Sapphire Launches its Radeon VII, AMD Reserving 1.80 GHz Boost for Direct Sales?

[btarunr]

Sapphire is among the first AMD add-in-board (AIB) partners to launch a Radeon VII graphics card. The card sticks to AMD reference board design, which the company unveiled at its CES 2019 keynote. Interestingly, its GPU engine boost frequency is set at 1750 MHz, which is less than the 1800 MHz boost frequency figure that was mentioned by the company earlier. Could it be that AMD is reserving 1800 MHz for cards directly sold on AMD.com? The memory frequency is unchanged at 1000 MHz, which works out to an HBM2 memory bandwidth of 1 TB/s. Sapphire's box for this card lists out key specifications upfront, and also features the Vega II logo. It's likely that the card will be sold at the baseline price of $699, given that there are no other variants of this card, not even custom-design.



View at TechPowerUp Main Site

 

[kastriot]

If this card was 499$ with GDDR5 it would sold much better, this way AMD is just buying some time for navi i guess.

 

[Nkd]

If this card was 499$ with GDDR5 it would sold much better, this way AMD is just buying some time for navi i guess.

NO point. at that rate just buy a vega 64 or even a vega 56. it needs that memory throughput for the performance increase. Vega is desperate for that lol.

Also Can we please stop calling it vega II? I understand that is what it looks like but AMD clearly stated its Radeon 7 not labeled as Vega 2. I don't really care but it would be nice to represent the card as what it actually is called.

 

[ratirt]

NO point. at that rate just buy a vega 64 or even a vega 56. it needs that memory throughput for the performance increase. Vega is desperate for that lol.

Also Can we please stop calling it vega II? I understand that is what it looks like but AMD clearly stated its Radeon 7 not labeled as Vega 2. I don't really care but it would be nice to represent the card as what it actually is called.

Totally agree. People tend to twist words and call stuff what they think it is despite the original though or meaning. GDDR5 in Vega would've been a disaster. Even though I really wanna look at the benchmarks and I'm curious is it just the shrink and clock boost or there's more in that new card. If it has a 2080 performance I think I will switch to this new AMD card.

 

[ShurikN]

ASRock also has a "custom sticker" version, but no boost info on that one for now.

 

[Nkd]

Totally agree. People tend to twist words and call stuff what they think it is despite the original though or meaning. GDDR5 in Vega would've been a disaster. Even though I really wanna look at the benchmarks and I'm curious is it just the shrink and clock boost or there's more in that new card. If it has a 2080 performance I think I will switch to this new AMD card.

Same here. I was going to grab an RTX card but all the good ones seem to be closer to 800. and I am terrified of the space invaders, I guess I am just paranoid like that. I wanna see how good it holds boost clocks, I think if we can tweak it to hold max boost clocks it might be faster then benchmarks like vega. Also I hope that HBM2 an be pushed another 100-200mhz. On top I want division 2 anyways.

 

[TheGuruStud]

Same here. I was going to grab an RTX card but all the good ones seem to be closer to 800. and I am terrified of the space invaders, I guess I am just paranoid like that. I wanna see how good it holds boost clocks, I think if we can tweak it to hold max boost clocks it might be faster then benchmarks like vega. Also I hope that HBM2 an be pushed another 100-200mhz. On top I want division 2 anyways.

Stock mem clock should be fine. I don't think there could be any gaming benefit over say 750GB/s. Save that power for gpu clock.

 

[ratirt]

Same here. I was going to grab an RTX card but all the good ones seem to be closer to 800. and I am terrified of the space invaders, I guess I am just paranoid like that. I wanna see how good it holds boost clocks, I think if we can tweak it to hold max boost clocks it might be faster then benchmarks like vega. Also I hope that HBM2 an be pushed another 100-200mhz. On top I want division 2 anyways.

I'm sure it will be faster than Vega. ( but deep down inside I still just hope it will be faster) The question remains. how much faster? 2080's performance and I'm getting one for sure. RTX's are out for me. I'm done with Nvidia.

 

[FordGT90Concept]

Was anyone able to verify the card was actually running at 1800 MHz at CES? 1750 MHz rounds to 1.8 GHz...

1.8 GHz could also be binned chips that Sapphire isn't getting shipments of. Maybe Sapphire manufactured the reference cards and these are the left overs?

 

[Patriot]

Was anyone able to verify the card was actually running at 1800 MHz at CES? 1750 MHz rounds to 1.8 GHz...

1.8 GHz could also be binned chips that Sapphire isn't getting shipments of. Maybe Sapphire manufactured the reference cards and these are the left overs?

Sapphire is the primary odm for reference cards. They also make all the instinct and radeon pro cards.

 

[londiste]

Seems to be missing a source link:
http://www.sapphiretech.com/productdetial.asp?pid=97FA4A9F-D64C-45E6-A0C4-B8911A7EE2B3&lang=eng
Specs are unlikely to be different between vendors and even AMD themselves. This is a reference card with stickers.

Besides, this matches perfectly with what AMD presented. Remember, 1800MHz is peak clock and later it was clarified that Peak is higher than Boost.
- Base: 1400MHz
- Boost: 1750MHz
- Peak: 1800MHz

 

[fullinfusion]

If this card was 499$ with GDDR5 it would sold much better, this way AMD is just buying some time for navi i guess.

If you say so. Gddr5 and 5x can't compare to the direct interaction between the hbm plus look at how much realistate it take! And they aren't buying time friend, they're just making your Navi right. Navi? Only you can figure it out on what it'll bring huh. You might be surprised.

 

[bibob94]

i think.. i should link you guys this

, hes make a good point there..

 

[Turmania]

2018 was a terrible year for GPU. Unfortunately 2019 started where 2018 left off.hoping second half would bring excitement with Navi and then maybe Nvidia being forced to bring out 7nm versions.

 

[xtreemchaos]

too much money for me, £500 is my max id pay for a single gpu, i dont care how good or fast it is.

 

[xkm1948]

Wonder whether AMd will be sending these out for reviews.

 

[ArbitraryAffection]

Honestly I don't believe 1800 MHz really as a usual operating clock because in the Vega 10 Whitepaper it says that the chip reaches 1700 MHz. Which it can, but requires a lot of manual tuning, and overclocking and not all chips can reach it. I was honestly expecting the Vega 20 parts to run around 1600-1700 Mhz in games tbh.

 

[Valantar]

Really looking forward to benchmarks and seeing what clocks this actually ends up at. Don't know whether to be optimistic and hope for high sustained boost clocks due to 7nm, or to take the TDP and "peak boost clock" labeling as a bad sign and expect this to be just a minor boost over Vega 64 despite the node shrink. Doubling the HBM2 stacks should only represent a <20W power draw increase, so that should be negligible overall. Stock Vega 64 on air boosts to around 1500MHz (1479-1557) in games according to AnandTech's review (perhaps better with tweaking, but that's hardly applicable given that <1% of users are likely to do so), so this would need to sustain at least 1700MHz to represent a meaningful increase - and even that would be rather weak at 13% (not accounting for any perf increases due to memory bandwidth, of course). 1800MHz is only a 20% increase from 1500MHz, after all - at the same power draw with a full node shrink, that's ... worrying.

I really, really, really hope Navi brings some proper architectural efficiency improvements. Nvidia did it with Maxwell, so it must be possible, and I'd love an upgrade candidate for my Fury X to actually arrive at some point.

 

[ArbitraryAffection]

Really looking forward to benchmarks and seeing what clocks this actually ends up at. Don't know whether to be optimistic and hope for high sustained boost clocks due to 7nm, or to take the TDP and "peak boost clock" labeling as a bad sign and expect this to be just a minor boost over Vega 64 despite the node shrink. Doubling the HBM2 stacks should only represent a <20W power draw increase, so that should be negligible overall. Stock Vega 64 on air boosts to around 1500MHz (1479-1557) in games according to AnandTech's review (perhaps better with tweaking, but that's hardly applicable given that <1% of users are likely to do so), so this would need to sustain at least 1700MHz to represent a meaningful increase - and even that would be rather weak at 13% (not accounting for any perf increases due to memory bandwidth, of course). 1800MHz is only a 20% increase from 1500MHz, after all - at the same power draw with a full node shrink, that's ... worrying.

I really, really, really hope Navi brings some proper architectural efficiency improvements. Nvidia did it with Maxwell, so it must be possible, and I'd love an upgrade candidate for my Fury X to actually arrive at some point.

Don't hold your breath for ground-breaking efficiency gain with Navi, I think those will come with the post-GCN design. Unless Navi Is that design, I did hear some rumours about a Navi 2560 Stream processor chip reaching Vega 64 performance levels. (many grains salt needed). but I don't believe it in all honesty. Also Vega 20 will have some "per core" gains over Vega 10 and will probably be somewhat faster "per clock" than Vega 64 even tho this SKU uses fewer CU's, because of the additional bandwidth I think. Also there's no way it will reach the claimed 2080 performance bracket without some "per core" improvements.

 

[nemesis.ie]

No surprise Sapphire havea product out, they are the ODM for AMD branded cards AFAIK.

Was anyone able to verify the card was actually running at 1800 MHz at CES? 1750 MHz rounds to 1.8 GHz...

1.8 GHz could also be binned chips that Sapphire isn't getting shipments of. Maybe Sapphire manufactured the reference cards and these are the left overs?

No, there is no rounding from MHz to GHz. It's not like bits/bytes with 1024 to a MB. It's standard 1000s I think. i.e. 1750MHz == 1.75GHz.

If you just mean "standard rounding" where you round up at a 1/2, maybe, but I think that's still a stretch, especially when they always quote MHz as the number looks bigger sand it is marketing ...

I'm surprised they done put 1,750,000,000Hz.

Incoming: Next nV product will be listed as "2,500,000,000 Hz. JH ("LJ" to his mates) is already quoting 1/4 FP numbers for the TFLOPS of their cards instead of the usual 32bit ...

 

[Valantar]

Don't hold your breath for ground-breaking efficiency gain with Navi, I think those will come with the post-GCN design. Unless Navi Is that design, I did hear some rumours about a Navi 2560 Stream processor chip reaching Vega 64 performance levels. (many grains salt needed). but I don't believe it in all honesty. Also Vega 20 will have some "per core" gains over Vega 10 and will probably be somewhat faster "per clock" than Vega 64 even tho this SKU uses fewer CU's, because of the additional bandwidth I think. Also there's no way it will reach the claimed 2080 performance bracket without some "per core" improvements.

Considering Navi is slated for wide adoption in consoles, and as such has a real-world in-use life span of 5+ years, I'd say it needs efficiency improvements. If not, we'd end up with a very, very weird gaming market where consoles are borderline obsolete even at launch. Even if consoles move to a more PC-like product cycle (not just mid-lifespan performance upgrades like the One X or PS4 Pro, but more of a continuous, incremental performance increase at the same price every 2-3 years), it sounds highly unlikely that console makers would be willing to move to an entirely new architecture just a few years after boldly launching a "new generation" of consoles. That would require very significant software/firmware/driver work, which sounds like too much for a mid-cycle refresh/update as they wouldnt get the "new console generation!" sales boost to balance out development costs.

Then again, it could be that console refresh cycles and AMD's architecture development times just don't align - it could happen, if the arch update turns out to be very challenging - but again that sounds unlikely given the mutual dependence between these parties. Console makers could theoretically go to Nvidia, but it's highly unlikely they'd get the performance they want for the prices they want, and that would likely break (or significantly hurt) compatibility with all previous software. It's not like they can re-use PC drivers, after all. As such, console makers need a significant update from AMD for ~2020, and are unlikely to adopt an untried, 1st-gen architecture, while AMD needs the continued business of console makers. If Navi doesn't bring a noticeable efficiency improvement, I don't see how they could launch a new generation of consoles at that point.

Incoming: Nect nv product will be listed as "2,500,000,000 Hz. JH ("LJ" to his mates) is already quoting 1/4 FP numbers for the TFLOPS of their cards instead of the usual 32bit ...

Next up: near-infinite performance* with the next generation of cards.

*(in 1-bit calculations).

 

[eidairaman1]

If this card was 499$ with GDDR5 it would sold much better, this way AMD is just buying some time for navi i guess.

Totally agree. People tend to twist words and call stuff what they think it is despite the original though or meaning. GDDR5 in Vega would've been a disaster. Even though I really wanna look at the benchmarks and I'm curious is it just the shrink and clock boost or there's more in that new card. If it has a 2080 performance I think I will switch to this new AMD card.

Watch this please.

 

[Deleted member 158293]

Seriously awaiting single slot water blocks for this card. Would've really liked a shorter PCB, but this still could be my first small form factor build in alooooong time

 

[Berfs1]

I mean lots of vega 56s and 64s could overclock past 1600 MHz (56 requires a 64 bios flash), but the VII should pull much less power at the same clocks. The first gen Vegas are capable of hitting those clocks, but at high power draws.

 

[Valantar]

Seriously awaiting single slot water blocks for this card. Would've really liked a shorter PCB, but this still could be my first small form factor build in alooooong time

I still can't fathom Nvidia's insistence on having 2-slot I/O. Even if my case has room for two slots, I really like that my EK water block makes it single-slot

I mean lots of vega 56s and 64s could overclock past 1600 MHz (56 requires a 64 bios flash), but the VII should pull much less power at the same clocks. The first gen Vegas are capable of hitting those clocks, but at high power draws.

With a full node shrink between the two, that's a pretty low bar. 40% lower power at equal performance isn't unheard of across a full node shrink, which even accounting for the extra power draw of two more stacks of HBM2 would mean running ~1500MHz at ~180-200W if extrapolating from the V64. I suppose that tells us something of how far up the power/perf curve they're pushing this, just like V56 and 64.