Another 6 months before VEGA?

razor1 said:
And I have no idea how you have come up with Vega 11 and 10 are the same performance. There are videos of AMD saying flagship GPU. That is big Vega.

By all accounts, Anarachist, AMD isn't know for not showing anything else than the best they can or talking about things in the best way possible when it comes to theoretical numbers, or what the have at hand. I expect no different here. That is twice they have shown actually, three times they have shown, around gtx 1080 performance for their "flagship GPU".
Click to expand...
I just find it incredibly unlikely they had a huge performance regression from Polaris to Vega. I'm not the one implying they're the same performance.

Shintai said:
No, its you that seem to mess things together. Vega 11 will fit somewhere between Vega 10 and Polaris 10.
Click to expand...
You're the one that said Vega 10 is on par with a 1080, yet a revamped Polaris is up against 1070. So what, Vega 11 is between 1070/1080 and Vega 10 is equal to 1080? That's really bad for two chips. Might as well have just canned Vega 11 at that point if small and large perform equally.
 
Anarchist4000 said:
I just find it incredibly unlikely they had a huge performance regression from Polaris to Vega. I'm not the one implying they're the same performance.
Click to expand...


Its not a regression that's 100% the performance of Polaris (P10). And that is if you had 100% scaling of all units with out the doubling of power, so you end up with a bit more efficiency.
 
Anarchist4000 said:
You're the one that said Vega 10 is on par with a 1080, yet a revamped Polaris is up against 1070. So what, Vega 11 is between 1070/1080 and Vega 10 is equal to 1080? That's really bad for two chips. Might as well have just canned Vega 11 at that point if small and large perform equally.
Click to expand...

Who said a revamped Polaris is up against 1070? Polaris 10 will be rebranded as 500 series as well tho.

With your logic, Polaris should never have been released either.
 
Last edited:
Shintai said:
So said a revamped Polaris is up against 1070?
Click to expand...

Yeah that is a pipe dream, 30% more performance lol at the same power usage where P10 needs 30% greater memory bandwidth than a 1070 and 25% more computational performance.
 
razor1 said:
Yeah that is a pipe dream, 30% more performance lol at the same power usage where P10 needs 30% greater memory bandwidth than a 1070 and 25% more computational performance.
Click to expand...

Yep, just fixed the typo as well. Vega 11 looks to be what goes against 1070.
 
Shintai said:
Who said a revamped Polaris is up against 1070? Polaris 10 will be rebranded as 500 series as well tho.

With your logic, Polaris should never have been released either.
Click to expand...
Shintai said:
https://videocardz.com/66253/amd-radeon-rx-580-ashes-of-the-singularity-results-leaks-out

RX580 ~ GTX1070.
Click to expand...
Videocardz said:
What’s RX 580? Is it a new chip? Is it a rebranded and overclocked RX 480? I don’t think we can answer this yet. Ashes of the Singularity benchmarks are hard to compare, but there are plenty of results with GTX 1070 in this area (7100 points in Standard 1080p preset).
Click to expand...
It looks like you did. Or maybe I'm misintrepreting your 580 equivalent to 1070 and article you linked saying as much?
 
Shintai said:
You are comparing 2 different games. AOTS and AOTS Escalation. 1.5 was released in december.
http://forums.ashesofthesingularity.com/481330/ASHES-DEV-UPDATE-January-2017

Tho benchmark performance wise there seem to be no difference.
Click to expand...

Ah cool and thanks :)
So that means the GTX1070 is still faster than the 580, here is a benchmark with the 2.12 version that is using the next up High setting and still around 10% faster that the 580 score, also using a 5820K.
http://ashesofthesingularity.com/me...-details/64bef3e2-be5c-4992-9613-4b1ce43a65d1

So still not as fast as that GTX1070 that is at a higher setting, why I asked about the versions.
But as you appreciate and to just say generally:
AoTS is only ideal when done by reviews that gives a reasonable baseline reference mark and their methodology-tools-system.
Videocardz jumped the gun saying it matches GTX1070 as the AMD score should be compared against the better GTX1070 on same CPU.
Better GTX1070 scores because the 580 is a single score but importantly done by AMD so the environment would not be messed up, score improvements will possibly come with drivers/ but it should be compared to the better GTX1070 scores as we do not know if the clocks are low/high/at spec for the 580 but probably at spec.
 
Last edited:
pendragon1 said:
dude, stop. that is still a totally different version, different settings...
Click to expand...
You read where Shintai says the benchmark is the same and it also got an uptick?
The settings are different with them being higher for the GTX1070 and it still performed faster, they both used the 5820K.
You did notice I asked about the versions and received a response before my post :)
Anyway settings are not an issue in my context that was explicitely pointing out the higher settings with the 1070 was still faster and by around 10%, which goes against what the headlines were inferring in terms of expectations.
TBH it makes more sense that AoTS is not used outside of reviews, but plenty like it when it shows such results in a leak against the slower scores and ignore faster ones using same CPU, hyped up by some sites (why I feel videocardz has jumped the gun with their headline and skews early expectations).
Cheers
 
Last edited:
NKD said:
https://videocardz.com/66306/amd-demonstrates-radeon-vega-in-star-wars-battlefront-again

so in this demo today we finally see the vega card running with a basic heatsink and debugging pcb attached. The actual card is so frickin short. I mean if that is flagshipped vega color me impressed. because my case won't fit anything larger than the motherboard length as it hits the powersupply. So I am intrigued by the length of that card.
Click to expand...


Well it uses HBM, why have a large area foot print when using that?
 
razor1 said:
Well it uses HBM, why have a large area foot print when using that?
Click to expand...

well I thought it would require larger heatsink to cool that shit, lol. But may be its vapor chamber cooler they got under the hood.
 
NKD said:
well I thought it would require larger heatsink to cool that shit, lol. But may be its vapor chamber cooler they got under the hood.
Click to expand...


If its using around 250 watts, is what I'm expecting, shouldn't need anything larger than whats on a 390x.
 
NKD said:
https://videocardz.com/66306/amd-demonstrates-radeon-vega-in-star-wars-battlefront-again

so in this demo today we finally see the vega card running with a basic heatsink and debugging pcb attached. The actual card is so frickin short. I mean if that is flagshipped vega color me impressed. because my case won't fit anything larger than the motherboard length as it hits the powersupply. So I am intrigued by the length of that card.
Click to expand...
i posted that vid in this thread earlier, post #429. the pcb is clearly at least(edit: ok maybe its about an inch) an inch longer than the mobo. then the debug pcb sticks off the back. mobo is a standard atx so its ~9.6", so card is ~10.5-11" so I think you might still be SOL.
upload_2017-2-22_19-44-43.png
 
Last edited:
pendragon1 said:
i posted that vid in this thread earlier, post #429. the pcb is clearly at least(edit: ok maybe its about an inch) an inch longer than the mobo. then the debug pcb sticks off the back. mobo is a standard atx so its ~9.6", so card is ~10.5-11" so I think you might still be SOL.
View attachment 17627
Click to expand...
I have little clearance but 1070 and 1080 were both too long. We will see how this pans out. I have rx 480 right now. Sold off my 1070 since I have freesync monitor. I will stay optimistic.
 
pendragon1 said:
i posted that vid in this thread earlier, post #429. the pcb is clearly at least(edit: ok maybe its about an inch) an inch longer than the mobo. then the debug pcb sticks off the back. mobo is a standard atx so its ~9.6", so card is ~10.5-11" so I think you might still be SOL.
Click to expand...

That's just an engineering sample so not much point worrying about it too much. Decent chance the final card will be shorter, especially AIB versions
 
1070 performance is pretty @#$%@# bad for this long a wait. And it won't be cheap with HBM2.

Another DOA.
 
if its at 1070 performance levels, there is no use of even making this card, cause Fury is already at 1070 performance levels. I think its going to gtx 1080 levels or a bit higher. not too much higher though.
 
  • Like
Reactions: N4CR
like this
razor1 said:
if its at 1070 performance levels, there is no use of even making this card, cause Fury is already at 1070 performance levels. I think its going to gtx 1080 levels or a bit higher. not too much higher though.
Click to expand...

Fury is equal to a 1070 ehh.
 
outside of power consumption yeah, its close, about 5-10% variation. (Fury X)
 
I would expect leaks that are more inline with what is going to be released in March. By then AIBs should be very much in progress of making them.

Now if the full version Vega 11 competes well with the 1070 that would be good since then AMD would have a very solid in-between card from the 480/1060 and the 1070 - hitting that low $300 segment which is absent now.

Nvidia would be forced to lower the 1070 price (good for us) and/or have a refresh of the Pascal line.

I almost find what Nvidia could do if Vega is strong is way more interesting. They have so many skews already in the market place plus so many new Pascal skews they can launch. Titan Black (full GP-102, $1200), Titan Ultimate (GP-100 with HBM2 memory, $1500) of course Ti 1180 Ti and then possible refresh of the top end Pascal cards. Then Nvidia could just slash prices to ridicules levels (except that would be very rare for Nvidia to do and I would not expect that). I really don't see Nvidia afraid to test the waters with an ultra high end and expensive gaming/professional card, AMD has already done that with the Fury Pro Duo sort of but kinda chicken out as well as a hyper gaming card since it was not in the end.
 
{NG}Fidel said:
Fury is equal to a 1070 ehh.
Click to expand...

Depends upon ones context, 4K average across many games they are reasonably close and like Razor says around 10% for the GTX1070 across diverse range of games at 1440p, but that increases to 20% with custom AIB cards at 1440p and these should be the primary focus on comparison as consumers mostly buy AIB rather than blower reference type cards as the AIB are usually cheaper/faster (at manufacturing stock and without throttling or without requiring excessive parameter changes)/quieter.
The problem is that the Fury X was pretty much limited in what could be done to it.
https://www.computerbase.de/thema/grafikkarte/rangliste/#diagramm-performancerating-2560-1440

Personally I prefer where possible AIB for both manufacturers as the 'reference' models rarely give us a true indicator of the GPU performance even when not extreme OC'd.
Cheers
 
CSI_PC said:
Personally I prefer where possible AIB for both manufacturers as the 'reference' models rarely give us a true indicator of the GPU performance even when not extreme OC'd.
Click to expand...

Sorry but what does that mean? Are you saying reference cards are not representative??? Come off it, I don't believe that.
 
Syphon Filter said:
Sorry but what does that mean? Are you saying reference cards are not representative??? Come off it, I don't believe that.
Click to expand...
Really?
You realise you have to really tweak parameters for reference models from both AMD and Nvidia to actually stop them throttling from either temps or power?
And for reviews the point is to not change such parameters for their general performance reference table as it totally changes the context of such performance/behaviour.
Anyway if you had clicked on the link I provided you would see:
1070FE at 1440p is 11% faster than Fury X.
Gigabyte 1070 Gaming is 21% faster than Fury X.

So reference models are not representative.
Cheers
 
Syphon Filter said:
You sure you're comparing apples to apples there?

Gigabyte gaming models carry OC from factory.
Click to expand...
Sigh.
We are not talking about the extreme performance AIBs here.
And you read my original post where I stated the problem with the Fury X is that unfortunately there are not much improvements IHV can do to this.
If you think you can run a reference 1080 or 1070 without changing the temp and power targets without throttling you are wrong, same goes with the AMD 480.
All IHV carry some base OC but it is primarily the cooling/indirectly power-temp targets that help to really carry them if one does not change parameters on the reference design.
Computerbasede has done plenty of articles showing such issues, so has PCGameshardware.

Just because Fury X cannot be purchased as an improved IHV product is not the fault of Nvidia when it is compared to AIB 1070s.
Same can be said about Nvidia Pascal Titan compared to custom IHV GTX1080s, but you do not hear people complaining about that being unfair.
Or the 980ti custom models compared to various Nvidia reference Titan designs.

Please go back and read my OP.
The custom AIB performance results are more relevant to most consumers and reflects the performance of a GPU, whether that be AMD or Nvidia and this is excluding overclocking behaviour.
Thanks
 
Last edited:
CSI_PC said:
1070FE at 1440p is 11% faster than Fury X.
Gigabyte 1070 Gaming is 21% faster than Fury X.
Click to expand...

That's what you said. I'm saying The 1070 Gaming is OCed.

EDIT: I get what you're saying. My question is should people compare OC cards to non-OC cards?

I'm not having a go, you're getting defensive for some reason. I'm just asking a question.
 
Last edited:
Syphon Filter said:
That's what you said. I'm saying The 1070 Gaming is OCed.

EDIT: I get what you're saying. My question is should people compare OC cards to non-OC cards?

I'm not having a go, you're getting defensive for some reason. I'm just asking a question.
Click to expand...

And you missed my context.
You normally don't compare custom AIB to reference, but cards such as Fury X and Titan are only reference designs, with minimal improvements from AIBs.
If the manufacturers do not provide options for the design to be improved, then it is still relevant it should be compared to the average custom AIBs of other models.

This is part of the problem, people go on about how a Fury X is close to a 980ti or 1070, but most by a large margin buy the custom AIB models of those products because they are quieter and do not throttle by default.

You really should read just how much reference designs throttle without changes to power target/fan profiles/etc, and how well they sell compared to custom AIBs with better cooling/settings.
The only reason to use reference design is to use the weakest performance , and as my point does not indicate the GPU as it is hobbled due to cooling/defaults, the OC assositaed with AIBs when it is not the extreme expesnive custom models is not excessive.

Cheers
 
Last edited:
290x ref sorta sucked without turning fans up, even then... it got hot. Aftermarket was like a whole new card, big averages merely by maintaining or slightly exceeding stock rated clocks. Ref cards would throttle badly. We saw that with the reference edition 1080 too.
 
With Pascal Boost clock, it's paramount the Power Target is set as high as possible as default, and temps kept as low as possible, to promote the highest boost bin. Even without factory OC, just better power delivery, higher default Power Target and lower temps from custom cards will allow most Pascal cards to boost close to 1.9GHz by default, IIRC.

Which is massive, compared to the base and advertised boost clocks.
 
Here's a tasty bit from Ars
https://arstechnica.com/information...essor-more-cores-bandwidth-memory-than-intel/
In two-socket systems, 64 of the PCIe lanes are used for inter-socket communication. That leaves 64 from each socket available for I/O. The inter-socket communication uses AMD's "Infinity Fabric," the (somewhat ill-defined) high-speed cache coherent interconnect that's also used within Zen. The processor uses the same pins for PCIe as it does Infinity Fabric.

Alternatively, those same I/O channels can be used for direct-attached GPUs (which is to say, not using PCIe). That's comparable to what Nvidia is doing with its NVLink interconnect.
Click to expand...

If this same tech is enabled in the desktop then a Ryzen+RX Vega combo might have some secret sauce.
 
ecmaster76 said:
Here's a tasty bit from Ars
https://arstechnica.com/information...essor-more-cores-bandwidth-memory-than-intel/


If this same tech is enabled in the desktop then a Ryzen+RX Vega combo might have some secret sauce.
Click to expand...
Unfortunately you will only see this on server motherboards, considering how much the upper X370 motherboards are (and these are pretty simplistic in comparison) this would not be cheap for consumers even if they did consider it,which I doubt because it would be too niche as well with its costs that would be much higher than X370 boards.
Same way the closest we see NVLink outside of HPC is the Quadro GP100 (expensive card) with its more simplified NVLINK (replacing SLI) rather than also putting it on the Pascal Titan that is not cheap for consumers anyway.
And that is a simplified NVLINK not the full mesh GPU-GPU-CPU.
Funny enough that Quadro GP100 is on sale and available around same time as the 1080ti, yeah none of us are going to buy that Quadro though, check its price :)

Cheers
 
Last edited:
Syphon Filter said:
That's what you said. I'm saying The 1070 Gaming is OCed.

EDIT: I get what you're saying. My question is should people compare OC cards to non-OC cards?

I'm not having a go, you're getting defensive for some reason. I'm just asking a question.
Click to expand...

You have to be reasonable. FE Pascal cards tend to settle at around 1850 mhz, any Aib version with better cooling will sit at 2ghz.

Reference 980ti settled at around 1130mhz, if you have a dud card it will "only" hit 1400. That's a 25% bump, reviews don't necessarily have to account for that, but readers should be aware that when they see Fury X and 980ti neck and neck that one can gain 25% extra performance from simple OCing and the other cannot
 
You must log in or register to reply here.
Back
Top