4 shader, raster and tesselation engines (4 x 768 = 4096 sp) , 192 TMUs, 4 or 8 ACE, 64 ROPs, 1/16 fp64 rate. 380 - 400 sq mm

R9 370X - 1536 sp, 2 /4 GB GDDR5, 2 shader, raster and tesselation engines ( 2 x 768 = 1536 sp), 2 or 4 ACE, 32 ROPs, 1/16 fp64 rate. 250 sq mm

Tonga was a testbed for a few of the architectural improvements (tesselation , fill rate, color compression) which will make its way to GCN 1.3 or GCN 2.0 found in R9 3xx series. Tonga has served its purpose both as R9 285 and the M9 295x in the iMac with Retina Display. But its a highly imbalanced product and inefficient in terms of both power and die size / cost. Do not make the mistake of judging the R9 3xx based on Tonga. :thumbsdown:

I agree about not judging GCN 1.2 based on Tonga alone. And if you want to judge it based on Tonga, then it should at least be the R9 M295X, not the far less efficient R9 285 (which, as I've said, I think is trash silicon from Apple-bound wafers).

That said, I do disagree somewhat with your guesstimates of what chips are going to be released. I think there's too much room in between your hypothesized 380X and 370X. Here's what I am speculating - at least, it's what I would have tried to do if I was in charge of AMD GPUs:

• Fiji: R9 290X. Pretty much what everyone is guessing; 4096 SPs with 8GB HBM and strong Double Precision performance in the FirePro version. TDP will be around 250W-300W; AIBs will do just fine with their existing coolers, but AMD's Water Cooled Edition will allow them to avoid having their terrible reference blower hurt their standing in charts. This card should be able to beat the Titan X by at least a couple of percent.
• Grenada: R9 380X. 3072 SPs with a 384-bit bus and 6GB of GDDR5. I disagree with you on this one, because I don't think a card like this would need HBM, or benefit particularly from it. Unlike with Hawaii, this card won't need full Double Precision capabilities, so this can be cut out to save die space. The bus can also be reduced in size compared to Hawaii because of GCN 1.2's delta color compression. That will also reduce die size and manufacturing costs. Power consumption and performance should come in at about the same spot as Nvidia's GTX 980 (180W or so).
• Antigua: R9 370X. This would be the least changed card, basically just "Tonga done right". 2048 SPs with a 256-bit bus and 4GB of GDDR5. The reference version would have only a single power connector, and slot in at about 120W-140W maximum power consumption.
• Trinidad: R9 360X. 1024 SPs with a 192-bit bus and 3GB of GDDR5. The bus width could be cut down from Pitcairn because of GCN 1.2's delta color compression, again reducing the transistor budget and die size. This card would be designed to compete with GM107 (GTX 750/750Ti), slotting in at a maximum TDP of 75W with no external power connector.
• All cards would, of course, use GCN 1.2 architecture (or better, if internal improvements have been made since then). They would be manufactured on GloFo 28nm SHP, assuming that 20nm or smaller processes are not yet viable. The cards would be upgraded with fixed-function HEVC decoding, similar to what currently exists in Nvidia's GTX 960 (AMD has this officially planned for Carrizo, so they probably already have a design that can be used.) I don't know how feasible this is, but if idle power consumption in multi-monitor mode can be toned down, this should definitely be done as well.
• And finally, the kicker:
• Bermuda: R9 395 X2. Two modified Fiji dice on a single interposer - with special hardware to allow both to communicate internally over an ultra-high-bandwidth link and present themselves to the system as one giant GPU with 8192 SPs. This would be the heaviest lift of all, but if AMD could somehow pull it off, it would absolutely demolish everything else...

btw getting the exact same Tonga chip to fit into a 110-130w desktop product is not gong to happen.

Why not? The existing R9 M295X Tonga (fully enabled) in Retina iMac has a maximum power consumption of about 125W. So it definitely can be done. That chip is fabbed on TSMC; port it over to GloFo 28nm SHP and do some tweaking, and I bet at least a few more watts can be squeezed out. There's probably some transistor budget to be saved (I suspect, though I can't prove, that there's some "junk DNA" in Tonga left over from Tahiti that could be removed).

 

[raghu78]

That said, I do disagree somewhat with your guesstimates of what chips are going to be released. I think there's too much room in between your hypothesized 380X and 370X. Here's what I am speculating - at least, it's what I would have tried to do if I was in charge of AMD GPUs:

• Fiji: R9 290X. Pretty much what everyone is guessing; 4096 SPs with 8GB HBM and strong Double Precision performance in the FirePro version. TDP will be around 250W-300W; AIBs will do just fine with their existing coolers, but AMD's Water Cooled Edition will allow them to avoid having their terrible reference blower hurt their standing in charts. This card should be able to beat the Titan X by at least a couple of percent.
• Grenada: R9 380X. 3072 SPs with a 384-bit bus and 6GB of GDDR5. I disagree with you on this one, because I don't think a card like this would need HBM, or benefit particularly from it. Unlike with Hawaii, this card won't need full Double Precision capabilities, so this can be cut out to save die space. The bus can also be reduced in size compared to Hawaii because of GCN 1.2's delta color compression. That will also reduce die size and manufacturing costs. Power consumption and performance should come in at about the same spot as Nvidia's GTX 980 (180W or so).

Of all the chips R9 380X is the one which requires HBM the most. For two reasons reducing power consumption and for being the flagship discrete mobile GPU which competes against GM204. You have to realize that AMD will not have any 14nm chips before Q3 2016. I doubt AMD are going to give up on the high end notebook GPU market. PCI-E MxM GPUs are restricted to chips with 256 bit memory bus

http://en.wikipedia.org/wiki/Mobile_PCI_Express_Module#2nd_generation_configurations_.28MXM_3.29

AMD does not have as efficient color compression as Nvidia Maxwell which is one of its third generation color compression technology. So for AMD to have a high end mobile GPU which can compete with GM204, HBM is the only way to go.

• Antigua: R9 370X. This would be the least changed card, basically just "Tonga done right". 2048 SPs with a 256-bit bus and 4GB of GDDR5. The reference version would have only a single power connector, and slot in at about 120W-140W maximum power consumption.

Sorry but Tonga imo has no role in the R9 3xx product stack. Its highly imbalanced in terms of resources and very inefficient in terms of perf/sq mm and perf/watt. There is no role for a 360 sq mm chip in the <= USD 200 market. AMD will replace Tonga with Trinidad which will have much better perf/sq mm and perf/watt.

• Trinidad: R9 360X. 1024 SPs with a 192-bit bus and 3GB of GDDR5. The bus width could be cut down from Pitcairn because of GCN 1.2's delta color compression, again reducing the transistor budget and die size. This card would be designed to compete with GM107 (GTX 750/750Ti), slotting in at a maximum TDP of 75W with no external power connector.

Again the leaks for the XFX R9 370 with TDP info show that desktop Trinidad is not a sub 75w product. My specs are likely to be the correct one. mobile Trinidad should be below 75w.

• All cards would, of course, use GCN 1.2 architecture (or better, if internal improvements have been made since then). They would be manufactured on GloFo 28nm SHP, assuming that 20nm or smaller processes are not yet viable. The cards would be upgraded with fixed-function HEVC decoding, similar to what currently exists in Nvidia's GTX 960 (AMD has this officially planned for Carrizo, so they probably already have a design that can be used.) I don't know how feasible this is, but if idle power consumption in multi-monitor mode can be toned down, this should definitely be done as well.

improved microacrchitecture than Tonga for better perf/sp , perf/sq mm and perf/watt. Improved video encoding / decoding. improved display connector support - HDMI 2.0 / Displayport 1.3

And finally, the kicker:

• Bermuda: R9 395 X2. Two modified Fiji dice on a single interposer - with special hardware to allow both to communicate internally over an ultra-high-bandwidth link and present themselves to the system as one giant GPU with 8192 SPs. This would be the heaviest lift of all, but if AMD could somehow pull it off, it would absolutely demolish everything else...

This will happen in Q4 2015 or early Q1 2016. my guess is AMD will have it in mid-late Q4 to catch the holiday shopping season.

Why not? The existing R9 M295X Tonga (fully enabled) in Retina iMac has a maximum power consumption of about 125W. So it definitely can be done. That chip is fabbed on TSMC; port it over to GloFo 28nm SHP and do some tweaking, and I bet at least a few more watts can be squeezed out. There's probably some transistor budget to be saved (I suspect, though I can't prove, that there's some "junk DNA" in Tonga left over from Tahiti that could be removed).

Just because there is a mobile Tonga with 125w does not mean its suitable for desktop cards. I expect AMD's next gen mobile flagship to be R9 380X and HBM based. There will be Trinidad based mobile SKUs which will easily be below 75w for mobile versions with perf competitive with mobile GTX 960 and much better perf/watt than mobile Tonga. :thumbsup:

 

[StereoPixel]

CodeXL 1.7

So, basically, new lineup is... only Fiji as new card?

Similar architecture to Tonga, just bigger. And with HBM.

It's just a software/driver bug. Fiji is GCN 1.3+ or 2.0.
Hawaii's W8100 was posted in CodeXL as Tonga.
http://www.elisanet.fi/kaotik/img-ftp/amdgpus.png

Dave Baumann from AMD said about W8100:

The CodeXL capture came out in May, W8100 was released in July. You don't make a product change that late. If one is wrong, which is it more likely to be - software or hardware...?

 

[Spanners]

When did this forum turn into 4chan?

A single image macro doesn't really turn the forums into anything, but I'm sure you'll get agreement from the Mods as I've seen them use that exact phrase in the past, uncoincidentally.

 

[Head1985]

Info from Obr:
Hawaii with HBM
390x wont beat TITANX.
390x 40% Faster than 290x
390x 15%Faster than GTX980
Why is Fiji so late?Because NV launched TITANX with 12GB so fast that AMD needs redone Fiji to use with 8GB.
Fiji should be here for months now with original schedule.

 

[5150Joker]

Info from Obr:
Hawaii with HBM
390x wont beat TITANX.
390x 40% Faster than 290x
390x 15%Faster than GTX980
Why is Fiji so late?Because NV launched TITANX with 12GB so fast that AMD need redone Fiji to use with 8GB.
Fiji should be here for months now with original schedule.

Who or what is Obr? And you can't just go and retool something like that in a matter of months to go from 4 GB to 8 GB. Sounds like a bunch of bologna to me.

 

[Azix]

AMD doesn't really need to aim for titanx. Maybe 980 ti at most.

then again if its the full maxwell, maybe they do. Price ignored

 

[raghu78]

Info from Obr:
Hawaii with HBM
390x wont beat TITANX.
390x 40% Faster than 290x
390x 15%Faster than GTX980
Why is Fiji so late?Because NV launched TITANX with 12GB so fast that AMD needs redone Fiji to use with 8GB.
Fiji should be here for months now with original schedule.

And this is why its absolute rubbish. The design of such complex GPU chips cannot be redone in the space of few months and that too so close to release. The memory controller for R9 390X must have been designed from the start with the dual link interposer tech and ability support 8 GB HBM.

My take is clear. R9 390X with 4096 sp and 8 GB HBM beats Titan-X and is 55-60% faster than R9 290X. R9 380X with 3072 sp is 15% faster than GTX 980. wanna take a bet. :thumbsup:

 

[5150Joker]

And this is why its absolute rubbish. The design of such complex GPU chips cannot be redone in the space of 3 months. The memory controller for R9 390X must have been designed from the start with the dual link interposer tech and ability support 8 GB HBM.

My take is clear. R9 390X with 4096 sp and 8 GB HBM beats Titan-X and is 55-60% faster than R9 290X. R9 380X with 3072 sp is 15% faster than GTX 980. wanna take a bet. :thumbsup:

You also stated you wish to bet that with R390/390X AMD will regain a discrete gpu market share of 35-40%. Now how long do you suppose a market impact of R390/390x would take to show from JPR? One or two quarters? We can revisit these posts around February 2016 and see what the JPR results show. If it shows that AMD hardly made a dent, don't toss out excuses, just eat crow and accept you had an overly optimistic outlook of it all. If I'm wrong, I'll more than gladly admit I was off the mark.

 

[jpiniero]

1) No credible website or in fact no website in the world in the last 1.5 years even once leaked the possibility of R9 390X being comprised of dual Tonga chips.

Do you really think AMD can afford to make a chip the size of a theoretical 4096 core Tonga would be though? It'd be over 550 easy if it were on 28 nm. AMD doesn't have the fanbois to justify the costs. That's why the dual Tonga idea makes sense since they could sell a single die 'mobile' version to Apple (for the Broadwell iMac) and perhaps other OEMs.

 

[Bobisuruncle54]

Do you really think AMD can afford to make a chip the size of a theoretical 4096 core Tonga would be though? It'd be over 550 easy if it were on 28 nm. AMD doesn't have the fanbois to justify the costs. That's why the dual Tonga idea makes sense since they could sell a single die 'mobile' version to Apple (for the Broadwell iMac) and perhaps other OEMs.

None of that makes sense. AMD needs to present a brand new lineup that Nvidia can't respond to until Pascal is ready.

 

[Techhog]

Info from Obr:
Hawaii with HBM
390x wont beat TITANX.
390x 40% Faster than 290x
390x 15%Faster than GTX980
Why is Fiji so late?Because NV launched TITANX with 12GB so fast that AMD needs redone Fiji to use with 8GB.
Fiji should be here for months now with original schedule.

So, that would make it like 30% slower than the Titan X and 980 Ti? If it costs more than $500 and uses more power than Hawaii, they might as well throw in the towel and stick to well-priced mid-range GPUs. They're never going to catch up.

 

[ShintaiDK]

Financially they dont need to "catch up". And they dont really have the R&D for it anyway.

Before the massive rebrand, it was also deemed impossible that it wouldnt all be new GPUs.

 

[jpiniero]

None of that makes sense. AMD needs to present a brand new lineup that Nvidia can't respond to until Pascal is ready.

AMD's kind of broke you know. It'd be nice if they shrank to 20 nm but it'd only be for the power savings since the $/transistor isn't any cheaper than 28 nm.

 

[Techhog]

Financially they dont need to "catch up". They dont have the R&D for it anyway.

Before the massive rebrand, it was also deemed impossible that it wouldnt all be new GPUs.

Some of us want competition. I know you're in favor of the entire computing world being dominated by a single company, but that's not how I feel. This would likely mean that AMD will permanently be a generation behind, and that's just terrible. It destroys brand perception and ensures that they'll be gone before the end of 2017. (which is why you think it's fine, obviously)

I also want to note that I don't care about your opinion on anything because of that bias of yours. (And no, I'm not talking about Nvidia)

 

[ShintaiDK]

Do you think you get competition due to whatever the 390X performs and cost?

Or is the actual competition part happening in another lower segment?

 

[raghu78]

You also stated you wish to bet that with R390/390X AMD will regain a discrete gpu market share of 35-40%. Now how long do you suppose a market impact of R390/390x would take to show from JPR? One or two quarters? We can revisit these posts around February 2016 and see what the JPR results show. If it shows that AMD hardly made a dent, don't toss out excuses, just eat crow and accept you had an overly optimistic outlook of it all. If I'm wrong, I'll more than gladly admit I was off the mark.

sounds fair. Lets say if AMD gets to 35 - 40% by Q4 2015 you admit you were wrong. If not I will admit I am wrong. But let me say at the outset you write off AMD only at your risk. They have been extremely competitive with each and every generation and whatever they have been building is going to be very competitive. AMD has a huge first mover advantage to HBM and the last time AMD got first to a memory standard (GDDR5) Nvidia did not catch up for 21 months. I predict a 18+ month lead this time.

Nvidia's first FINFET chips are rumoured to be dumb Maxwell shrinks to reduce the risks of transition to FINFET. The last time Nvidia tried a new architecture with a new memory controller and a new process node it was a disaster - GF100 aka GTX 480. It took another 6+ months and a new base revision GF110 aka GTX 580 to fix the problems. AMD has chosen the right time to transition to HBM - high yielding and very mature 28nm node. So I would expect Nvidia to come out with Maxwell 16/14nm shrinks and then make the transition to Pascal, HBM and NVLink on 16/14nm FINFET.

 

[5150Joker]

Some of us want competition. I know you're in favor of the entire computing world being dominated by a single company, but that's not how I feel. This would likely mean that AMD will permanently be a generation behind, and that's just terrible. It destroys brand perception and ensures that they'll be gone before the end of 2017. (which is why you think it's fine, obviously)

I also want to note that I don't care about your opinion on anything because of that bias of yours. (And no, I'm not talking about Nvidia)

AMD hasn't been competitive in things like notebooks for over 3 years yet the world is still turning. And you obviously care about his opinion seeing how you replied to his post.

 

[Kenmitch]

Do you think you get competition due to whatever the 390X performs and cost?

Or is the actual competition part happening in another lower segment?

Pricing competition based on performance of AMD's new offerings at the minimum. Most won't buy AMD's offerings but silently wish they hit it out of the park so they can get a better deal on the inferior NVidia product....Nothing changed.

 

[Maxima1]

AMD hasn't been competitive in things like notebooks for over 3 years yet the world is still turning. And you obviously care about his opinion seeing how you replied to his post.

Yeah, and you notice the premium you pay for when you go Nvidia.

Dell's Broadwell Latitude 3550 has this junk for similar price as the old model

NVIDIA GeForce 830M 2GB Graphics (DDR3)

Dell's Haswell model had this

AMD 8850M 2GB GDDR5